Abstract Motivation: Addressing today’s social, environmental, and technological challenges requires not only strong strategies but also approaches that coordinate stakeholder activities across ecosystems. While many tools help multi-stakeholder partnerships analyze the past and envision desirable futures, stakeholders often struggle to translate those visions into actionable steps because their goals and values in the present are misaligned. This misalignment leads to planning paralysis or fragmented efforts. The Collective Action Planning (CAP) method: CAP is a facilitation method that integrates diverse stakeholder goals and perspectives to create complementary, actionable solutions. Grounded in systems thinking and pragmatism – particularly the concept of simplexity, which balances complex understanding with simple action – CAP connects requirements, constraints, and activities across ecosystems, emphasizing how pilot experiments interact with existing structures. Research Design: Developed through an echeloned Design Science Research (eDSR) process involving sixteen validation cycles in seminars, trainings, and projects, CAP now comprises five steps. Practitioner benefits: We illustrate CAP use through the case of building a circular ecosystem for reusable to-go cups. CAP enhances sensemaking and coordination in multi-stakeholder partnerships, helping actors move from shared vision to collective action – in short, it is a method for acting together.
Train managers face a challenging work environment, characterized by a heightened risk of passenger aggression within spatially confined environments. Further, the rarity of threatening incidents makes skill development difficult. This paper presents the design and development of a Virtual Reality (VR) training artifact aimed at equipping train managers with de-escalation skills through immersive VR simulation of threatening onboard scenarios. Developed in collaboration between SJ (Sweden's largest train operator), University West, and VR developer Tenstar, the project operationalizes Design Thinking (DT) as the micro-level method within an Action Design Research (ADR) framework. Two Minimum Viable Products (MVPs) were developed as boundary objects to surface tacit practitioner knowledge and validate scenarios before committing to costly VR development. A modular JSON-based architecture decouples scenario content from the VR implementation, enabling rapid iterations without modifying the underlying VR environment. Evaluation indicates strong perceived utility and confirms that the design process successfully captured the complexity of real onboard work practices. The paper contributes a methodological approach for developing VR training artifacts in complex, practice-embedded contexts.
B2B pricing in high-velocity wholesale environments is typically delegated to frontline sales agents, yet time pressure and information asymmetry lead to systematic over-discounting that erodes margins. While data-driven automation promises a remedy, the absence of rejected quotes from historical transaction data creates a fundamental censoring problem: the margin-maximizing price is likely unrecoverable, and black-box optimization cannot be validated against the true business objective. Static floor-and-target rules, the dominant alternative, ignore transaction-level heterogeneity. We present a hybrid intelligence Decision Support System developed with a large German finishing trades wholesaler processing over 3 million annual transactions. The artifact uses K-Nearest Neighbors as a Case-Based Reasoning mechanism to retrieve historical transactions with comparable elasticity profiles, defined by customer characteristics, product attributes, order size, and market timing, and constructs transaction-specific price corridors from the empirical discount distribution of the matched set. Glass-box retrieval enables domain experts to inspect neighbors, assess comparability, and iteratively recalibrate similarity weights, treating model output as a refinable heuristic rather than an authoritative optimum. An offline simulation on 160,000 recent transactions estimates revenue uplifts of 8 to 12
As Large Language Models (LLMs) become integral to design science research, a new software paradigm known as ‘promptware’ has emerged. In this paradigm, the prompts that drive artifact development are themselves critical, first-class software artifacts requiring systematic documentation and engineering rigor. However, traditional version control tools capture only code changes, leaving the rationale behind AI-assisted decisions invisible and irreproducible. LLM Buddy is a desktop-based research-instrumentation environment for eADR that captures prompts and LLM responses across interfaces, links them to the files they shaped, and preserves the resulting design rationale as an auditable, rollbackable research trace. It provides multimodal prompt capture, prompt-to-file association, automated rollback, and structured session records for eADR reporting. Developed and evaluated through six Elaborated Action Design Research (eADR) iterations, LLM Buddy captured 1,555 prompts during a longitudinal field deployment at the University of South Florida, supporting the identification of recurring prompt interaction patterns, including what the authors term Conversational Forking, and aiding recovery from a critical file corruption incident through its automated versioning system.
