Proceedings

This study applies Niklas Luhmann's social systems theory to conceptualize blockchain-based business reputation systems as integrated sociotechnical systems. The authors propose a design lens that aligns social layers (observation, selection, communication, system trust, and elements/relations) with technical layers. The methodology uses conceptual design research, illustrated and analyzed through a stylized market scenario.

The problem. Existing research on reputation systems is predominantly technical, neglecting how human interpretation, social behaviors, and incentives shape system outcomes. As a result, current blockchain-based solutions lack an integrated conceptual foundation, making it difficult to resolve social challenges like fake ratings, lack of feedback incentives, and reputation inflation. This gap prevents these systems from establishing trust beyond their technical infrastructure.

Key findings.
– The study conceptualized a social design layer for reputation systems based on Luhmann's systems theory, aligning social and technical mechanisms (conceptually supported).
– The role of the social layer was illustrated through a market scenario where tipping serves as a voluntary reputation signal and rating disclosure serves as an economic incentive (conceptually supported).
– The study does not contain empirical testing, statistical analysis, or quantitative validation of the proposed framework, which remains a key limitation (no statistical support).
– The conceptual nature of the study introduces limitations, such as a high level of abstraction that constrains immediate practical guidelines and assumes buyers are consistently willing to tip.

This study investigates how workflow-integrated chatbots can optimize enterprise knowledge retrieval. Employing a design science research methodology, the authors developed a prototype in the finance department of a medium-sized German company and evaluated its usability and performance.

The problem. Employees often waste significant time searching for fragmented internal information or clarifying ambiguous documentation. While large language models and retrieval-augmented generation show promise, organizations struggle to design systems that ensure user trust, data privacy, and seamless day-to-day adoption.

Key findings.
– The chatbot prototype achieved a mean usability score of 68.4 on the Chatbot Usability Questionnaire, placing it slightly above the industry average benchmark of 68. (Supported)
– The prototype demonstrated a high conversational accuracy rate of 88.2% (247 out of 280 interactions) during user testing. (Supported)
– The chatbot's ability to reliably interpret and reason with Excel-based tabular data was mixed, highlighting an ongoing technical limitation of the model. (Mixed)
– The study is limited by its small sample size (15 employees in the field study), single organizational context, and specific language and platform setup. (Limitations)

This study designs and evaluates NeuroAdaptX, an explainable AI interface that adapts presentation-level properties like structure, modality, and density to self-reported neurodiversity profiles. Grounded in Cognitive Load Theory and Cognitive Fit Theory, the author establishes design principles for cognitive accessibility. The system's utility is evaluated through a randomized between-subjects online experiment with 216 participants against a static baseline.

The problem. Standard explainable AI systems often assume a universal user profile, neglecting cognitive diversity such as ADHD, autism spectrum conditions, and dyslexia. This lack of customization can impose excessive cognitive load, degrade comprehension, and distort trust. Prior research has focused on content relevance rather than structural presentation accessibility, leaving a critical gap in design knowledge for inclusive interface adaptation.

Key findings.
– Hypothesis 1 (supported): NeuroAdaptX significantly lowered perceived cognitive load compared to the static baseline interface (p < .001).
– Hypothesis 2 (supported): The adaptive interface led to significantly higher objective comprehension of the AI's explanations (p = .004).
– Hypothesis 3 (supported): Users reported significantly higher explanation satisfaction and quality with the neuro-adaptive design (p < .001).
– Hypothesis 4 (supported): The treatment significantly improved calibrated and appropriate trust compared to the baseline (p < .001).
– Hypothesis 5 (not supported): Exploratory testing showed no statistically significant moderation effect of specific neurodiversity profiles on cognitive load or comprehension, suggesting directionally consistent benefits across ADHD, ASC, and dyslexia groups.
– Hypothesis 6 (supported): The adaptive interface significantly improved perceived informational justice (p = .001).
– Limitations: The study's vignette-based online format abstracts from high-stakes environments and repeated-use dynamics, and profiles were determined via self-report rather than clinical diagnostics.

This study proposes a preliminary design theory for structuring and orchestrating discourse strategies in conversational agents. Utilizing Habermas's Theory of Communicative Action (TCA) as a kernel theory, the authors synthesize existing research to define prescriptive design principles that address cognitive, affective, and behavioral outcomes.

The problem. Prior frameworks for explainable AI and human-agent interaction are siloed and fail to provide an overarching theory of human-machine discourse. Furthermore, older interaction models operate on the naive assumption that agents are inherently honest, failing to address risks such as user manipulation, deception, or emotional persuasion in organizational settings.

Key findings.
– Synthesizes a conceptual design theory identifying five core discourse strategies: instrumental, strategic, expressive, normative, and communicative action.
– Conceptually supports the Instrumental Strategy for achieving high task performance and efficiency while minimizing cognitive load.
– Conceptually supports the Communicative Action Strategy (such as providing transparent explanations) for improving user trust, learning, and overall satisfaction.
– Conceptually supports the Expressive Strategy, grounded in the CASA paradigm, for eliciting emotional trust and perceived social presence through human-like cues.
– Identifies that Strategic (deceptive or covert) and Communicative Action (transparent) strategies are mutually exclusive and cannot be deployed simultaneously.
– Limitation: The proposed principles are based on a conceptual synthesis of prior literature and have not yet been empirically validated in real-world user testing.

This study employs a Design Science Research (DSR) approach to develop and evaluate a web-based conversational financial co-pilot. The designed prototype integrates a stateful, session-based memory with a Large Language Model (LLM) to deliver tailored financial advice based on a user's unique goals. The utility and trust-building capacity of the artifact were evaluated through a qualitative focus group with five domain experts.

The problem. Current digital financial services face a major trust deficit due to static robo-advisors that offer limited personalization and ungrounded conversational agents that lack logical constraint enforcement. Consequently, existing digital platforms fail to replicate the competence-based trust that human advisors establish, which hinders their adoption and scalability. This research addresses the challenge of designing conversational interfaces that can demonstrate genuine competence to earn user trust.

Key findings.
– The qualitative evaluation showed that stateful memory can significantly enhance perceived system competence, though its performance proved fragile to evolving user preferences.
– The system's failure to consistently enforce negative constraints (such as recommending excluded assets) was identified as a critical breach of competence that severely damages trust.
– A significant gap was observed between superficial persona-matching and true hyper-personalization, indicating that users judge personalization by the depth of conversational inquiry before a recommendation is made.
– To build sufficient trust in high-stakes financial domains, conversational AI must be paired with institutional trust signals, such as clear liability frameworks, risk disclosures, and graceful management of out-of-boundary requests.
– The study's findings are constrained by a small evaluation sample (n = 5 experts) and its cross-sectional, short-term design, necessitating future quantitative and longitudinal validation.

This study details the design, development, and evaluation of a web-based support system for anomaly detection in Ambient Assisted Living. The researchers developed a hybrid system utilizing unsupervised machine learning algorithms, specifically Isolation Forest and LSTM-Autoencoder, combined with SHAP-based Explainable AI to interpret alerts. The artifact was evaluated through a quantitative analysis of model agreement on unlabeled sensor data alongside a qualitative focus group with six domain experts.

The problem. Applying machine learning to automate health-related anomaly detection in eldercare is severely limited by a scarcity of labeled training data. Furthermore, existing unsupervised algorithms operate as opaque black boxes, providing alerts without explanation. This lack of interpretability prevents caregivers from establishing the trust required to act on automated alerts.

Key findings.
– Quantitative evaluation supported the implementation of a hybrid model, confirming that Isolation Forest and LSTM-Autoencoder algorithms detect orthogonal, complementary types of anomalies (point outliers versus temporal pattern breaks).\n- Qualitative evaluation supported the existence of a 'Model Selection Paradox,' showing that giving users algorithmic choices and displaying conflicting outputs significantly erodes trust and increases cognitive load.\n- Qualitative evaluation supported the need for contextualized explanations, finding that raw feature importance scores are insufficient for decision-making without semantic baseline comparisons.\n- Focus group feedback supported a user preference for push-based notification interactions rather than continuous pull-based dashboard monitoring.\n- The study's limitations include a small expert focus group sample (n = 6), a short-term evaluation period, and reliance on sensor data from a single provider.

This study details the development and qualitative evaluation of an interactive AI-as-a-Service (AIaaS) toolbox designed to help small- and medium-sized enterprises (SMEs) independently assess, design, and adopt AI initiatives. Applying a design science research approach, the authors developed a set of design requirements, principles, and features, and instantiated them in an interactive prototype.

The problem. SMEs face significant pressure to adopt artificial intelligence to stay competitive, yet they frequently lack the internal expertise and financial resources to execute these projects. This lack of guidance often leads to exploratory, technology-first experiments that fail to align with strategic business needs, resource capacities, or regulatory requirements.

Key findings.
– The conceptual design knowledge (consisting of 4 requirements, 8 design principles, and 21 design features) was qualitatively supported and validated by SME practitioners.\n- Providing concrete, archetypal use cases (DP4) was rated as the most highly relevant design principle, receiving a perfect score of 5 out of 5 from all interviewed SMEs.\n- A design conflict was identified between the linear progress sequence of the prototype and the practitioners' preference for a flexible, non-hierarchical 'pick-and-choose' navigation style, which was addressed in final design adjustments.\n- Results are based on qualitative evaluations with a limited sample of 11 practitioners from 5 SMEs assessing a non-functional prototype rather than a fully deployed system in a live environment.

This study presents the design, development, and evaluation of XAI-MAP, an interactive decision-tree-based tool designed to recommend specific explainable AI (XAI) methods based on use-case characteristics. Utilizing a design science research approach, the researchers gathered requirements from interviews and literature to build the prototype, which was then evaluated via a task-based study with 20 AI developers.

The problem. Selecting appropriate XAI methods for specific applications is challenging for AI developers who lack specialized XAI expertise. Existing selection guidelines are often fragmented, operate only at a coarse method-class level, or assume prior methodological knowledge, leaving a gap for context-sensitive, practical decision support.

Key findings.
– The decision-tree-based prototype was evaluated by 20 developers, yielding a mean System Usability Scale (SUS) score of 87.5, which corresponds to a subjective rating of 'excellent' and indicates high perceived ease of use and low interactional effort.
– Quantitative feedback from the User Experience Questionnaire (UEQ-S) supported the tool's pragmatic quality, with developers rating it as supportive, simple, and structured.
– Users identified semantic friction points regarding domain-specific terminology (such as 'segmentation' and 'end users'), indicating that while the interaction logic was intuitive, terminology clarity requires refinement.
– The study's quantitative results are limited by a small evaluation sample size (n=20) and a simulated, task-based setting, meaning the survey metrics represent descriptive trends rather than statistically generalizable findings.

This study introduces and evaluates the Contextual Orchestration and Reward Shaping Engine (CORSE), a modular decision-support architecture designed to improve the reliability of large language model (LLM) outputs. The authors develop and formalize CORSE's orchestration and reward-shaping mechanisms using an Action Design Research methodology. The system is experimentally validated using LLaMA 3.3 and Claude 4.5 on 50 synthetic financial advisory profiles to evaluate its feasibility and performance.

The problem. Deploying LLMs in high-stakes, regulated environments like financial advising is challenging due to the models' sensitivity to prompt design and inability to consistently incorporate structured constraints. Typical prompting techniques or post-hoc evaluation approaches fail to prevent fluent yet materially incorrect or non-compliant outputs. There is a lack of closed-loop architectures that dynamically route prompts and enforce explicit domain-specific rules during inference.

Key findings.
– Under controlled simulation conditions, the full CORSE architecture (combining contextual orchestration and reward-shaping evaluation) achieved statistically significant improvements in pass rates and mean composite reward scores over both baseline and orchestration-only configurations across Claude 4.5 and LLaMA 3.3 models (all p-values = 0.0000).
– The performance difference between the Baseline and Contextual Orchestration-only (CO-only) configurations was not statistically significant (p = 0.2967 for Claude 4.5 pass rate, p = 0.2179 for LLaMA 3.3 pass rate), indicating that adaptive reward shaping is the primary driver of recommendation quality.
– CORSE enabled rapid error correction, with an average of 0.84 iterations required to resolve failure modes per client profile.
– The study's findings are limited by its reliance on a controlled synthetic financial-planning setting and prototype metrics, which may not fully capture the complexities of live, institutional advisory environments.

This study aims to ground the design of assistive technologies for blind and visually impaired people through empirically validated problem understanding and actionable guidance. Adopting an echeloned Design Science Research approach, the authors conducted 25 semi-structured interviews with visually impaired individuals to identify core challenges and define system requirements.

The problem. Many existing assistive technologies are built for mass markets and fail to align with the context-sensitive barriers and everyday practices of visually impaired individuals. Additionally, current research lacks a comprehensive understanding of how emerging artificial intelligence tools can be designed to address these complex daily challenges without causing user strain.

Key findings.
– Seven distinct problem areas (interaction with surroundings, assistive tools, daily life, well-being, public interaction, external dependencies, and financial aspects) were empirically identified and validated through qualitative coding.
– The research derived two major design objectives (Managing Interaction with the Surroundings and Managing Assistive Tools) and formulated thirteen concrete design requirements to guide future developers.
– Because this is an initial design science and qualitative analysis study, findings are qualitatively supported by grounded theory methodology rather than statistical hypothesis testing, with recognized limitations including a German-specific context and a sample potentially biased toward higher technology affinity.

