1. Introduction
Strategic planning today takes place in conditions of high uncertainty and complexity, which are further intensified by growing dynamism and ambiguity (Reference HeadHead, 2022). In such environments, forecast-centric decision models have well-documented limitations, as we cannot forecast or predict what will happen, and can only understand why things happen in retrospect (Reference Snowden and BooneSnowden & Boone, 2007). As a result, relying solely on trend following and extrapolation is no longer a sufficient basis for robust strategies (Reference Bradfield, Wright, Burt, Cairns and van der HeijdenBradfield et al., 2005). Scenario methods are becoming more prominent because they open up possibility spaces through future-oriented and networked reasoning (Reference Bradfield, Wright, Burt, Cairns and van der HeijdenBradfield et al., 2005). Scenario processes, particularly within small and medium-sized enterprises, are frequently considered too complex and in need of extensive explanation. Time and cost pressures also inhibit adoption, despite the rising demand for support (Reference Cornelisse and van KlinkCornelisse & van Klink, 2024). Conversely, compact “quick methods” often fail to accurately represent the actual problem landscape as they omit key drivers or interdependencies (van der Heijden, 2005) (Reference Fink and SiebeFink & Siebe, 2011). Generative AI (GenAI) shifts the frontier in this context. Some aspects of knowledge work can be automated to reduce time and costs, and tasks can be augmented to improve output quality. Field evidence indicates efficiency and quality gains in knowledge-intensive work. For example, a controlled BCG study reported more tasks completed, shorter completion times, and higher output quality under AI assistance (Reference Dell’Acqua, McFowland, Mollick, Lifshitz-Assaf, Kellogg, Rajendran, Krayer, Candelon and LakhaniDell’Acqua et al., 2023). In terms of scenario practice, AI can systematically identify factors, propose projections, support consistency assessment, draft scenario descriptions, and continuously monitor indicators. Meanwhile, interactive workshop formats are emerging that provide real-time support for projections, consistency checks and consequence analysis (Reference Ködding, Dumitrescu and HartmannKödding & Dumitrescu, 2022) (Reference Geurts, Gutknecht, Warnke, Goetheer, Schirrmeister, Bakker and MeissnerGeurts et al., 2022).
In practice, scenario management plays a critical role in long-term product strategy and portfolio decisions. Companies use them to anticipate shifts in customer needs, technologies, regulation, and competition, aligning product roadmaps and life cycles accordingly.
However, in many SMEs, scenario work is still workshop-centered, PowerPoint-driven, and largely offline. Classical scenario processes usually occur as facilitated workshop sequences, including scoping, identifying factors, building projections, assessing consistency, and synthesizing narratives. These manual steps are recorded in static documents with limited structural connection between drivers, projections, and subsequent strategic decisions. Consequently, traceability, iterative refinement, and digital continuity across decision stages are weak.
In contrast, we present an AI-enabled, modular approach to end-to-end scenario management for small and medium-sized enterprises (SMEs). The proposed system addresses this gap by conceptualizing scenario management as a continuous, digitally supported design process in which AI components assist specific analytical tasks, while human actors retain responsibility for framing, plausibility assessment, and strategic judgment. In this sense, scenario management can be understood as a design capability that supports developing future-robust products in uncertain conditions. The core is a combination of a scenario wizard for low-barrier step-by-step use and an integrated, cloud-based expert system for deep parameterization. Functionally, the approach covers all stages of scenario management—from scenario field definition, factor analysis, projection creation, and consistency assessment to scenario development, narrative elaboration, impact analysis, and evaluation. The modules are coupled and decoupled via clear interfaces to facilitate reuse and future development. The UX principles emphasize graspability and intelligibility, including scenario one-pagers, future-space maps, and uncertainty signals.