Detecting emerging failure patterns in automotive aftersales requires analyzing large volumes of heterogeneous repair data at very low initial occurrence rates. Machine learning can cluster similar failures, but domain expertise is crucial for early novelty detection in context-dependent environments. We present a Human-in-the-Loop Information System that integrates expert feedback into model-driven clustering. Using Design Science Research, we derived four design principles based on Cognitive Fit Theory and practitioner insights, instantiated in a web-based prototype. In empirical evaluation, field analysts indicated that aligning cluster information with analytical tasks reduced cognitive load and enhanced reasoning. This study contributes design knowledge for Human-in-the-Loop Information Systems, demonstrating how human expertise and model-driven approaches can collaborate effectively to support failure detection.
Higher education teachers need to produce lecture material in multiple formats (slides, notes, audio, and video), but workflows are fragmented across tools and roles. We present a prototype artifact for agentic lecture production, developed and evaluated through a design science process. The core design contribution is to conceptualize each lecture as a computational boundary object that centralizes the transformation chain, preserves process state, supports reproducibility, and makes human–Artificial Intelligence (AI) transitions explicit and traceable. The artifact integrates markdown-based authoring of slide headers, bullet points, and speaker manuscripts, template-aware slide generation, lecture management, Text-To-Speech (TTS), avatar/video pipeline orchestration, and export management in an orchestrated production pipeline. We describe the artifact design, its research and practice significance, and a formative evaluation focused on feasibility, usability, and operational value. Results indicate that the prototype improves workflow coherence and traceability, supports faster iteration from content drafting to multimedia output, and contributes preliminary design knowledge for human-in-the-loop, AI-supported educational production systems.
Flood hazard information in the United States is conveyed primarily through regulatory floodplain maps that are costly to update and difficult to integrate with modern geospatial analytics. This paper presents TerrainGrade, a design science artifact that generates cell-level flood susceptibility estimates using open national hydrologic data within a machine learning pipeline implemented on an H3 hexagonal grid (approximately 85 m cells). TerrainGrade estimates inundation likelihood across multiple flood scenarios and aggregates the results into a Flood Susceptibility Index (FSI). Training labels are derived by converting modeled river flows to water levels and comparing them with local terrain elevation using the Height Above Nearest Drainage (HAND) metric. In a pilot covering 2.2 million cells in the US state of Maryland, spatial cross-validation demonstrated out-of-fold AUC values of 0.984, 0.978, and 0.969 for increasing flood severity thresholds. Defined FEMA flood zones receive substantially higher susceptibility scores despite not being used for training. TerrainGrade provides a scalable method for generating continuous flood susceptibility surfaces from open federal datasets, enabling flood risk screening and resilience analysis across large geographic domains.
Systematic literature reviews (SLRs) are central to rigorous research but remain resource-intensive and dependent on fragmented toolchains. At the same time, artificial intelligence (AI)-based support for review tasks often lacks transparency and limits researcher control. This paper presents LitFlow, a web-based platform for AI-augmented SLRs developed following the echeloned design science research (eDSR) methodology. LitFlow integrates multi-database search, criteria-based screening, structured data extraction, and audit-trail generation within a single workspace. Its augmentation approach provides AI recommendations with confidence scores, justifications, and source references, while final decisions remain with the researcher. The platform is built on a community-extensible architecture. A formative evaluation with five researchers confirmed the perceived value of the integrated workflow and the augmentation-oriented design. Participants also raised socio-technical concerns, including potential anchoring effects from AI recommendations, which inform directions for future iterations. LitFlow contributes a working demonstration of transparent, researcher-controlled AI support across the full SLR workflow.