This study utilizes Design Science Research to develop and evaluate a senior-friendly pedestrian routing artifact that integrates barrier avoidance and supportive amenities. The routing engine was implemented using open-source street and elevation data, and subsequently evaluated with older adults. The methodology includes a preliminary test and a field-based walking comparison against a mainstream navigation baseline.

The problem. Standard pedestrian navigation tools optimize primarily for distance or travel time, ignoring physical barriers, safety considerations, and supportive amenities critical to older adults. This misalignment can restrict late-life mobility, leading to decreased social participation and increased isolation. There is a lack of empirical design knowledge to guide practitioners in building senior-friendly, explainable routing applications.

Key findings.
– In a descriptive comparative preference test, 64% of participants (9 out of 14) preferred the senior-friendly route over the Google Maps baseline, while 29% reported no preference.
– Descriptive mean ratings favored the senior-friendly route across all dimensions, including perceived safety (+17.35%), comfort (+12.24%), physical load (+10.20%), and suitability for walking alone (+7.14%).
– Due to the small sample size (n=14) and descriptive nature of the final evaluation, formal statistical significance testing was not conducted on the quantitative route comparisons.
– Qualitative interviews validated that route choices are highly situational, depending heavily on dynamic context-aware factors (weather, lighting, and season) and social conditions (walking alone versus accompanied).
– The study is limited by its small sample size, potential peer influence during group-based walking evaluations, and geographic confinement to a single mid-sized German city.

This study introduces the Verified Adaptive Learning in Deployment (VALID) framework, a five-layer governance architecture designed to support the transition from human-in-the-loop oversight to human-on-the-loop supervision in autonomous contract execution contexts. Using Elaborated Action Design Research (eADR), the framework was developed and iteratively refined through three design cycles. Its performance was evaluated using a multi-AI agent simulation across 150 matched runs against a simulated human baseline incorporating modeled cognitive biases.

The problem. While organizations increasingly delegate complex tasks to autonomous AI agents, existing governance frameworks provide limited guidance on how these agents can earn and sustain interorganizational trust. There is a lack of architectures that support continuous, performance-based agent learning while ensuring strict compliance with constitutional constraints. Consequently, organizations struggle to transition safely from direct human oversight to high-level supervisory roles.

Key findings.
– The VALID framework successfully supported trust progression, with agents demonstrating a statistically significant increase in trust from a neutral baseline of 0.50 to a mean of 0.805 across 150 simulated runs.
– The hypothesis that governed agents would maintain high decision quality and avoid hesitation was supported, with agents achieving a stable 48.9% win rate and 0% avoidance behavior, compared to a 47.1% avoidance rate in the simulated human baseline.
– The initial assumption that higher trust scores would naturally produce greater operational sustainability was not supported; high-trust agents (trust >= 0.85) achieved only 53.8% sustainability compared to 70.0% for lower-trust agents because they accepted commitments beyond organizational delivery capacity.
– A primary limitation is that all findings are based on modeled computational outcomes in a controlled simulation under ideal technical conditions rather than empirically observed real-world behavior.

This study presents and validates a socio-technical machine learning development framework called UML-Med designed to build trustworthy and deployable clinical prediction applications. The researchers used Design Science Research Methodology (DSRM) to structure the framework and validated its practical application by building a patient no-show prediction model using five years of electronic health record data (7.9 million appointments) from a pediatric hospital.

The problem. Despite promising predictive performance, machine learning models face limited adoption in real-world clinical settings due to a trust gap caused by prediction uncertainty, lack of explainability, and limited methodological reproducibility. Traditional model evaluations rely on static performance metrics like accuracy or area under the curve (AUC) without assessing model stability and predictive variance under realistic data variations.

Key findings.
– Incorporating uncertainty quantification alongside traditional metrics enabled the selection of a more stable and reliable model (AdaBoost) that had a lower prediction uncertainty (mean spread of 0.0073) and stable cross-validation performance. (Supported)
– Evaluating models purely on predictive accuracy (AUC) was shown to be insufficient for clinical deployment, as the highest-performing models (Bayesian Neural Networks and Gradient Boosting) exhibited greater predictive variance and uncertainty. (Supported)
– Model-agnostic explanations (SHAP) demonstrated that patient behavioral history and scheduling details (such as past cancellations, rescheduling frequency, and late-day appointment hours) are more powerful predictors of no-shows than demographic variables. (Supported)
– Limitation: The framework's validation was restricted to historical electronic health records from a single pediatric ambulatory clinic, meaning model performance and feature weights may drift or differ across different geographic settings and clinical populations.

This study utilizes design science research to conceptualize and map the design space of graphical user interface (GUI) test automation. The authors analyze 103 GUI-based test automation frameworks to build a structured, multi-dimensional taxonomy. This taxonomy is integrated with a cluster analysis to identify and describe recurring design patterns, or archetypes, in test automation.

The problem. The current landscape of GUI-based test automation is fragmented and diverse, ranging from deterministic manual approaches to autonomous AI solutions. Practitioners struggle to make informed decisions because the underlying design choices of various tools and frameworks remain implicit and lack a structured comparison framework. This leads to tool-centric selection rather than strategic, design-aligned configuration.

Key findings.
– Developed a taxonomy containing four meta-dimensions (context, test strategy, system view, execution) and nine design dimensions to classify GUI test automation.
– Identified five distinct design archetypes (human-guided, deterministic, autonomous learning-based, rule-based, and learning-guided) through a k-modes cluster analysis of 103 frameworks.
– Evaluated the taxonomy's practical applicability by mapping it to five established real-world tools, demonstrating how different design choices align with specific testing goals.
– This design science research study did not conduct empirical statistical hypothesis testing; the identified archetypes are descriptive and require further empirical validation with active practitioners in real-world settings.

This study utilizes a Design Science Research approach to develop and evaluate a design theory for cross-source information seeking in software development. The authors design and instantiate a local-first, modular Retrieval-Augmented Generation (RAG) prototype named SPOCK that operates on commodity hardware. The system's feasibility, usability, and performance are evaluated through formative benchmarks and a naturalistic field study in a small-to-medium enterprise (SME).

The problem. Software developers spend nearly a third of their workday searching and integrating fragmented information across heterogeneous sources, leading to high cognitive overhead and reduced efficiency. While cloud-hosted Retrieval-Augmented Generation (RAG) offers a solution, privacy concerns and high infrastructure costs of Large Language Models (LLMs) prevent adoption in resource-constrained small and medium-sized enterprises (SMEs). Conversely, using compact, locally-deployed Small Language Models (SLMs) introduces high risks of hallucination and model confusion when processing noisy, multi-source contexts.

Key findings.
– Formative benchmarks empirically demonstrated that a hybrid retrieval strategy combining keyword scoring with vector-based semantic similarity achieves a more balanced search performance (overall Mean Reciprocal Rank of 0.67 and Precision@10 of 0.79) than keyword-only or embedding-only approaches in isolation.
– Field evaluation confirmed that a centralized knowledge index (DP-X1) reduces developers' search effort and context-switching overhead compared to traditional fragmented search workflows.
– The deployment of evidence-driven traceability features (DP-RS1), such as collapsible 'evidence chips' linked to original documents, was shown to increase developer trust and facilitate rapid technical verification of generated outputs.
– System administrators confirmed the feasibility of local-first, containerized Small Language Model (SLM) deployment (DP-GE1) for meeting data privacy compliance on standard hardware (8-16 GB RAM), though latency was identified as a potential adoption risk.
– Acknowledged limitations of the study include its evaluation within a single SME, which may limit immediate transferability to other organizational environments, and the limited generative depth inherent to compact SLMs.

This study introduces AIDE, a human-in-the-loop, AI-assisted pipeline designed to systematically extract and formalize latent design knowledge from non-design science publications. The authors demonstrate its feasibility by applying it to a corpus of 99 interdisciplinary papers focusing on AI-enabled circular economy strategies.

The problem. Valuable design insights in applied fields like engineering and computer science often remain embedded implicitly in technical descriptions and evaluations, rather than being expressed as reusable design knowledge. Extracting these insights manually is highly resource-intensive and lacks a systematic, scalable methodology.

Key findings.
– The study successfully demonstrated the feasibility of the AIDE pipeline by extracting and formalizing one expert-validated design principle per paper from a corpus of 99 peer-reviewed publications.
– The pipeline processed all 99 publications in approximately 1.5 hours using Gemini Flash 2.5, establishing a balance between extraction quality and computational efficiency.
– The effectiveness of the pipeline is limited by the richness and clarity of the source text, as well as current large language model constraints regarding context windows and inference stability.

This study proposes a theoretical framework for anticipatory design science to complement traditional Design Science Research (DSR). It introduces strategies for temporal expansion to help researchers design technology while proactively imagining its long-term sociotechnical consequences.

The problem. Traditional design science is often limited by its focus on immediate, present-day problems, which temporally localizes solutions and ignores future social, ethical, and environmental impacts. Without tools to speculate about future worlds, designers frequently project current assumptions onto future users, potentially causing unintended negative consequences when technologies become mundane.

Key findings.
– Proposed three conceptual strategies for temporal expansion in design: problem problematization, designing speculative social worlds, and planning for evolving systems rather than final goals (Theoretical proposition; no empirical or statistical hypotheses were tested).
– Argued that designing with speculative 'possible worlds' helps expose hidden assumptions and potential negative externalities of technological systems before they are deployed (Conceptual analysis; no empirical or statistical hypotheses were tested).
– Recommended a shift in design evaluation criteria toward metrics like criticality, interestingness, and fruitfulness to better judge future-oriented designs (Conceptual recommendation; no empirical or statistical hypotheses were tested).
– Acknowledged limitations in that this is a conceptual and theoretical paper without empirical validation or statistical testing of the proposed design strategies in real-world organizations.

This study employs a design science research approach to establish prescriptive design knowledge for the ethical integration of artificial intelligence within emergency medical dispatch systems. The methodology combines a systematic literature review of 34 papers with qualitative analysis of eight semi-structured interviews with experts and laypeople. The resulting design requirements and principles were evaluated and refined through an online focal group workshop with four domain experts.

The problem. The implementation of artificial intelligence in safety-critical, high-stress emergency dispatch environments raises complex ethical challenges, such as algorithmic bias, opaque decision-making, and the potential erosion of human accountability. While technological capability is rapidly advancing, there is a lack of concrete, practice-oriented design guidelines that translate abstract ethical concepts into actionable requirements for developers and practitioners. This gap risks a premature and unreflective shift of decision authority from human dispatchers to automated systems.

Key findings.
– The study successfully formulated nine distinct design requirements and six actionable design principles to guide the ethical development and deployment of AI in emergency dispatch systems.
– The proposed design principles were conceptually validated and supported by a consensus of domain experts during an ex-ante evaluation workshop, though traditional statistical hypothesis testing was not conducted due to the qualitative design science methodology.
– Key principles advocate for positioning AI strictly as non-binding decision support, using adaptive and purpose-bound data collection, incorporating medically validated gender-specific symptom profiles, and providing low-threshold, cue-based explanations.
– The study's findings are limited by its ex-ante nature, meaning the principles were evaluated conceptually rather than through the ex-post observation of a fully implemented system in a live, real-world operational environment.

This study develops a conceptual framework designed to guide researchers in identifying, positioning, evaluating, and formulating research problems in Design Science Research (DSR). Drawing on both design science and design thinking methodologies, the authors propose a workflow comprising the Problem Funnel, the Problem Space Matrix, seven explicit quality criteria, and five actionable design guidelines. The utility of the framework is demonstrated through two illustrative application examples: sustainable behavior in residential buildings and enterprise systems migration.

The problem. While identifying a research problem is recognized as the essential first step in Design Science Research, existing methodologies offer very limited operational guidance on what constitutes a high-quality problem. Consequently, researchers frequently struggle with either overly abstract 'grand challenges' that are intractable in a single project, or overly specific 'technical problems' that fail to generate generalizable, theory-driven knowledge. This lack of systematic evaluation criteria leads to poorly formulated problem statements and potentially irrelevant design artifacts.

Key findings.
– Developed the 'Problem Funnel' model to conceptually structure how broad, abstract challenges are progressively decomposed into research-tractable problems.
– Developed the 'Problem Space Matrix' to position problems along two dimensions (scope and structure), identifying the sweet spot for Design Science Research as focused yet complex problems.
– Proposed seven explicit, literature-grounded criteria for evaluating research problem quality: appropriate specificity, empirical grounding, stakeholder significance, tractable solvability, generalizable scope, solution novelty, and articulable clarity.
– Formulated five practical guidelines for problem identification and development to help researchers systematically navigate and refine their problem statements.
– Identified several common pitfalls in problem formulation, such as the Scope Trap, Solution-First Thinking, Gap-Spotting Without Problematization, and the Novelty Trap.
– Acknowledged that a key limitation of the proposed framework is the lack of solid empirical validation beyond the five seminar projects and two domain illustrations used to demonstrate its applicability.

This study conceptualizes prompts as Design Science Research (DSR) artifacts to integrate them into systematic socio-technical development and evaluation methodologies. The authors outline a two-level conceptualization, task support and prompt specification, to guide prompt engineering and research in non-deterministic large language model environments.

The problem. In existing Design Science Research, prompts are frequently treated as minor, static implementation details rather than intentionally designed components, leading to a lack of transparency and poor evaluation rigor. This under-specification exacerbates the gaps between stakeholder intentions and final model outputs, especially given the inherently non-deterministic nature of generative AI.

Key findings.
– The paper logically establishes that prompts meet all core criteria of design science artifacts, meaning they are human-made, purposeful, utility-based, evaluable, socio-technically embedded, and technology-oriented.
– The authors proposed a two-level conceptual framework comprising task support (defining expectations, boundaries, and context) and prompt specification (encoding structured instruction techniques like chain-of-thought or few-shot prompting).
– The study delineates structured guidelines for prompt evaluation, emphasizing that because LLM behavior is non-deterministic, evaluation must rely on repeated executions, variance reporting, and failure-mode analysis rather than one-off tests.
– As a purely conceptual and argumentative paper, no empirical hypotheses were formulated or statistically tested, which represents a limitation that future empirical studies must address.

This study employs an eight-year Action Design Research (ADR) methodology to design, build, and evaluate a Healthcare Ecosystem Orchestration (HEO) artifact. The HEO artifact coordinates cross-level patient navigation across supra-regional, regional, and local healthcare environments. Its utility is evaluated using a naturalistic, ex post assessment of 48,516 real-world navigation cases.

The problem. Industrialized healthcare systems are highly fragmented, leading to significant inefficiencies and resource misallocation as patients navigate autonomous organizations. Traditional digital triage systems and symptom checkers are evaluated primarily on diagnostic accuracy, failing to translate clinical assessments into actionable navigation paths under level-specific capacity constraints. Consequently, there is a design gap in how to coordinate patient flows across diverse ecosystem levels without resorting to hierarchical control.

Key findings.
– The central hypothesis that the distribution of recommended urgency levels significantly differs across ecosystem levels was statistically supported (p < 0.001).
– Emergency-level triage recommendations were significantly higher at the regional hospital level (33.6%) compared to the supra-regional insurance level (9.3%) and the local general practitioner level (6.5%).
– Self-care recommendations were highest at the local general practitioner level (41.6%), representing a 5.4-fold difference compared to the regional hospital level (7.8%).
– The study developed and validated four core design principles: Compliant Self-Service Navigation, Tripartite Separation of Concerns, Level-Adaptive Configuration, and Cross-Level Resource Orchestration.
– Limitations include a study context restricted to the German healthcare system with its rigid separation of ambulatory and inpatient care, and the lack of assessment of long-term downstream metrics like downstream provider workload.

This study demonstrates and evaluates a reference architecture designed to construct process mining event logs from fragmented, multimodal data. By instantiating five core construction functions (parsing, abstraction and concretization, extraction, integration, and enrichment), the authors build a unified event log using CRM records, email metadata, and call transcripts. The system is evaluated via a real-world sales process case study in a small enterprise.

The problem. Organizations often store process data in highly fragmented and heterogeneous formats across different systems (such as CRMs, email clients, and phone records), preventing comprehensive process mining. Conventional approaches struggle to integrate these unstructured and structured modalities into a single coherent event log. This lack of integration leaves critical operational gaps, particularly in small and medium-sized enterprises that lack unified ERP systems.

Key findings.
– The instantiated architecture expanded the event log from a baseline of 9 activities and 142 events (CRM-only) to 23 activities and 723 events, providing significantly richer process insights.
– Temporal consistency was fully supported, with normalized timestamps achieving 100% monotone ordering across all 22 evaluated cases.
– Episode-based abstraction was supported, with 89.8% of evaluated conversational episodes rated as topically coherent by a domain expert.
– Automatic case resolution was supported, achieving an 86.4% correct automatic case assignment rate with the remaining 13.6% safely flagged as ambiguous for human review.
– Activity extraction labeling was only partially supported and showed mixed results; only 53.8% of proposed labels were rated as correct or mostly correct, with performance heavily dependent on the presence of strong keyword cues.
– Sentiment analysis at the episode level achieved a moderate accuracy of 68.3% (F1-score of 0.676), acting as an informative but noisy contextual signal.
– Study limitation: The evaluation is restricted to a single SME sales process and a small sample size of 22 cases.

This study employs a Design Science Research approach to formulate and evaluate prescriptive design principles for scaling and governing hyperautomation. The researcher derived 17 meta-requirements and 11 design principles from a structured literature review and 11 semi-structured expert interviews. The resulting design knowledge was evaluated using a mixed-methods approach combining a quantitative survey of 100 managers and ex-post qualitative expert feedback.

The problem. While organizations increasingly adopt hyperautomation to integrate diverse technologies, most existing frameworks focus only on task-level automation. There is a lack of actionable guidance on how to systematically coordinate heterogeneous technologies and manage human-AI collaboration in complex processes. This research gap often results in operational fragmentation, high integration debt, and scaling failures within enterprises.

Key findings.
– Survey results showed robust statistical support for the 17 identified implementation requirements, with an overall mean agreement score of 5.37 out of 7.
– The 11 proposed design principles were statistically supported with an overall mean agreement score of 5.21 out of 7, with 'security and compliance by design' (DP8) and 'risk-tiered exception governance' (DP4) receiving the highest endorsements.
– The principle prioritizing API-based integration over UI-based RPA (DP2) received the lowest statistical support (mean of 4.72, with only 54% of respondents agreeing), indicating perceived barriers like legacy constraints and limited API availability.
– Ex-post expert validation (n = 11) yielded even higher mean agreement scores (5.98 for requirements and 5.84 for principles), emphasizing that integration-centric discipline is crucial despite its perceived difficulty.
– A limitation of this study is that the evaluation is based on perceived agreement and expert critique rather than longitudinal performance metrics from actual field deployments.

This study conceptualizes a new mode of engagement in Design Science Research (DSR) termed the 'curator', where researchers analyze and synthesize unstructured practitioner accounts to build generalizable design knowledge. To define and demonstrate this approach, the authors outline a structured curation process and apply it to an illustrative project analyzing systems engineering trends from professional blog posts.

The problem. Traditional DSR often positions the researcher as a direct developer of design artifacts, which can limit the speed and scale at which design knowledge is accumulated. Furthermore, valuable, emergent practices documented informally by practitioners on platforms like professional blogs remain difficult to systematically process, structure, and generalize within current DSR frameworks.

Key findings.
– The 'curator' mode of engagement is established as a conceptually distinct DSR role, specialized in generating abstract, filtered, and generalized knowledge at an accelerated pace relative to traditional player or observer roles.
– A structured five-step curation process comprising data sourcing, analysis, generalization, abductive logic, and boundary control is proposed and conceptually supported.
– Using a dataset of approximately 92,766 words from Medium.com, the curation method successfully synthesized a process-theory-based framework characterizing the evolution of the systems engineering design function across four defining eras: Agile adaptation, DevOps adaptation, platform alignment, and coexistence with Artificial Intelligence.
– The study outlines potential limitations of this method, including the risk of analyzing transient industry trends that might rapidly fade, the volatility of online source data, and the risk of generating artifacts that may lack long-term theoretical resonance.

This study develops and evaluates an interactive 3D digital twin system integrated with a machine learning model to simulate and forecast land surface temperatures. Using design science research, the authors built a prototype for the city of Lahti, Finland, to help urban planners virtually test the thermal effects of modifications like green roofs, water bodies, and vegetation.

The problem. As rapid urbanization exacerbates the urban heat island effect, city planners lack interactive, forward-looking decision support systems to evaluate thermal impacts prior to physical construction. Existing tools are largely descriptive and retrospective, failing to offer interactive workflows where practitioners can easily simulate and compare alternative planning interventions.

Key findings.
– The optimized machine learning model achieved high predictive performance, with technically supported accuracy metrics of R² = 0.96, RMSE = 1.16°C, and a correlation coefficient of r = 0.98.
– User evaluations supported the system's usability (mean 3.92/5) and usefulness (mean 4.12/5) as an interactive tool to compare alternative planning scenarios with stakeholders.
– Trust in the prediction results was only moderate (mean 3.70/5), indicating that high predictive performance alone does not ensure user confidence and that system designs must explicitly incorporate transparency and data explainability.
– The behavioral evaluation is limited by a small sample size of five domain experts and students, which restricts statistical generalizability.

This study outlines the development and qualitative validation of the Collective Action Planning (CAP) method, a five-step facilitation framework designed to translate abstract multi-stakeholder visions into practical, near-term actions. The framework was iteratively refined and validated through sixteen cycles in academic, expert, and real-world project settings.

The problem. Multi-stakeholder initiatives often struggle to bridge the gap between long-term sustainability visions and immediate operational plans. Misaligned goals, complex ecosystems, and present-day constraints frequently lead to analysis paralysis or fragmented, uncoordinated efforts.

Key findings.
– The Collective Action Planning (CAP) method was qualitatively validated across 16 cycles, showing utility in helping partners transition from shared abstract visions to concrete operational plans.\n- Facilitators reported that CAP successfully helped uncover hidden challenges, structured constructive stakeholder dialogues, and clarified expected learning outcomes.\n- Evaluation revealed two distinct adoption patterns: a structured approach that builds a comprehensive system perspective, and a flexible approach that uses specific prompts to generate creative solutions.\n- Limitation: The method has only been validated within the context of circular ecosystems in European geographies, and its applicability to other industries or non-European cultural contexts remains untested.

This study proposes a conceptual modeling tool-based approach to Green Lean Six Sigma (GLSS) to support organizations in redesigning business processes for environmental sustainability. Developed as a design science research artifact using the ADOxx metamodeling platform, the prototype integrates environmental metrics into eleven traditional quality techniques. The solution was evaluated using illustrative manufacturing scenarios, expert interviews, and a System Usability Scale (SUS) study.

The problem. Despite growing demand for resource-efficient business processes, there is currently no universally accepted Green Lean Six Sigma standard or structured framework. Consequently, practitioners lack clear guidelines on how to modify traditional quality techniques for environmental goals. Furthermore, there is insufficient research on how to effectively codify and document GLSS project results using modeling software.

Key findings.
– The design of 11 GLSS conceptual model types successfully demonstrates how environmental metrics (such as the 'Voice of the Environment') can be integrated into traditional quality management techniques.
– In a System Usability Scale (SUS) study with 10 participants, 90% of users rated the prototype highly in terms of being easy to use, easy to learn, and instilling confidence.
– Qualitative feedback from three process management and BI experts confirmed that the modeling tool effectively supports all phases of the DMAIC cycle and enhances practitioners' understanding of environmental inputs and outputs.
– The research is limited by its early-stage iteration, a small evaluation sample (10 students as proxies), and a limited set of quality techniques, with real-world field evaluations planned for future work.

This study outlines and demonstrates the use of the Critical Incident Technique (CIT) as a semi-naturalistic evaluation approach in Design Science Research (DSR). It presents a five-step methodology and illustrates its application using a case study of a stakeholder engagement tool for Business Process Management (BPM) projects. The paper assesses the technique's feasibility, participant experiences, and researcher advantages.

The problem. Evaluating DSR artifacts in fully naturalistic, real-world settings is often expensive, time-consuming, and difficult to execute. Conversely, artificial evaluations like simulations often lack real-world validity and fail to engage participants deeply. There is a gap in existing methods for an intermediate approach that captures context-sensitive insights while maintaining feasibility.

Key findings.
– Verisimilitude: Qualitative data from practitioner interviews supported that using recalled real-world critical incidents provides high authenticity and realism compared to artificial scenarios.
– Increased Engagement: Participants demonstrated greater engagement and interest when evaluating the artifact against real incidents they personally experienced rather than simulated scenarios.
– Deep Reflection: The CIT method was shown to encourage deep reflection, allowing participants to uncover unexpected insights and shift their perspective on past incidents.
– Resource and Cost Efficiency: From the researchers' perspective, the CIT evaluation was highly resource-efficient, requiring only 20 hours of participant interaction to obtain rich data that would be costly and complex to capture in a fully naturalistic field study.
– Limitations: The findings are qualitative and illustrative, based on a small sample of 10 practitioners, and do not empirically establish superiority or statistical impact over alternative evaluation approaches.

This study investigates the design of mobile applications to coordinate spontaneous volunteers during storm surge crises. The researchers followed a design science research methodology to build an interactive mobile prototype and qualitatively evaluated it with citizens and emergency response experts.

The problem. Spontaneous volunteers provide critical help during disasters, but integrating them into official response efforts is difficult due to unstructured communication and safety hazards. Current mobile warning apps are limited to one-way alerts, leaving a gap in two-way coordination and collaborative volunteer management.

Key findings.
– A centralized, structured information dashboard and stepwise registration process were qualitatively supported by both citizens and experts as effective for quick onboarding.
– Concise digital safety briefings were qualitatively supported as valuable preparation tools, though experts noted they cannot fully replace on-site physical instruction.
– Peer-to-peer communication channels received mixed support, with volunteers valuing self-organization while experts raised concerns about misinformation and lack of moderation.
– Gamification and motivational features (such as progress badges) received mixed support, as some volunteers found them motivating while others deemed them inappropriate for disaster contexts.
– Study limitations include a small qualitative sample of 12 digitally literate participants and evaluation based on simulated walkthroughs rather than real-world deployment.

This study utilizes an Action Design Research approach to develop and refine an immersive, 3D game-based pretraining environment on the Roblox platform. The researchers evaluated the design across two cycles, first with an alpha-version exploratory test among student teachers (n=12) and subsequently via a large-scale field study with university students (n=82 experimental, n=107 control) to observe cognitive and motivational readiness.

The problem. Novice learners often face severe cognitive overload and motivational anxiety when introduced to highly complex business simulation games. Traditional pretraining strategies typically rely on passive instructional formats like video or text, which do not actively engage learners. Consequently, there is a design gap in understanding how interactive, immersive virtual worlds can be systematically structured to prepare students cognitively and motivationally without inducing additional extraneous cognitive load.