This paper presents the design, implementation, and simulation-based evaluation of a multi-agent agentic AI system that supports job seekers across the full lifecycle of a job search. Applying the elaborated Action Design Research (eADR) method, we diagnose the challenges candidates face within AI-driven recruitment systems and design a governed, hybrid multi-agent architecture integrating ten specialized LLM-powered agents, a coordinator agent, multi-tier memory, and human-in-the-loop (HITL) oversight. We implement the system as a 17,000-line prototype using Python, LangGraph, Claude Sonnet, PostgreSQL, and a React dashboard, then evaluate it through simulation using actual candidate profiles with live LLM calls. Results confirm the architecture correctly orchestrates agents, enforces governance, and produces actionable outputs across career alignment, job discovery, application orchestration, and interview preparation. The source code is publicly available on GitHub.
The diffusion of generative AI has made AI Literacy a critical societal issue, yet a large proportion of novice adult learners remains underserved by existing educational approaches. Existing approaches either lack theoretical grounding, fail to reach novices with low digital literacy, or do not address adults outside formal learning contexts. To address this gap, we design and evaluate a theory-grounded, practice-oriented AI Literacy course for novice adult learners outside formal learning settings using Action Design Research. The prototype implements a multi-session course artifact including guided instruction and hands-on interaction with LLMs to transition learners without prior experience toward AI-assisted, self-directed learning. Grounded in a digital literacy framework and Dynamic Skills Theory, the artifact aims to iteratively develop AI Literacy as critical knowledge and functional skills. Distinctive design features, such as low-tech metaphors, are implemented to create an end-user-oriented course structure and novice-friendly communication of content. The prototype was evaluated in multiple field iterations with novice adult learners. Longitudinal data was collected using a mixed-methods design, combining self-reported and behavioral data. Results consistently indicate substantial improvements in interaction strategies and conceptual understanding of AI across heterogeneous learners. We provide a reusable course design and synthesize five design principles to support the design and refinement of courses, especially those for novice adult learners outside formal learning settings.
The rapidly evolving enterprise resource planning (ERP) market demands innovative educational approaches that leverage artificial intelligence (AI) capabilities. Current ERP education faces significant challenges: educators struggle with complex system functionality while students require personalized, hands-on education that many higher education institutions cannot efficiently provide. This research in progress explores how AI can enhance current practice-based ERP education curricula to best benefit both student learning and alleviate pressure on instructors. Following a design science research approach, we leverage discriminative AI capabilities to design and implement a conversational ERP teaching assistant (TA). Our solution integrates with SAP S/4HANA to allow for real-time, personalized support based on students’ input data. Preliminary evaluation based on user tests and expert feedback confirms that our ERP TA fosters learning by helping to troubleshoot students’ work, which saves instructors significant time and mental resources. Conversational rigidity and remaining gaps in theoretical business process content remain limiting factors. As our research continues, we plan to expand the curriculum base and incorporate generative AI capabilities to enhance conversational abilities.
This paper presents a design framework for energy management systems (EMS) that positions openness and information security as interdependent dimensions, jointly shaping innovation in digital energy ecosystems. Rather than considering them as opposing forces, the framework frames them as characterized by a dynamic interplay that requires continuous recalibration in response to evolving technological and organizational contexts. Accordingly, we conceptualize openness as a facilitator of innovation, interoperability, and data-driven collaboration among diverse actors, including utilities, regulators, technology providers, and end-users. Yet, we also acknowledge that openness amplifies vulnerability by increasing the number of interfaces, data exchanges, and system dependencies, thereby exposing EMS to potential external exploitation.
Households are increasingly expected to adjust electricity use in response to dynamic system conditions, yet most lack the information needed to interpret fluctuating prices, carbon intensities, and generation patterns. Addressing this gap, we designed a smartphone application that simplifies, contextualizes, and personalizes key system signals to support intuitive household decision-making. Following a multi-year design science research process, we developed a customizable master indicator, complementary single indicators, and alert-based cues. We evaluated these features in a staggered field experiment with high-frequency smart-meter data from 23 households in Luxembourg. The results provide exploratory evidence that environmental cues can activate selective pro-environmental behavior, price responsiveness remains limited in high-income contexts, and urgent, risk-framed signals trigger stronger behavioral responses than encouragement-based cues. The study advances understanding of transparency-based digital interventions in demand response settings and offers actionable guidance for practitioners and policymakers seeking to make dynamic consumption more accessible.
Older adults’ independent mobility enables out-of-home participation, well-being and health, yet pedestrian navigation systems still optimize primarily for distance or time, often overlooking barriers, safety thresholds, and supportive infrastructure that shape late-life walking decisions. We present a senior-friendly pedestrian routing artefact developed through echeloned Design Science Research, translating lived mobility constraints into prescriptive design knowledge. Based on 11 semi-structured interviews, we derive initial Design Requirements (DRs) and Design Principles (DPs) for barrier-aware, amenity-sensitive routing and execution-relevant explanations. We instantiate these in an OpenStreetMap pedestrian network enriched with amenities (benches, toilets, and shelters) and height data, and implement an A*-based routing engine with configurable costs and explanation payloads. In a field-based walking study, 14 older adults compared artefact-generated routes with baselines and provided ratings and qualitative feedback; the senior-friendly route was preferred overall. Thematic analysis further showed that infrastructure maintenance, seasonal conditions, traffic exposure, and social context shape route acceptance. We synthesize these insights into refined DRs and DPs emphasizing context-aware hazard modeling, multi-route transparency, landmark-grounded explanations, social-context sensitivity, and stage-appropriate information. Our contributions provide actionable guidance for practitioners developing senior-friendly pedestrian navigation systems.
Generative artificial intelligence (GenAI) challenges several assumptions that have long underpinned design science research (DSR), including that artifacts can be clearly bounded, evaluation results can be treated as evidence about stable contributions, and prescriptive knowledge can travel across contexts with limited boundary specification. This paper critically reviews how empirical DSR studies engage with these tensions when designing GenAI artifacts. Using a PRISMA-guided systematic literature review, we analyze the emerging corpus through a configuration-centric lens that examines how studies define the artifact, stabilize evidence, formulate reusable knowledge, and address governance in use. The review shows that most GenAI artifacts are not bounded tools but socio-technical configurations that combine models, prompts, retrieval mechanisms, orchestration logic, and human oversight. It also identifies recurring weaknesses, including under-specified artifact boundaries, limited configuration disclosure, unstable evaluation evidence, incomplete portability conditions, and governance concerns that are acknowledged but rarely operationalized as design features. Based on these findings, we argue that GenAI should be treated as a boundary condition that intensifies key DSR challenges and requires adaptations in artifact conceptualization, evaluation, reporting, and the formulation of reusable design knowledge.
Archetypes are widely used in information systems (IS) research to structure and interpret complex socio-technical phenomena. Despite their widespread use, however, archetype development often lacks methodological rigor. Many studies rely on ad hoc and situational configurations of methods, which limit comparability and transparency. To address this gap, we developed a taxonomy of 114 archetype development approaches identified through a systematic literature review in the IS domain. The taxonomy captures common contexts of use, objects of analysis, and procedures employed in archetype construction. The analysis reveals dominant and less common archetype development practices clarifying how archetypes are understood and constructed in IS and giving guidance for future studies. Beyond its theoretical contribution, the taxonomy also supports practical applications, such as constructing archetypes of IT services as a structured solution space. Overall, our taxonomy strengthens the methodological foundations of archetype development and provides reusable design knowledge for future studies in IS and related fields.