Key findings.
– The evaluation was exploratory and interpretive in nature, meaning no formal statistical significance testing or causal inferences were conducted to compare the groups.\n- Descriptively, participants across the immersive pretraining groups reported low levels of extraneous cognitive load and moderate-to-high germane cognitive load during the pretraining phase.\n- Enjoyment was descriptively high and anxiety remained low across all pretraining groups, but perceived autonomy was relatively low due to the highly linear progression of the task structure.\n- A technical breakdown during the field deployment showed that more than half of the students who encountered a bug did not return after the system was fixed, demonstrating a practical need for robust auto-save and progression recovery features in virtual learning environments.

This study presents an integrative literature review of 64 conference papers to evaluate how researchers design information technology artifacts using Large Language Models (LLMs). The authors analyze current design practices to identify challenges and propose theoretical guidelines for generative AI integration. The study is limited to conference papers published in 2025.

The problem. Designers increasingly integrate generative AI into systems, but they frequently treat these complex models as black boxes. This obscures what is actually designed versus what the underlying AI naturally provides, threatening the reproducibility and durability of design knowledge. Consequently, there is a lack of methodological guidance to navigate the non-deterministic nature and rapid evolution of generative AI.

Key findings.
– Descriptive analysis showed that 82.8% (53 out of 64 papers) black-boxed the design features of the LLM, failing to describe its core functionality, reasoning, or behavior in detail.
– Relying on prompting without further model contextualization was the dominant approach, observed in 60.9% (39 papers) of the reviewed studies, whereas only 32.8% (21 papers) contextualized their LLM applications.
– Attempts to generalize prescriptive design knowledge were absent in 60.9% (39 papers) of the sample, which did not abstract findings into formal design principles.
– The study conceptually synthesized five primary challenges (obscure composition, opaque contextualization, fragile consistency, knowledge erosion, and lack of guidance) and proposed three corrective guidelines to help researchers structure LLM-enabled artifact design.
– A noted limitation is that the literature review focused exclusively on 2025 conference publications, meaning journal-level publications with potentially more mature methodologies were not analyzed.

This study explores how the Consortium Research (CR) methodology can be applied to large-scale, enterprise-grade open source projects. Using the Eclipse Dataspace Components (EDC) project as a qualitative case study, the researchers examine how combining CR with open-source practices bridges the gap between academic rigor and practical industry relevance.

The problem. Sustaining both scientific rigor and practical relevance in Design Science Research is challenging in highly dynamic, multi-organizational, and technology-intensive fields. Traditional collaborative research methods are typically localized and lack the structured governance needed to scale and accumulate design knowledge across diverse stakeholders.

Key findings.
– The study relies on a qualitative, interpretive single-case research design rather than statistical testing; therefore, findings are supported qualitatively through case analysis rather than statistical significance.
– Combining Consortium Research (CR) with open source practices was shown to be a viable approach to scale collaborative design science research, enabling over a dozen organizations to contribute to a shared, production-ready artifact.
– Open source governance models (such as those of the Eclipse Foundation) effectively complement CR's role structure by providing structured mechanisms for decision-making, contribution management, and IP handling.
– Major collaborative challenges were identified, including personnel turnover, misalignment of short-term and long-term goals, and hidden strategic agendas, which require active management and soft skills to resolve.
– Study Limitation: Generalizability is limited as the findings are based on a single in-depth case study of the Eclipse Dataspace Components project within a specific industry context.

This study applies a Design Science Research (DSR) approach to develop and evaluate a design theory for cross-organizational master data management. The researchers combined a hermeneutic literature review with semi-structured industry expert interviews to design a framework and integrate it into a reference architecture for decentralized data spaces.

The problem. Traditional master data management depends on centralized registries or commercial data brokers to ensure data quality, creating strategic dependencies and compliance risks. Existing decentralized identity technologies lack the semantic integration required to manage master data, leaving a gap in how organizations can maintain accurate and sovereign data exchange without central intermediaries.

Key findings.
– The design theory, comprising five Design Principles (DPs) and eight Design Features (DFs), was qualitatively supported by industry experts who validated its utility in establishing organizational trust, semantic interoperability, and usage policy integration.
– The feasibility of the design was supported through successful integration and instantiation within the International Data Space Association (IDSA) Reference Architecture Model 4.0.
– Expert evaluation highlighted challenges (not yet statistically tested) regarding legal enforceability across jurisdictions, operational complexity, and key management overhead.
– The study's findings are limited by a small qualitative evaluation sample (four experts) and a primary focus on technical feasibility over broader empirical business adoption.

This study develops and evaluates design principles for the ethical deployment of automated mental workload monitoring systems (Auto-MWMS) in industrial environments. Using design science research, the authors construct a conceptual framework from literature and refine it through qualitative interviews with IT and industrial experts. The resulting principles aim to balance operational efficiency with human-centric values like privacy and autonomy.

The problem. Integrating humans-in-the-loop within the Industrial Internet of Things (IIoT) increases cognitive demands, highlighting the need to monitor workers' mental workload to prevent errors. However, such automated monitoring risks introducing intrusive surveillance, violating worker privacy, and compromising employee autonomy. While existing literature discusses these ethical challenges, a practical gap remains in translating these abstract concerns into actionable system design principles.

Key findings.
– All nine developed design requirements, including compliance with legal regulations, respect for privacy, and minimization of errors, were qualitatively supported by domain experts as comprehensively capturing the key ethical challenges of Auto-MWMS.
– All eight proposed design principles (such as restricted data use, pausing of data collection, and customizable worker permissions) were qualitatively supported by practitioners as understandable, relevant, and suitable for system design.
– There was no statistical hypothesis testing performed, as the study relied entirely on qualitative expert evaluation.
– The study acknowledges limitations, including a small sample size of three expert interviews and the fact that the design principles have not yet been implemented or tested in real-world IIoT systems.

This study uses a Design Science Research approach to develop and evaluate a multi-agent AI tutoring system that orchestrates specialized scaffolding mechanisms (conceptual, procedural, strategic, and metacognitive). The prototype was iteratively refined with researchers, educators, and students, followed by a quantitative analysis of scaffold effectiveness in a controlled concept mapping task.

The problem. Generic generative AI models handle user queries uniformly, which often encourages cognitive offloading and passive copying rather than active skill development. Standalone AI tools lack pedagogical grounding, are unable to sequence different support mechanisms dynamically, and do not maintain persistent learner profiles to gradually reduce assistance over time.

Key findings.
– Metacognitive scaffolds delivered first in the tutoring sequence produced the highest concept map growth (M = 2.66, SD = 3.50, Cohen's d = 0.96), which was statistically supported (95% CI [1.24, 1.91], excluding zero).
– Scaffolds delivered later in the sequence (strategic, procedural, and conceptual) did not yield statistically significant effects on learning behavior, as their 95% confidence intervals overlapped with zero.
– A focus group evaluation demonstrated that simultaneous exposure to a novel interface, complex tasks, and multi-agent interactions can cause cognitive overload in novice learners.
– The study's limitations include a formative evaluation context, a student sample restricted to one European business school, and pending summative learning outcome measurements.

This study proposes a conceptual framework to guide the development of design principles within design science research. It introduces a mode-based reasoning framework, an abstraction-alignment model, and five navigation moves to help researchers systematically align design requirements, principles, and features. The utility of the framework is demonstrated through a retrospective reconstruction of a longitudinal research project on virtual companionship.

The problem. Developing design principles that are both actionable and generalizable is a persistent challenge in design science research. Existing methodologies often focus on rigid activity sequences or structural formulas without explaining how researchers actually navigate different abstraction levels. Consequently, developed principles are frequently either too abstract to offer concrete guidance or too closely tied to specific implementations to be transferable.

Key findings.
– The study conceptually supports a mode-based framework that models design principle development as a dynamic, iterative process across Framing, Reflection, and Synthesis modes.\n- The abstraction-alignment model conceptually supports that design principles are most stable when design requirements, principles, and features align at a compatible abstraction level (referred to as the conceptual sweet spot).\n- The framework identifies and outlines five recurring epistemic actions (Abstraction, Specialization, Instantiation, Alignment, and Stabilization) that describe horizontal and vertical movements within the design knowledge space.\n- No statistical hypotheses were proposed or quantitatively tested in this conceptual study.\n- The primary limitation is that the framework was illustrated using a single retrospective reconstruction of a virtual companion project rather than validated across multiple live research domains.

This conceptual study proposes a theoretical framework to integrate serendipity into Design Science Research (DSR). The author develops a structured, four-phase model and guiding principles to systematically recognize, evaluate, and transform unexpected design outcomes into rigorous knowledge.

The problem. Traditional DSR methodologies treat unexpected findings and anomalies as deviations or noise to be resolved through iterative refinement. This orientation restricts researchers from leveraging emergence and surprise as productive sources for reframing problems and generating discovery, particularly in highly uncertain or complex environments.

Key findings.
– This is a purely conceptual and theory-building paper; it contains no empirical testing, statistical hypotheses, or quantitative data.
– Conceptualized a four-phase serendipity cycle model that details how designers can transition an anomaly from noise to signal: (1) encountering an unexpected trigger, (2) recognizing and reflecting, (3) making an abductive leap, and (4) materializing the insight into design knowledge.
– Proposed three guiding principles to enable serendipitous discovery: constructing artifacts with deliberate incompleteness, cultivating a prepared mind, and practicing collective, concurrent, and reflexive evaluation.
– Identified limitations of the framework, noting that the boundary conditions between the proposed phases are not yet empirically tested, decision gates for when to pursue versus park anomalies are not fully operationalized, and structural funding regimes may constrain its adoption.

This study adopts a trauma-informed design perspective to investigate how integration-related information systems are experienced by refugees. Following an echelons design science research approach, the authors analyzed qualitative data from 35 semi-structured interviews with Ukrainian adults residing across 22 Finnish cities to formulate actionable system design principles.

The problem. Existing public-sector information systems are typically built on rational-user and efficiency assumptions that ignore how trauma and displacement affect cognition, emotion, and trust. As a result, routine digital administrative interactions often induce severe anxiety, cognitive overload, and avoidance behavior, thereby worsening digital and social exclusion.

Key findings.
– Qualitative thematic analysis supported that language barriers, fragmented digital ecosystems, opaque system logic, and repetitive disclosure requests transform routine administrative tasks into emotionally distressing experiences.
– Qualitative evidence supported that trauma-related vulnerability amplifies cognitive overload and system avoidance, whereas informal human mediation (e.g., volunteers and community groups) acts as a critical, necessary workaround.
– The research successfully derived seven concrete design principles for trauma-informed systems, which focus on safety over efficiency, predictability, graduated engagement, integrated human support, contextual legibility, data dignity, and ecosystem coherence.
– Limitations include a highly specific sample (Ukrainian adults under Temporary Protection Directive in Finland) and the fact that these design principles have not yet been instantiated or evaluated in real-world environments.

This study develops and evaluates the Circular Transition Assessment Tool (CTAT), a digital self-assessment system designed to help plastics manufacturers evaluate and advance their transition toward a circular economy. Utilizing a design science research approach, the authors formulated meta-requirements and design principles to build a tool that integrates multi-level maturity scoring with automated, expert-validated recommendations. The artifact's utility and usability were iteratively evaluated through qualitative workshops and quantitative surveys with industry professionals.

The problem. Plastics manufacturers face intense regulatory and environmental pressure to transition from linear to circular production models, yet they lack domain-specific tools to accurately assess their readiness. Existing maturity models are often too generic, purely conceptual, or fail to connect with practical organizational decision-making, leading to misallocated resources. Furthermore, there is a lack of research on designing digital assessment tools that balance automated, transparent logic with human expert validation to support credible sustainability transitions.

Key findings.
– Quantitative survey results (N=10) supported that the assessment tool is intuitive, accessible, and highly usable for plastics-industry organizations.
– The evaluation supported the operational feasibility and expected positive decision-making impact of integrating the tool into routine organizational practices.
– The assumption that the tool's standard framework easily captures the highly complex, heterogeneous reality of all real-world circular economy processes was not fully supported, yielding a high number of neutral ratings regarding fidelity.
– The empirical findings are limited by a small sample size (eight companies total across two cycles) and a geographical focus on the German plastics sector, meaning the tool may require adaptation for other regions or industries.

This study adopts an echeloned Design Science Research approach to design, instantiate, and evaluate a generative AI chatbot named externalAIze for externalizing tacit knowledge. The artifact is built on four core design principles, phased reflective questioning, curated input with free-text fallback, adaptive motivational feedback, and human-in-the-loop validation, and was evaluated with 53 participants.

The problem. Organizations face a high risk of losing valuable, experience-based tacit knowledge due to workforce mobility, restructuring, and retirement. Traditional human-mediated elicitation methods are expensive and hard to scale, while existing conversational AI options lack prescriptive design guidance and frequently impose high cognitive loads on users.

Key findings.
– The hybrid interaction modality (curated input options with free-text fallback) was statistically supported as being significantly less demanding in terms of articulation effort compared to a fully open-ended text interface, while providing more expressive flexibility than a purely closed-ended interface.
– Phased reflective questioning based on systematic reflection was supported as a meaningful method to help users surface tacit knowledge, though its perceived effectiveness was lower for participants with no prior familiarity with generative AI systems.
– Empathetic and adaptive motivational feedback positively influenced user engagement and reflection, but was not supported as a standalone solution for older participants or those in later career stages who reported higher hindrances to articulation.
– A human-in-the-loop knowledge validation step was supported as an effective mechanism for users to correct errors and misinterpretations, significantly improving the perceived quality of the final explicit knowledge summary.
– Limitations of the study include a relatively small sample size of 53 participants, reliance on subjective self-report questionnaires, and evaluation using a generic, simplified weekend-trip scenario rather than complex, specialized business processes.

This study conducts a PRISMA-guided systematic literature review of 35 empirical articles to analyze how Design Science Research (DSR) designs and evaluates generative artificial intelligence (GenAI) artifacts. The authors propose a configuration-centric framework to assess artifact boundaries, evaluation evidence, knowledge portability, and governance in the GenAI era.

The problem. Generative AI challenges traditional DSR assumptions because its probabilistic nature, prompt sensitivity, and reliance on external components make it difficult to treat artifacts as stable, bounded objects. Consequently, there is a lack of clarity on how to establish rigorous evaluation, cumulative design knowledge, and practical validity for GenAI-enabled systems.

Key findings.
– The analysis of the corpus shows that GenAI artifacts are typically socio-technical configurations rather than standalone tools, with pipeline-style architectures present in 25 out of 35 studies and retrieval augmentation in 20 out of 35 studies.
– A recurring weakness identified in the reviewed literature is that evaluation evidence often outpaces configuration disclosure, with many studies failing to report stable model identifiers, prompt libraries, or retriever settings.
– While 23 out of 35 studies attempt to formulate portable design knowledge (such as design principles or blueprints), these contributions are frequently not configuration-complete, risking brittle prescriptions.
– The study finds that although ethics and governance concerns are widely acknowledged in the analyzed papers, they are rarely operationalized or evaluated as core functional design features.
– Acknowledged limitations of this review include the emergent and heterogeneous nature of the analyzed literature, its concentration in specific archetypes (such as RAG systems and assistants), and the potential exclusion of non-English or grey literature.

This study introduces 'fruitfulness' as a complementary evaluative dimension for Design Science Research (DSR). It develops a conceptual model that shifts the evaluation of design knowledge from immediate problem-solving utility to its capacity to expand future design options.

The problem. Traditional DSR evaluations focus primarily on the immediate 'fitness' of an artifact in solving a specific problem. This orientation offers limited guidance for assessing future-oriented or speculative designs whose primary value lies in opening new trajectories rather than immediate optimization.

Key findings.
– This is a purely conceptual study and does not present empirical testing or statistical hypotheses.\n- The paper proposes a conceptual model of 'fruitfulness' to assess how design knowledge expands future design trajectories within a possibility lattice.\n- The study introduces a taxonomy of four artifact types (configurational, adjacent-exploratory, latent-activating, and boundary-probing) to differentiate how designs support option-making, though these concepts are not empirically validated.

This design science study explores how hands-on entrepreneurship education can help individuals feel capable of actively shaping future societal and ecological paths. The researcher developed and evaluated a university course integrating sustainability concepts with entrepreneurial action, analyzing qualitative reflection papers from 20 participating students.

The problem. In an era of complex ecological and social crises, individuals often experience a sense of 'unable-ness' or paralysis, viewing the future as abstract and unchangeable. Conventional educational and organizational frameworks lack actionable design knowledge to help people overcome this perceived lack of agency and collectively build pathways toward desirable futures.

Key findings.
– Qualitative analysis of student reflection papers supported that integrating entrepreneurial action with collaborative future-making practices successfully cultivated a sense of personal 'able-ness' to shape desirable futures, with two individual exceptions where students remained unoptimistic or felt a lack of belonging.
– The qualitative data supported that acquiring knowledge about severe ecological crises does not lead to paralysis when paired with solution-oriented, collaborative entrepreneurial tasks.
– The study qualitatively validated eight design principles (DP1-DP8) emphasizing multi-disciplinary teamwork, value negotiation, systemic thinking, and safe spaces for prototyping as effective frameworks for cultivating future makers.
– Limitations of the findings include a small qualitative sample of 20 higher education students, potential desirability bias in self-reported reflections, and specific dependency on the educational context.

This study introduces Testbed Research, a design-oriented methodology adapted from Action Design Research (ADR) to guide the design and management of digital ecosystems. The authors develop a five-stage, process-oriented framework and evaluate its application across 30 real-world ecosystem initiatives conducted between 2016 and 2026.

The problem. Non-digital enterprises frequently struggle to establish and sustain digital ecosystems due to the high complexity of managing heterogeneous partners with conflicting interests. Existing design science research methods lack structured guidelines for coordinating multiple autonomous actors and do not account for the role of researchers as neutral orchestrators or trust anchors.

Key findings.
– This study does not perform statistical hypothesis testing, instead evaluating the framework qualitatively through practical application.
– The five-stage Testbed Research methodology facilitated the creation of digital ecosystem proofs of concept within 12 to 18 months across 30 real-world projects.
– The research team established a neutral 'trust anchor' role, which resolved controversial partner negotiations over risk and reward allocation during ecosystem formation.
– A detailed pay-per-part case study demonstrated that the method aligned a diverse group of partners, including an original equipment manufacturer, a bank, a service provider, and a mechanical engineering company, to realize collaborative value.
– Limitations include that the framework's performance was evaluated within a specific set of 30 initiatives, and the influence of different funding models on stakeholder dynamics remains untested.

This study analyzes 114 archetype development approaches in the Information Systems domain through a systematic literature review to construct a comprehensive taxonomy. The taxonomy classifies these approaches based on their context, objects of analysis, and development procedures to establish a clearer methodological foundation.

The problem. Although archetypes are widely used to categorize and analyze complex socio-technical phenomena, their construction often lacks methodological rigor and standardization. This fragmentation leads to ad-hoc research configurations, limits transparency, and makes it difficult to compare findings across different studies.

Key findings.
– The study successfully developed and validated a taxonomy comprising 43 characteristics across 11 dimensions to structure archetype development (supported by empirical analysis of 114 papers).
– Analysis showed that 97% of archetype developments are conducted ex-post and 98% are descriptive rather than prescriptive (statistically supported by frequency distribution).
– The most common purpose of archetypes is to serve as a design guide (65%), followed by conceptual maps (42%), assessment instruments (37%), and theory builders (30%) (statistically supported; non-exclusive categories).
– A significant methodological deficit was identified, with 80% of reviewed archetype studies containing no evaluation of their developed archetypes (statistically supported).
– The study's limitations include potential interpretive bias during taxonomy development and restriction of the analyzed corpus exclusively to the Information Systems domain.

This study examines how artificial intelligence (AI) systems can be designed and integrated to facilitate authentic learning in higher education. Utilizing an echeloned design science research (eDSR) approach, the authors analyzed 200 AI systems deployed across European institutions and interviewed 11 experienced educators to construct a design problem space. From these insights, they derived and qualitatively validated ten meta-requirements for AI learning support systems.

The problem. The rapid adoption of large language models in education has sparked concerns that students may outsource critical cognitive tasks, leading to deskilling and a lack of authentic learning. Additionally, educators often lack the competencies, time, and structured institutional support needed to effectively integrate these tools into their curricula. This creates a significant gap in prescriptive design guidance for educational AI systems that promote deep pedagogical engagement rather than simple cognitive shortcuts.

Key findings.
– A market review of 200 European educational AI systems revealed that current tools predominantly focus on administration, content-specific learning, or task feedback, while completely neglecting substantive student-to-student conversations.
– The study identified three central qualitative problem dimensions: educators' lack of competencies and resources, students' uncritical reliance on AI outputs, and insufficient institutional adoption support.
– Ten meta-requirements (R1-R10) addressing content, user experience, and governance were developed to guide the design of AI-supported learning platforms.
– Qualitative validation with experts supported the relevance and coherence of the meta-requirements, although educators noted that institutional regulations could restrict the flexibility of proposed assessment templates (R4).
– The study's findings are qualitative and subject to limitations including a geographic focus on European institutions and a small evaluation sample size of 11 initial interviewees, 12 focus group participants, and 3 validation interviewees.

This study utilizes an Action Design Research approach to iteratively design and evaluate a voxel-based augmented reality (AR) authoring tool. It compares a semi-structured, example-based scaffold against an unstructured baseline to examine how structural guidance affects the creativity and cognitive load of novice users.

The problem. Augmented reality technologies are largely limited to passive content consumption because existing authoring systems require complex programming and 3D modeling skills. Providing unstructured environments to lower these barriers can overwhelm novice users, yet adding guidance risks restricting creative autonomy and causing anchoring bias.

Key findings.
– Statistically Supported: Artifacts created with semi-structured example-based scaffolds were rated by external experts as significantly more creative than those created without scaffolds.\n- Not Statistically Supported: There was no statistically significant difference in participants' subjective perceptions of creativity support between the scaffolded and baseline conditions.\n- Not Statistically Supported: Differences in total cognitive load, germane load, and extraneous load between the scaffolded and baseline groups were not statistically significant.\n- Study Limitations: The findings are limited by a small sample of 36 novice users and a specific voxel-based design paradigm, meaning the results may not generalize to experienced users or alternative AR systems.

This study investigates the design requirements for AI-driven entrepreneurial coaching systems using a design science research approach. The researchers conducted semi-structured interviews with startup founders and professional coaches in Germany to understand the limitations of current coaching models and generic AI assistants.

The problem. Human startup coaching is heavily capacity-constrained and episodic, which often results in coaches lacking visibility into founders' day-to-day decisions. Meanwhile, generic LLM-based assistants lack persistent venture context and provide overly agreeable feedback, which risks substituting founder judgment instead of facilitating deep reflection and learning.

Key findings.
– Proposed and empirically supported four core design requirements for AI coaching systems: task-oriented venture progression support, structured venture building, assumption-driven validation, and reflective developmental support.
– Identified the 'scaffolding-substitution' boundary as a critical design tension, showing that AI systems should provide 'developmental friction' via critical questioning rather than immediate, unchecked answers.
– Outlined a hybrid service design model for support organizations where AI manages structured preparation (e.g., business model canvas, experiment planning), leaving scarce human coaching capacity for high-value reasoning under ambiguity.
– Acknowledged study limitations, noting the qualitative sample was situated specifically within German university-embedded incubator environments focusing on early-stage technology ventures.

This research-in-progress study designs and evaluates an AI-based conversational teaching assistant integrated directly with SAP S/4HANA to support ERP education. Following a design science research approach, the authors developed a hybrid conversational agent utilizing discriminative AI and conducted a preliminary evaluation using pilot user tests and qualitative expert feedback.

The problem. ERP systems are highly complex, making hands-on classroom training resource-intensive as instructors spend substantial time troubleshooting individual student errors. Standalone generative AI tools lack integration with real-time student transaction data and are prone to hallucinations, limiting their ability to provide accurate, context-specific feedback.

Key findings.
– Qualitative feedback from academic and professional instructors supported the proposition that the ERP teaching assistant saves significant troubleshooting time and mental resources.
– Preliminary user testing supported the finding that students with higher digital literacy navigated the conversational assistant more effectively, whereas students with less technology experience struggled to resolve dialogue misunderstandings.
– Structural testing supported the utility of using a discriminative AI model to successfully prevent hallucinations, though users noted conversational rigidity and gaps in theoretical curriculum content as current design limitations.

This study investigates how blended learning environments can be designed using a Design Science Research (DSR) approach to support the self-determination of professional adult learners. By combining Self-Determination Theory with literature analysis and learner needs, the study establishes a conceptual design framework and five initial design principles.

The problem. Adult professional learners often struggle with motivation in blended learning environments due to a lack of clear course structure, insufficient conceptual scaffolding, and weak alignment with real-world practice. Existing systems rarely design intentionally for psychological needs like autonomy, competence, and relatedness, resulting in poor learner engagement.

Key findings.
– Five initial design principles were conceptualized to address learning structure, conceptual clarity, active engagement, contextual relevance, and learning flow.
– The proposed design principles and the prototype have not yet been empirically or statistically tested, meaning the theorized motivational benefits remain hypothetical.
– Preliminary qualitative validation was limited to early feedback from only two students, who supported the feasibility of the design direction.
– Diagnostic survey responses from the targeted student group indicated that perceived autonomy was rated the lowest motivational factor, followed by relatedness, competence, and satisfaction, though statistical significance tests for these differences were not reported.
– A key limitation is the early stage of the research, which lacks full implementation, expert evaluation, and large-scale empirical testing.

This study details the design and pilot implementation of an LLM-adversarial assessment pattern where students are tasked with critiquing a flawed, AI-generated 'Dirty Audit' to test their evaluative judgment. The research employs a Design Science Research framework to develop a scalable and AI-resilient assessment model in a digital marketing course. To prevent academic misconduct and AI-assisted cheating, the methodology incorporates a written critique scaffold followed by a proposed synchronous oral verification layer.

The problem. The rapid advancement of generative AI allows students to easily produce polished and professional marketing audits, which invalidates traditional artifact-creation assignments. Consequently, educators face a validity crisis because grades no longer reliably measure students' analytical competence or independent understanding. There is a critical research gap in designing scalable assessments that leverage AI as an adversarial tool rather than merely attempting to restrict its use.