Design Science Research has given the field a rigorous foundation for building and evaluating artifacts, but it has never fully solved the method-selection problem. Researchers know the stages. They know the evaluation criteria. What they often lack is structured guidance on which generative design methods to apply at each point in the process, and that gap quietly undermines both the creativity and the transparency of the work. This paper addresses that gap directly. The hybrid DSR framework introduced here embeds Kumar's seven modes of innovation within each stage of Peffers et al.‘s six-step process model, anchored by Hevner et al.‘s DSR guidelines and Gregor and Hevner's contribution types and artifact levels. The central question driving the work is straightforward: can a structured integration of Peffers’ process with Kumar's innovation modes reduce method-selection friction and support more balanced, creative rigor in complex research environments? The answer, grounded in four completed DSR cycles and structured expert interviews, is yes; however, the mechanism matters. The problem is not the tension between rigor and creativity. It is the absence of curated, stage-aligned method options that leaves researchers defaulting to implicit choices at precisely the moments when deliberate ones would produce better outcomes. The framework operationalizes innovation not as a departure from process; rather, as the deliberate selection of human-centered, generative methods at each DSR stage. The DSR. Navigator, a prototype AI-driven tool, instantiates the revised framework in practice. This research-in-progress contributes a mid-level method artifact to the DSR community and sets the stage for rigorous validation and broader platform development.
Demographic changes are driving an urgent need for Ambient Assisted Living (AAL) technologies to support independent living for the elderly. A critical capability of such systems is the automated detection of health-related anomalies in daily routines, potentially serving as a partial replacement for human monitoring. While Artificial Intelligence (AI) offers promising solutions, the application of Machine Learning (ML) in this domain is severely hampered by the scarcity of labeled reference data, rendering supervised approaches impractical. Furthermore, unsupervised models often operate as “black boxes,” lacking the interpretability required for caregivers to trust and act upon alerts. Following Peffers et al.’s Design Science Research (DSR) methodology, this paper presents the design, development, and evaluation of a web-based artifact that utilizes unsupervised algorithms (Isolation Forest and LSTM-Autoencoder) to detect anomalies in unlabeled sensor data. To ensure practical relevance and trust, the system integrates Explainable AI (XAI) using SHAP values to contextualize alerts. The artifact was evaluated through a dual-perspective approach: a quantitative analysis of model agreement on unlabeled data, and a qualitative study involving a focus group with AAL experts ( n=6 ) and a technical review by a domain specialist. The results highlight the tension between algorithmic precision and human interpretability, contributing validated design guidelines for trust-aware, unsupervised monitoring systems.
This paper addresses the critical challenges faced by organizations in designing and managing digital ecosystems within the ongoing digital transformation. Previous research has made notable progress in tackling technical and infrastructural concerns yet practical implementation, value creation, the integration of diverse partners remain underexplored and difficult to operationalize. Existing design-oriented research methods, such as Action Design Research (ADR), provide a strong foundation but lack a stepwise, process-oriented approach that fully recognizes the active roles of researchers as facilitators, knowledge providers, and trust anchors in the context of digital ecosystems. Moreover, current methods do not sufficiently accommodate the heterogeneity of ecosystem partners and the emergence of scientific artifacts during ecosystem formation. To bridge these gaps, we propose the Testbed Research approach, an adaptation of ADR specifically tailored for the complex realities of digital ecosystem development. Drawing on empirical evidence from 30 ecosystem initiatives conducted between 2016 and 2026, we demonstrate the methodological innovations and practical benefits of Testbed Research. Our findings contribute to both theory and practice by offering a robust framework that supports sustainable value creation and the effective orchestration of digital ecosystems, ultimately bridging the divide between academic insight and organizational practice.
Design principles (DPs) are a central vehicle for prescriptive design knowledge in design science research (DSR), yet researchers still struggle to develop DPs that are both actionable and projectable. We propose a conceptual scaffold for understanding DP development as navigation in a design knowledge space. It integrates (1) a mode-based framework distinguishing modes of framing, reflection, and synthesis, (2) an abstraction-alignment model showing why design requirements, DPs, and design features must align at compatible abstraction levels, and (3) navigation moves that characterize recurring epistemic actions involved in shifting across abstraction levels and relating problem and solution knowledge. We illustrate the scaffold by reconstructing a project on virtual companionship, showing how a candidate DP evolved through feature abstraction from an existing system and how alignment stabilized the DP. Overall, the paper offers orienting guidance for DP development by helping researchers make abstraction choices, conceptual alignment, and reasoning moves more explicit, thereby supporting transparent and cumulative design knowledge construction.