Key findings.
– In the first design cycle involving 151 students, the written critique format resulted in grade clustering in the A- to B+ range, demonstrating a ceiling effect that lacked the power to distinguish between independent work and AI-assisted submissions.
– Automated testing using an LLM red-team confirmed that generative AI alone could not reliably detect the subtle, context-specific errors embedded in the 'Dirty Audit' or construct traceable strategic recommendations, supporting the task's adversarial resistance.
– The study identified a limitation in asynchronous written assessments, showing they remain susceptible to AI-assisted completion and are insufficient for validation without synchronous defense.
– To address the ceiling effect and verify authorship, a seven-minute Interactive Oral Assessment (IOA) verification layer was designed for the second cycle, though its empirical efficacy remains to be fully evaluated in future research.

This study evaluates how a hybrid semantic graph-vector retrieval system impacts cognitive cost, user satisfaction, and task performance in retrieving presentation slides. The researcher conducted an online experiment with 56 participants who constructed a slide-based proposal pitch under different retrieval conditions.

The problem. Creating presentations often requires reusing legacy slides from large repositories, which is cognitively demanding and frequently lacks narrative consistency. Traditional search engines and hierarchical folders fail to support users in navigating complex, multimodal content and structuring cohesive stories.

Key findings.
– Hybrid graph-vector-based exploration led to statistically significantly higher user satisfaction compared to traditional hierarchical folder navigation (Supported).
– Hybrid graph-vector-based exploration statistically significantly improved task performance in slide retrieval and creation (Supported).
– Hybrid graph-vector-based exploration did not lead to a statistically significant decrease in cognitive effort compared to traditional navigation (Not Supported).
– Hypotheses regarding the quality of LLM-supported narratives and their mediating effects were not tested in this preliminary, work-in-progress paper, which is also limited by a small sample size (N=56).

This study develops a design science research framework to balance open innovation and information security within digital energy ecosystems. The researchers conceptualize openness and security as interdependent dimensions and apply their proposed Digital Energy Ecosystem (DEE) framework to the case of the Swedish National Data Hub.

The problem. The digitalization of smart grids creates a critical tension between the need for open, collaborative architectures that drive innovation and the necessity of robust cybersecurity to protect critical energy infrastructure. Existing management approaches often treat openness and security as opposing forces rather than co-evolving, interdependent design requirements.

Key findings.
– The study introduces the Digital Energy Ecosystem (DEE) framework, which models system performance as a function of innovation value, security risk, and a governance loop factor; formal statistical hypothesis testing was not conducted in this design science paper.
– Application of the framework to the Swedish Data Hub vignette demonstrated that regulatory delays reduced governance effectiveness (from a target of 1.2 to 0.7) and increased residual risk (from 1.6 to 2.2), explaining why the project was placed on hold.
– The conceptual and qualitative analysis supports the design proposition that systemic viability in digital energy infrastructures depends as much on institutional readiness and regulatory clarity as it does on technical cybersecurity solutions.

This research-in-progress paper proposes the conceptualization and development of a design knowledge library for distributed ledger technology (DLT) to enable the systematic reuse of design knowledge in future projects. The research design outlines a multi-phase approach involving a systematic literature review, expert interviews, and the implementation of two e-commerce prototypes.

The problem. Design science research (DSR) knowledge is rarely systematically accumulated or reused across projects, leading to inefficient and fragmented development processes. In the context of DLT, practitioners lack concrete, structured guidelines to bridge the gap between high-level technology adoption decisions and actual system implementation features.

Key findings.
– Identified initial design elements consisting of five requirement categories, nine design principles, and twelve implementation features from a preliminary review of seven DSR publications (Supported qualitatively by the initial literature review).
– Mapped complex relationships (n:m) between design requirements, such as preventing data manipulation, and corresponding principles and features (Supported qualitatively by the initial literature review).
– Outlined a structured four-phase development and evaluation plan for the library, including future validation through two e-commerce prototypes (Proposed framework, empirical validation remains for future work).
– Recognized limitations, as this is a research-in-progress study based on a small preliminary literature sample without full empirical validation or statistical testing.

This study presents the design, development, and simulation-based evaluation of a multi-agent AI system built to support job seekers across the entire career search lifecycle. Using the elaborated Action Design Research (eADR) method, the researchers developed a prototype featuring ten specialized LLM-powered agents managed by a supervisor coordinator and human-in-the-loop oversight. The system's functionality was evaluated using simulations with real candidate profiles and curated job postings.

The problem. While employers have rapidly adopted AI-driven systems to automate recruitment, candidates face a significant technological deficit, leading to high cognitive exhaustion and application abandonment. Existing tools only address isolated tasks like resume building rather than providing sustained, multi-step orchestration across the weeks-long job application lifecycle. This study addresses the need for a comprehensive, candidate-facing agentic system to balance this asymmetry.

Key findings.
– The prototype successfully sequenced and orchestrated all ten specialized agents to produce tailored outputs like customized resumes, cover letters, and company-specific interview prep materials (supported through simulation).
– The governance engine correctly enforced rate limits and personally identifiable information (PII) filtering while pausing for human-in-the-loop approval on high-risk actions such as submitting applications (supported through simulation).
– The study successfully validated all three proposed design principles regarding multi-agent specialization, hybrid human-in-the-loop checkpoints, and coordinator-led sequencing (supported through expert review and simulation).
– Limitations include the use of simulated, curated job postings rather than live APIs, evaluation on a small set of candidate profiles, and untested human-in-the-loop thresholds across broader demographics.

This research-in-progress study explores the design of a hybrid agentic workflow that integrates large language model (LLM) agents with traditional software components. The system combines structured Electronic Health Records (EHR) data and unstructured clinical notes to generate a consolidated summary and interactive visualization of a patient's life-long medical history.

The problem. Medical practitioners face significant time constraints when trying to review comprehensive, lifetime patient histories scattered across multiple structured databases and unstructured text documents. Current EHR systems are not capable of automatically consolidating and summarizing these heterogeneous data types into an intuitive, chronological overview.

Key findings.
– Statistical significance testing was not conducted in this research-in-progress study.
– The feasibility of the hybrid system architecture, which successfully divides tasks between LLM agents (for summarization) and traditional code (for data retrieval and visualization), was demonstrated through a proof-of-concept prototype.
– A set of preliminary design principles was derived to guide the development of clinical agentic systems, focusing on controlling LLM output variation and minimizing hallucinations.
– The current study is limited by the use of synthetic EHR data from a small sample of 14 simulated patients and the lack of formal clinical testing with medical professionals.

This study introduces a hybrid Design Science Research (DSR) framework that integrates Vijay Kumar's seven modes of innovation into Peffers et al.'s six-step process model. The researchers designed and evaluated this framework, alongside an AI-assisted decision-support prototype called DSR.Navigator, to guide method selection during DSR cycles.

The problem. While existing DSR models offer structural guidance, they lack curated, stage-aligned method options for researchers. This gap creates high cognitive load and forces researchers to rely on ad hoc or implicit method choices, which undermines both creativity and transparency.

Key findings.
– Preliminary expert evaluation supported that the revised round-based prototype is clearer and more usable than the original framework.
– Analysis of prior DSR cycles supported that structured integration of innovation modes reduces method-selection friction and enhances design logic transparency.
– The study is limited by its preliminary nature as a research-in-progress, with a formal evaluation of causal explanation and boundary conditions planned as the next step.

This study adopts a Design Science Research approach to design and build a prototype of a multi-agent, Large Language Model-based tutoring tool named ProofTutor to support university students' mathematical argumentation skills. The researchers consolidated requirements from nine semi-structured user interviews and a review of educational technology literature to construct their system. The resulting tool is designed to deliver formative feedback on mathematical proofs and offer conversational, step-by-step tutorial guidance.

The problem. Undergraduate students often struggle to construct coherent, logically valid mathematical proofs and express their reasoning in precise mathematical language. While personalized feedback is vital for overcoming these challenges, instructors in large lecture settings face significant resource constraints that limit opportunities for individualized guidance. Existing automated tutoring tools frequently lack the capability to provide nuanced feedback on the logical structure and coherence of proof-writing.

Key findings.
– Based on nine semi-structured student interviews, the study successfully identified and consolidated six core user requirements, including the need for individualized feedback, multi-modal interaction, and step-by-step explanations.
– The researchers derived and instantiated four design principles into a web-based multi-agent system prototype called ProofTutor, which utilizes specialized language models to separate tasks such as draft analysis, logical error identification, and feedback synthesis.
– Because this is a Research-in-Progress paper, there are no quantitative statistical evaluations or empirical efficacy results regarding the tool's impact on student learning at this stage, representing a key limitation that the authors plan to address in future evaluations.

This research-in-progress study employs an Echeloned Design Science Research (eDSR) approach to establish a design framework for AI-enabled high-risk decision-making systems. Collaborating with a large German financial institution, the researchers conduct semi-structured interviews and a think-aloud study with industry experts to define and validate design goals. The methodology focuses on developing early-stage design objectives and requirements that balance automated technical capabilities with strict regulatory compliance.

The problem. Organizations aiming to automate high-risk operational processes, such as loan approvals, face strict regulatory constraints under frameworks like the EU AI Act and GDPR. These regulations mandate rigorous human oversight, traceability, explainability, and personal data protection, which are difficult to implement using typical black-box AI models. Currently, there is a lack of comprehensive sociotechnical design knowledge to help organizations structure systems that optimize process efficiency while fully complying with these laws.

Key findings.
– The qualitative relevance and completeness of seven design objectives (DO1-DO7) and their corresponding requirements (DR1.1-DR7.1) were validated by domain experts.
– Regulatory, technical, and organizational challenges, specifically regarding data privacy, explainable AI, and human reviewer cognitive overload, were confirmed as major development hurdles (Supported qualitatively).
– Proposed economic challenges were not supported as active adoption barriers, as practitioners viewed automated decision-making as a necessary strategic investment rather than a cost-prohibitive expense (Not supported).
– A fundamental feasibility trade-off was validated between rigorous human oversight (which increases compliance) and process automation (which increases efficiency).
– Limitations: The findings are based on an early-stage, in-progress research project within a single German financial institution and lack summative or quantitative evaluation.

This study details the design, implementation, and preliminary evaluation of the Meeting Process Twin, an AI-enabled artifact designed to discover and analyze meeting-based decision processes. The tool extracts task-level events from multimodal inputs, such as audio transcripts and visual interactions, to map actual meeting progressions and compare them against formal, agenda-based models.

The problem. A significant gap exists in business process management because many crucial organizational decisions are made during meetings that do not generate structured digital event logs. This lack of visibility makes it difficult to systematically observe and analyze informal practices and deviations from formal agendas, which can create accountability and compliance challenges in governance-sensitive sectors.

Key findings.
– The technical feasibility of the Meeting Process Twin was verified across 54 city council meetings, revealing a low mean deduplicated process fitness of 0.314 and a shadow activity prevalence of 41.9% (supported by sensitivity analysis across similarity thresholds from t = 0.15 to t = 0.60).
– Of the detected shadow activities, 95.7% were statistically classified as innovative practices rather than compliance failures, highlighting regular pre-agenda and off-agenda discussion patterns.
– Limitations include that the current iteration is restricted to single-camera recordings and has only undergone preliminary feasibility testing, with full accuracy validation against human-annotated ground truths still in progress.

This study details the iterative design and implementation of an open-access web platform designed to foster multilingual language awareness in early childhood education. Using an echeloned Design Science Research approach, the researchers developed a functional prototype centered on multimodal, self-paced storytelling. The prototype was evaluated qualitatively through interviews and a focus group with educational experts, parents, and software developers.

The problem. Although linguistic diversity is common in early classrooms, digital educational resources are often strictly monolingual or lack pedagogically appropriate support for educators. Existing tools frequently encourage isolated child-device interaction rather than interpersonal engagement. Consequently, educators lack adaptable, open-access tools designed specifically for collaborative multilingual learning in resource-constrained environments.

Key findings.
– The evaluation relied entirely on qualitative feedback from a small sample (8 participants), meaning no statistical hypothesis testing was conducted.
– Evaluators qualitatively supported the platform's usefulness, noting that the simple, minimalist layout effectively reduced visual clutter and helped keep children focused.
– The manual, stepwise story progression was qualitatively supported as a feature that encourages active, discussion-based engagement rather than passive media consumption.
– The design concept of a single hybrid interface for all users was not supported by evaluators, who recommended separate, role-specific entry points for parents, educators, and children to resolve user confusion.
– Study limitations include the preliminary nature of the evaluation, the small sample size of eight participants, and the lack of validation in real-world kindergarten environments.

This study employs an echeloned Design Science Research (eDSR) approach to develop design knowledge for socio-technical systems that support sustainable reuse in low-code development. The methodology integrates a structured literature review of 77 papers with 20 semi-structured expert interviews across various industries. The study aims to construct and validate a foundational framework of design objectives and requirements for low-code reuse.

The problem. Organizations frequently struggle to scale low-code development initiatives because developers consistently recreate existing functionalities instead of reusing them. This issue leads to redundant code, higher maintenance costs, and decreased infrastructure flexibility. The root of the problem lies in the lack of organizational, individual, and procedural support mechanisms rather than limitations in the technical capabilities of low-code platforms.

Key findings.
– A validated problem statement establishing that effective low-code reuse goes beyond technical enablement and requires organizational, individual, and procedural support (supported qualitatively through literature and 20 expert interviews; no statistical hypothesis testing was conducted).
– Five validated design objectives, Systematic Reuse, Multi-Level Reuse, Cognitive Relief, Organizational Alignment, and Inclusiveness, designed to systematically foster low-code reuse (supported qualitatively; no statistical testing was conducted).
– Six specific design requirements, including ensuring platforms are usable for both professional and citizen developers and making reusable artifacts explicitly identifiable (supported qualitatively; no statistical testing was conducted).

This research-in-progress study uses an echelon design science research framework to systematically synthesize findings from 10 empirical studies on fixed-wing manned-unmanned teaming (MUM-T). The authors propose four theory-grounded nascent design principles to establish an initial prescriptive knowledge base for designing human-machine interfaces (HMIs). This foundational work serves as a starting point for future interface instantiation and evaluation with operational pilots.

The problem. Although manned-unmanned teaming is increasingly deployed in civilian and military aviation, the design of human-machine interfaces remains theoretically fragmented and specific to individual systems. Existing cockpit standards and policies lack prescriptive guidelines for managing cognitive challenges unique to airborne multi-platform supervision under severe time constraints. Consequently, designers lack generalized, theory-justified principles to build interfaces that effectively support human-autonomy collaboration.

Key findings.
– Proposed four nascent design principles (Context-Adaptive Automation Intervention, Hierarchical Delegation Granularity, Integrated Temporal-Spatial Plan Visualization, and Bidirectional Human-Autonomy Negotiation) synthesized from prior literature (Not yet empirically evaluated as a combined system in this research-in-progress).
– Justified the mechanisms of the proposed design principles using established kernel theories from cognitive engineering and human-automation interaction (Theoretically supported).
– Noted study limitations, including a small synthesis sample of 10 studies, a focus exclusively on fixed-wing platforms, and the exclusion of classified military data, which may restrict the generalizability of the findings.

This study investigates how digital information systems can be designed for and integrated into IT-distant, care-oriented environments. Using a multi-year, interpretive Design Science Research approach, the researchers developed and evaluated a socio-technical ecosystem in a German all-day school over four iterative design cycles. The evaluation combined qualitative methods including ethnographic field notes, interviews, and workshops.

The problem. Traditional information systems design assumes office-like settings characterized by standardized processes, stable user roles, and uninterrupted interactions. However, care-oriented and educational settings rely on relational, situational work and highly heterogeneous user groups. This mismatch frequently causes digital systems to be resisted, bypassed, or reduced to administrative burdens in practice.

Key findings.
– Stakeholder heterogeneity across staff, administration, children, and parents introduces divergent digital literacies and accountability demands (empirically supported by qualitative field observations).
– Care-oriented work demands constant attention shifts, making brief, interruptible digital interactions a primary design requirement over sustained, screen-centered use (empirically supported by qualitative field observations).
– System viability in care contexts depends heavily on multi-stakeholder emotional alignment and trust, where parental privacy and surveillance concerns are as critical as functional requirements (empirically supported by qualitative field observations).
– The study is limited by its focus on a single German after-school care facility and a specific hybrid terminal and NFC-wristband setup, which may restrict generalizability to purely software-based systems or other care environments.

This study describes the design and initial evaluation of an AI Strategy Companion developed to assist senior executives with strategic decision-making. Using an echeloned design science research framework, the authors built a prototype that structures and provides access to peer-derived, real-world leadership experiences.

The problem. Strategic decision-makers currently lack tailored decision support when navigating continuous organizational change. Because peer-level experiential knowledge is often tacit and fragmented, executives must rely on personal intuition or external consultants rather than systematic insights from peers in similar situations.

Key findings.
– Because this is an exploratory, early-stage research-in-progress study, no statistical hypothesis testing was conducted; findings are purely qualitative and based on a small sample of two senior executives.
– Qualitative feedback supported the feasibility of capturing and structuring executive reasoning through continuous, structured conversations.
– The evaluation supported the concept of the AI as a 'cognitive partner' that facilitates reflection and questioning of assumptions, rather than acting as an autonomous decision-maker.
– The participants supported the value of anonymous, spontaneous access to peer-derived insights, noting that it provides temporal and spatial independence during complex decision-making processes.

This research-in-progress paper proposes an evaluation framework for assessing prescriptive design knowledge within human-Generative AI (GenAI) co-design. The proposed methodology outlines system-level quality criteria and plans to evaluate them using a vignette-based study comparing different information retrieval techniques and prompt engineering levels.

The problem. Existing design science evaluation criteria assume human designers are the sole actors capable of interpreting and adapting abstract design principles. As GenAI tools increasingly act as co-designers, there is an unaddressed research gap regarding how to evaluate whether AI systems can successfully retrieve, interpret, and operationalize design knowledge.

Key findings.
– This is a conceptual research-in-progress paper, meaning no empirical hypotheses have been statistically tested or validated yet.
– The study successfully conceptualizes an evaluation framework comprising eight system-level quality criteria: activation, alignment, explainability, traceability, correctness, consistency, efficiency, and final quality.
– The paper proposes a vignette-based evaluation methodology using two generative information retrieval models (self-RAG and SearChain/MINDER) and three levels of prompt specificity to empirically test the framework in future research.

This study uses qualitative methods to explicate the problem space of customer-dependent incident management in digital support environments. Through semi-structured interviews with seven experts at a major industrial equipment manufacturer, the researchers analyze the operational challenges that arise when service progress relies on customer contributions.

The problem. Existing IT service management frameworks often overlook how customer-dependent readiness impacts service episodes, focusing instead purely on provider-side throughput. This contribution blindness creates operational bottlenecks, as teams lack systematic methods to qualify, monitor, and route tickets based on whether the customer has provided the necessary inputs or access.

Key findings.
– Identified three core challenges in digital incident management: unqualified early contribution signals that leave work readiness ambiguous, ad-hoc tool workarounds that consume specialist capacity, and contribution-blind routing that pushes unprepared cases into expert queues.
– Derived four design objectives for future contribution-aware systems, focusing on early readiness qualification, staged intervention logic, readiness-aware flow control, and transparent decision governance.
– Developed three evaluation criteria, operational benefit, decision defensibility, and stakeholder acceptability, to guide the assessment of subsequent software artifacts.
– Acknowledged study limitations, noting that as a qualitative research-in-progress paper, the findings rely on a small sample of seven interviews within a single industrial organizational context.

This study examines how Process Mining systems can be designed and embedded in organizations to support positive outcomes rather than being perceived as invasive surveillance. Using a design science research approach and an in-depth qualitative case study of a large utility organization, the authors identify key socio-technical mechanisms that mitigate electronic monitoring risks. From these mechanisms, they derive five preliminary design principles to guide the responsible implementation of analytics-based systems.

The problem. Advanced analytics tools like Process Mining offer detailed visibility into operations, but they run the risk of being perceived by employees as stress-inducing electronic performance monitoring. Prior research has focused primarily on the technical capabilities of these tools rather than on how they can be socio-technically designed to avoid triggering negative employee reactions. Consequently, there is a lack of actionable, design-oriented knowledge on how to configure and introduce these systems to foster trust and learning.

Key findings.
– Qualitative analysis supported that negative Electronic Performance Monitoring (EPM) effects did not materialize because Process Mining insights were consistently interpreted as signals of process performance rather than indicators of individual behavior.
– The study supported the formulation of five preliminary design principles: (1) Governance-Driven Trust, (2) Aggregated Process-Level Transparency, (3) Process Focus as a Boundary, (4) Guided Onboarding, and (5) Evidence-Based Steering without individual sanction.
– Empirical evidence supported that restricting visibility to aggregated, pseudonymized processes rather than individual metrics successfully shifted accountability from individual blame to process structures and supported collaborative problem-solving.
– Limitations: As a 'Research in Progress' study, the findings are limited to a single qualitative case study context of a large, highly regulated German utility provider, meaning the design principles require further long-term validation and evaluation.

This research-in-progress study explores how performance measurement systems can be designed to balance technical constraints and dynamic business requirements. The researchers apply imbrication theory as an analytical lens to investigate a case study of a medium-sized IT service provider developing a centralized dashboard platform.

The problem. Designing and implementing performance measurement systems is often resource-intensive, costly, and difficult to sustain due to technical and organizational constraints. A critical research gap exists in understanding how conceptual performance definitions, technical implementations, and organizational arrangements interrelate and influence one another over time.

Key findings.
– The study outlines that performance measurement system design consists of two interlocked subsystems: a technical subsystem (data aggregation and representation) and a conceptual subsystem (the collection of performance measures).
– Qualitative findings from the single case study suggest that material constraints of legacy, scattered data systems act as primary drivers for initiating technological change and system redesign.
– The implementation of the centralized system created new affordances, specifically multi-layered dashboards, which enabled more informed, role-specific executive decision-making routines.
– Because this is an exploratory, qualitative research-in-progress paper, the findings and design principles have not been statistically tested or systematically validated across multiple cases.

This research-in-progress study proposes a theory-driven large language model (LLM) agent framework to generate synthetic patient self-report data from clinician-recorded medical notes. The design integrates three behavioral theories to structure the LLM's workflow, prompt architecture, and a heuristic algorithmic module. The synthetic data's utility is evaluated by using it to fine-tune a ClinicalBERT model for downstream tele-triage predictive tasks.

The problem. Developing artificial intelligence for tele-triage is severely hindered by a scarcity of patient self-report data due to strict privacy regulations and high clinical expert costs. Existing efforts to generate synthetic text lack a robust theoretical foundation, relying instead on fragile prompt engineering that fails to capture the underlying psychological and situational mechanisms of patient reporting. This data mismatch leads to poorly performing, unreliable AI models that cannot be safely deployed in clinical settings.

Key findings.
– The proposed theory-driven multi-agent framework successfully transforms structured clinical notes into naturalistic, synthetic patient self-reports (Supported).
– Downstream predictive models fine-tuned on the generated synthetic data achieved significantly improved performance, with accuracy increasing to 0.715 and F1-score to 0.729, compared to training on original clinical notes (0.632 accuracy, 0.648 F1) (Supported).
– This study is a research-in-progress, and the preliminary findings are currently limited by a lack of extensive clinical validation and comparative evaluations with real patient-reported data (Acknowledged limitation).

This study reports on the ongoing design and initial testing of a redesigned master's thesis format based on Hevner et al.'s Design Science Research (DSR) framework. The proposed model shifts the assessment focus from text-heavy volume to a four-stage, milestone-based process spanning approximately 30 pages to combat the challenges of generative AI.

The problem. The widespread availability of large language models makes fluent academic writing a commodity, breaking the link between polished prose and individual student competence. Traditional, text-heavy master's thesis formats are highly vulnerable to AI exploitation, creating a need for assessment models that make cognitive effort and decision-making traceable.

Key findings.
– The findings are qualitative, context-bound, and not statistically generalizable, as they are based on an interpretive analysis of a pilot with 39 students at a single institution.
– Written text alone was found to be insufficient as evidence of competence; active discussion and defense of design decisions were necessary to verify student understanding.
– Traceability (explicitly linking problem insights to requirements and design choices) emerged as the strongest mechanism for establishing epistemic transparency.
– The staged, four-part milestone structure generated productive progress, with all 39 students in the cohort successfully continuing through the first three stages without dropping out.

This study examines how echeloned Design Science Research (eDSR) can be applied to plan, organize, and coordinate a multi-year, industry-embedded research program. By using the Generative AI-enhanced Knowledge Management (GAIK) project as a use case, the authors integrate eDSR with the University-Industry Linkages (UIL) framework to guide collaborative artefact creation. The methodology combines design science cycles with phased relationship development to systematically build a business-oriented GenAI toolkit.

The problem. Industry-university collaborative innovation projects are difficult to manage due to open-ended problem spaces, multiple stakeholders, and high technical uncertainty. Companies often seek immediate, practical tools while universities target generalized, reusable research findings, creating a tension that linear project management models cannot adequately resolve. Additionally, managing the complexity of diverse intermediate artefacts makes coordination and value articulation challenging.

Key findings.
– Developed a synchronization mechanism that successfully integrates the eDSR methodology with the UIL framework to co-structure design knowledge creation and collaborative relationships.
– Successfully demonstrated the framework's utility across four project iterations of the GAIK project, resulting in the development and initial launch of a modular, six-layer Generative AI toolkit for small and medium-sized enterprises.
– Statistical hypothesis testing was not applicable to this study, as the research used a qualitative Design Science Research (DSR) case study approach to evaluate framework applicability rather than quantitative empirical testing.

This study details the development and evaluation of LLM Buddy, an AI-augmented research environment designed to capture and trace LLM prompts and responses. The prototype environment was evaluated through a longitudinal field deployment spanning a six-iteration Elaborated Action Design Research (eADR) project.

The problem. Conventional version control tools track code changes but fail to record the prompts and decision-making context of AI-assisted development. This gap creates a traceability barrier, threatening the reproducibility, auditability, and rigor of AI-augmented design science research.

Key findings.
– LLM Buddy successfully captured and archived 1,555 distinct AI prompts across multiple language models during a six-iteration field deployment, demonstrating feasibility (supported empirically).
– The tool's automated snapshot system successfully enabled full recovery from an AI-induced file corruption incident, saving estimated development time (supported empirically).
– Analysis of the captured prompt histories supported the identification of a recurring non-linear interaction pattern termed 'Conversational Forking' (supported qualitatively).
– Acknowledged limitation: The evaluation was confined to a single-investigator eADR project, meaning broad cross-context generalizability remains to be verified in future work.

This study designs, instantiates, and evaluates a Human-in-the-Loop Information System (HITL IS) to support field analysts in automotive quality control. Grounded in Cognitive Fit Theory, the researchers developed four design principles and instantiated them in a web-based prototype. The prototype was evaluated through a survey-based study with 20 practitioners to measure cognitive workload and usefulness.

The problem. Automotive aftersales organizations face rising vehicle complexity and high volumes of heterogeneous repair data, making failure classification cognitively demanding. Automated machine learning models alone cannot handle ambiguous data and contextual nuances, yet current tools fail to support interactive collaboration between human experts and models, leading to high manual effort and inconsistent failure detection.

Key findings.
– The prototype demonstrated a statistically significant reduction in perceived mental demand (p = 0.009), time pressure (p = 0.004), and frustration (p < 0.001) compared to a neutral baseline midpoint.\n- Hypotheses that the prototype significantly altered perceived effort (p = 0.132) and perceived success (p = 0.336) compared to the neutral midpoint were not statistically supported.\n- The study is limited by a small sample size of 20 practitioners and reliance on a video-based prototype demonstration rather than a longitudinal field evaluation in a live environment.

This study designs and formatively evaluates a prototype system for agentic lecture production that orchestrates human-AI workflows. The system integrates markdown-based content authoring, automated slide generation, text-to-speech narration, and avatar video rendering into a single coordinated pipeline.

The problem. Educators experience high coordination overhead and tool fragmentation because multimedia lecture elements (slides, audio, video) are produced across disconnected platforms. This leads to version drift, duplicated effort, and a lack of traceability between original source content and final video outputs.

Key findings.
– The formative evaluation was qualitative and context-bound, meaning no statistical hypothesis testing was performed.
– Qualitative feedback from six university teachers indicated that the system successfully improved workflow coherence, traceability, and production speed.
– Operational feasibility was demonstrated during deployment, where five video lectures based on distinct textbook chapters were produced within a single working day.
– Users expressed support for the system's ability to maintain human oversight but noted a desire for more granular control over AI actions, such as specifying the exact number of slides generated.
– The study's findings are limited by a small participant sample size (n=6) and the lack of longitudinal data on long-term adoption.

This study presents the design, development, and evaluation of a mobile application designed to scaffold self-regulated learning (SRL) in higher education. Grounded in established SRL and feedback models, the standalone prototype guides students through planning, enacting, and reflecting on their learning sessions. The design was evaluated using expert feedback, a one-week user trial, and subsequent interviews.

The problem. University students frequently struggle to plan, monitor, and adapt their academic learning processes independently. While digital learning environments can track learning behaviors, existing tools often address only isolated parts of the learning cycle or present descriptive visual data that students struggle to interpret without guidance.

Key findings.
– Field test participants perceived the application as intuitive and useful for structuring their study sessions.
– Feedback visualisations tracking goals, tasks, and focus time were found to be useful for informing subsequent study habits, whereas mood and focus-prompt visualisations were reported as less actionable.
– While notifications appeared to support regular usage patterns for some participants, they were not found to clearly increase intrinsic motivation to learn.
– Due to the pilot study's limited scale (six participants over a one-week period), findings are descriptive and qualitative rather than statistically generalizable.

This study designs and evaluates a mobile decision support system that simplifies and personalizes dynamic electricity signals like prices and carbon intensities for households. The prototype was evaluated in an eight-week staggered field experiment tracking high-frequency smart-meter data. The methodology isolates the behavioral effects of single indicators, an aggregated master indicator, and alert notifications.

The problem. While dynamic tariffs and carbon-intensity variations offer pathways to decarbonization, consumers struggle to understand and react to highly volatile and complex utility data. Most households lack accessible, low-cost tools to translate these abstract signals into timely, actionable consumption adjustments. This cognitive barrier prevents active demand response even when users are environmentally or financially motivated.

Key findings.
– The Single Indicators feature produced a statistically significant cumulative increase in consumption of 44.33 kWh per user (p < 0.01), driven primarily by a highly significant hourly increase of 0.072 kWh during low-carbon green periods (p < 0.01).
– Price and Regional Share Single Indicators had no statistically significant effects on energy consumption, suggesting low price salience in this high-income setting.
– The Master Indicator's overall cumulative reduction of 8.5 kWh was not statistically significant.
– Disaggregated analysis of the Master Indicator revealed that urgent, risk-framed red signals led to a statistically significant hourly consumption reduction of 61 Wh (p < 0.1), whereas green signals had no significant effect.
– This exploratory study is limited by its small sample of 23 Luxembourg households, requiring larger-scale studies to confirm generalizability.

This study designs and evaluates AIMQA, an interactive conversational assistant powered by an LLM to support non-technical domain experts in assessing AI model quality within low-code/no-code (LCNC) platforms. Grounded in sensemaking theory, the companion translates technical performance metrics into context-aware, domain-specific explanations.

The problem. Although LCNC platforms democratize AI development, domain experts struggle to evaluate model quality because performance metrics are presented in technical, decontextualized terms. This lack of transparency forces reliance on technical specialists and hinders domain experts from confidently trusting, interpreting, and refining their models.

Key findings.
– A qualitative evaluation with 20 participants supported that linking quantitative metrics to concrete classification outcomes (such as a confusion matrix) was clear and understandable.\n- The LLM-based interactive dialogue was qualitatively supported as useful for helping users uncover the causes of specific model behaviors.\n- No participants rated the artifact's interpretability support as difficult to understand.\n- Limitations of the study include its small sample size of 20 participants evaluated in a qualitative, specific KNIME and Snowflake environment, alongside user demands for more explicit guidance on model refinement steps.

This study design and evaluates the Reproducible AI Documentation (RepAID) tool, which is integrated into the KNIME Analytics Platform. The tool extracts ML workflow metadata and utilizes Large Language Models (LLMs) to automatically generate standardized reproducibility reports. The researchers evaluated the prototype through comparative lab testing and semi-structured interviews with target users.

The problem. While low-code machine learning tools enable non-technical researchers to implement complex models, these platforms often lack transparency and code-sharing capabilities. This makes it difficult for researchers to satisfy rigorous journal and conference documentation standards manually. The research gap lies in the lack of automated, customizable tools to bridge the mismatch between low-code ease of use and strict reproducibility requirements.

Key findings.
– Lab evaluations qualitatively demonstrated that the RepAID tool successfully generated accurate, structurally consistent documentation that detailed internal workflow processes better than manual records (statistical significance not tested due to qualitative design).
– Qualitative feedback from six researchers supported the tool's ease of use and potential to reduce manual documentation effort, though results are restricted by the small sample size.
– Evaluation revealed limitations including the tool's inability to document external context (like dataset sources or hardware configurations) and a potential risk of user automation bias.

This study introduces TerrainGrade, a design science artifact that generates cell-level flood susceptibility estimates using open national hydrologic data. The methodology leverages a machine learning pipeline implemented on an H3 hexagonal grid to produce continuous flood probability indexes. The artifact was evaluated through a pilot implementation covering 2.2 million spatial cells in Maryland.

The problem. Traditional flood maps are costly to update and present risk as binary zones, making them difficult to integrate with modern geospatial and machine learning workflows. Furthermore, existing hydrologic models do not directly yield cell-based inundation layers suitable for continuous hazard assessment. This creates a gap for planners and insurers who require scalable, granular, and continuous flood risk screening.

Key findings.
– The machine learning pipeline demonstrated high predictive performance under spatial cross-validation, achieving out-of-fold AUC values of 0.984, 0.978, and 0.969 across three increasing flood severity thresholds (statistically supported).
– External validation against FEMA maps confirmed that designated flood zones received statistically significantly higher susceptibility scores (mean increase of 0.198, Welch t = 131.9, Cohen's d = 0.48), despite FEMA data not being used for training (statistically supported).
– The pilot study is limited to the state of Maryland and exhibits reduced predictive stability in flat coastal terrains, meaning further multi-state data is required to generalize findings across different geographic regions (acknowledged limitation).

This study introduces PaperMate, a canvas-based reading support system designed to help non-native academic readers comprehend complex scientific texts. Grounded in the Active Reader View, the prototype integrates simplification, structural guidance, key info extraction, and conversational clarification into a unified, user-controlled interface. The system was developed and refined using iterative design science research cycles, including laboratory and field testing.

The problem. Non-native English speakers face significant linguistic, structural, and conceptual barriers when engaging with English scientific literature, leading to comprehension errors and longer reading times. Existing reading assistants typically offer only isolated tools (such as translation or summarization) in fragmented interfaces, which fails to support a deep, holistic understanding of the text. Additionally, educational institutions lack the scalable resources needed to offer personalized reading support to these students.

Key findings.
– The study successfully designs and implements PaperMate, a unified canvas-based prototype that operationalizes five core design principles to facilitate text comprehension.
– Specific statistical hypothesis testing and quantitative performance outcomes from the laboratory evaluations (with 140 and 155 participants) are not reported in this article, as the text focuses primarily on the prototype's architectural design and development process.
– Preliminary proof-of-value evaluations qualitatively supported the system's feasibility and perceived usefulness, though comprehensive field validation in real-world environments is still ongoing.

This study designs, implements, and evaluates a theory-grounded, practice-oriented AI literacy course prototype for novice adult learners outside formal educational settings using Action Design Research. Grounded in a digital literacy framework and Dynamic Skills Theory, the multi-session course uses low-tech metaphors and an hourglass instructional design to transition learners toward self-directed AI use.

The problem. The widespread public availability of generative AI has made AI literacy essential, yet nearly half of adults in the EU lack basic digital literacy, leaving them underserved. Existing educational resources either lack theoretical grounding to build accurate mental models or focus strictly on formal educational environments rather than independent adult learners.

Key findings.
– In self-reported MAILs data, learners' AI literacy scores consistently improved from <=2 points to >=8 points on a 0-10 scale across course sessions (statistically supported).
– Performance tests and chatbot interaction logs confirmed that participants successfully progressed from basic search-like inputs to structured multi-turn conversation and prompting strategies (statistically supported).
– Acknowledged Limitations: The findings are based on a small, self-selected sample of participants (initial cohort of 12 older adults, followed by three similar small-scale iterations), which limits the immediate scalability and generalizability of the prototype.

This study presents and evaluates LitFlow, a web-based platform designed to support systematic literature reviews through transparent, researcher-controlled AI augmentation. Developed using design science research, the platform integrates multi-database searching, screening, data extraction, and audit-trail generation in a single workspace. The authors conduct a qualitative, formative evaluation using think-aloud interviews with five researchers to assess usability and workflow completeness.

The problem. Conducting systematic literature reviews is traditionally a highly time-consuming process that depends on fragmented, non-interoperable toolchains. While emerging AI-based tools attempt to streamline this process, they often prioritize automation over methodological transparency, which compromises researcher agency, decision auditability, and rigor.

Key findings.
– The evaluation was purely qualitative and exploratory, based on a limited sample of five researchers, meaning no statistical hypotheses were quantitatively tested.
– Participants consistently reported that the integrated workflow provided high value, specifically the multi-database search feature.
– Participants indicated they felt in full control of decisions, viewing the AI as a restrained assistant rather than an autonomous decision-maker.
– AI support in query refinement was qualitatively perceived as less useful than AI assistance during screening and data extraction.
– Participants raised concerns regarding potential AI anchoring effects, suggesting that viewing AI recommendations first could bias independent researcher judgment.

This study presents the design and evaluation of a web-based, hybrid intelligence Decision Support System (DSS) developed in collaboration with a large German wholesaler to assist in B2B pricing. The system utilizes K-Nearest Neighbors as a Case-Based Reasoning mechanism to provide frontline sales agents with transparent, transaction-specific price corridors. The artifact was evaluated through an offline simulation of 160,000 transactions and field-validated with experienced sales agents.

The problem. B2B pricing is frequently delegated to frontline sales agents who default to excessive discounting under time pressure, causing margin erosion. Standard data-driven automation is hindered by severe data censoring, as historical databases typically do not record rejected quotes, rendering optimal price points unrecoverable. Consequently, companies must find a way to guide agent negotiations without relying on opaque black-box optimization or rigid, non-adaptive static floor-and-target rules.

Key findings.
– An offline simulation of 160,000 recent transactions demonstrated estimated revenue uplifts of 8.2% to 12.0% across simulated negotiation scenarios (statistically supported in simulation).
– The same offline simulation demonstrated estimated profit uplifts of 26.8% to 38.9% across simulated negotiation scenarios (statistically supported in simulation).
– Field validation with experienced sales agents confirmed that the generated pricing corridors were plausible, realistic, and aligned with practical negotiation workflows (supported).
– A limitation is that the evaluation relies on offline simulation and is situated within the context of a single German finishing trades wholesaler; actual real-world transactional outcomes remain to be fully integrated and validated in live field environments.

This study explores the collaborative design and development of a Virtual Reality (VR) de-escalation training system for onboard train managers. By integrating Design Thinking within an Action Design Research framework, the researchers developed and evaluated low-cost Minimum Viable Products to capture complex real-world practices before committing to full VR development.

The problem. Onboard train staff face a high risk of passenger aggression within restricted and moving train carriages, but authentic practice opportunities are rare. Traditional classroom training fails to simulate the psychological stress and physical constraints of these encounters, while developing high-fidelity VR scenarios is typically cost-prohibitive and technically rigid.

Key findings.
– Qualitative evaluation supported the prototype's high utility and realism, with 21 experienced practitioners confirming its effectiveness as a safe space to experiment with different responses, although formal quantitative statistical testing was not conducted.
– Technical evaluation validated that the modular, JSON-based scenario architecture successfully decoupled content from code, enabling rapid, low-cost iterations without requiring engine re-programming.
– Limitation: The study is limited by its specific context (Sweden's largest train operator) and a small qualitative sample size of 21 participants, who indicated that the system would benefit from a wider variety of interactive choices.