2. Research methodology
This study adopts the Design Research Methodology (DRM) as outlined by Blessing and Chakrabarti (Reference Blessing and ChakrabartiBlessing & Chakrabarti, 2009). The DRM approach is structured into four phases: Research Clarification (RC), Descriptive Study I (DS-I), Prescriptive Study (PS), and Descriptive Study II (DS-II). This structure guides the work from problem framing to evaluation. The Research Clarification (RC) was conducted as a structured, concept-centric literature review following (Webster & Watson, 2002), focusing on the foundations of the scenario method, quality anchors (morphology / consistency), participation / communication / learning, governance / provenance / reuse, digital / AI support, and SME constraints. The RC synthesized a problem map and initial design implications to guide DS-I and the conceptual PS presented in this paper. Empirical PS/DS-II will be addressed in future work. In DS-I, we conducted a structured review of scenario methodologies and software practices and examined literature on AI support for foresight and scenario work. The review covered process variants (e.g., factor identification, projection generation, consistency assessment, scenario synthesis, communication formats) and examined where current tools fall short for smaller organizations. Based on the analysis, we identified requirements for a modular architecture, clear service interfaces, explainable outputs, and knowledge reuse across projects. These findings inform the Problem Analysis and State of the Art sections and provide the benchmark for later evaluation. In Section 5, the PS is presented as only a conceptual specification reflecting the current status. The solution is defined as a wizard and expert environment on a knowledge backbone but is not yet implemented. A field evaluation of DS-II in the mechanical engineering use case is planned to compare the concept with conventional tool chains, conduct ablation analyses of AI modules and assess the effects of longitudinal reuse. The results of the PS implementation and the DS-II evaluation will therefore be presented in a different paper in the future.
3. Problem analysis
To identify the problems, a structured, concept-centric literature review was conducted in Google Scholar, until conceptual saturation was reached. First, titles and abstracts were screened for conceptual relevance. Then, targeted full-text assessments were conducted on the most pertinent contributions. The evidence and design implications found were captured in a structured extraction sheet (N = 24 final candidate sources). Of these, 21 were selected and cited in the final manuscript based on their conceptual fit to Sections 3 (Problem Analysis) and 4 (State of the Art). The search string used is the following: (((“scenario planning” OR “scenario method*” OR “scenario technique*” OR “scenario management”) AND (SME* OR “small and medium*” OR “small firm*” OR “small business*”) AND (barrier* OR challenge* OR adoption OR “resource constraint*” OR cost OR time OR complexity)) OR ((“artificial intelligence” OR AI OR “machine learning” OR “natural language processing” OR “knowledge graph*” OR “topic model*” OR “text mining” OR “gen* AI” OR “large language model*” OR LLM*) AND (foresight OR “horizon scanning” OR “technology intelligence” OR “scenario planning” OR “scenario generation” OR “consistency check*”))).
Scenario work broadens strategic thinking under uncertainty. However, its practical usage remains constrained by a persistent gap between methodological ambition and organizational reality. Although it promises enhanced decision premises and learning, it is often experienced as opaque, resource-intensive, or poorly tailored to specific contexts (Reference ChermackChermack, 2022). Building on this evidence, we identify five problem areas: P1 – Perceived complexity and low graspability: Scenario approaches require participants to reason across multiple uncertainties, drivers, and interdependencies — a task that many stakeholders find challenging without clear guidance. Classical texts emphasize that scenarios are structured lenses for exploring plausible alternative futures, not forecasts. Without a shared frame, discussions can easily become fragmented or revert to linear extrapolation (Reference SchwartzSchwartz, 2007). Organizations often struggle to keep scenario conversations coherent and anchored in decision-making needs (Reference ChermackChermack, 2011). P2 – High effort and cost barriers, particularly for SMEs: The time and facilitation expertise required, as well as the need for workshops and analysis, creates adoption frictions. Various accounts note the substantial resources required and the associated opportunity costs (Reference ChermackChermack, 2022). German-language practice guides explicitly highlight that SMEs often lack the capacity for extensive scenario projects and therefore need lightweight, well-guided procedures (Reference SiebeSiebe, 2018). P3 – The risks of oversimplification and loss of system structure: “Quick methods” that skip factor structure and consistency checks can lead to limited possibility spaces and internally contradictory scenario sets. The sources of these methods emphasise that morphological structuring and consistency analysis are critical for maintaining fidelity to interdependencies and avoiding incompatible combinations of projections (Reference Siebe, Michl, Frank, Echterhoff and SiebeSiebe, Michl, et al., 2018) (Reference Siebe, Michl and SiebeSiebe & Michl, 2018). P4 – Weak fit to domain-specific needs and limited participation: Scenario processes are most effective when they engage diverse perspectives and are closely linked to the context of the focal organization. The literature describes scenario work as a dialogic, learning-oriented practice that benefits from stakeholder participation rather than merely technical analysis (Reference ChermackChermack, 2022). Recent engineering design studies also report on the value of interactive formats (e.g., on-the-fly checks and consequence exploration) in keeping groups engaged and anchored in domain-specific decisions (Reference Ködding, Koldewey, Dumitrescu, Shajek and HartmannKödding et al., 2023). P5 – Poor reuse of accumulated knowledge about the future: Organizations often reinvent drivers, projections, and narratives across projects. Lessons learned are not encoded into reusable structures (Reference ChermackChermack, 2022). Scenario scholarship emphasises the learning function of scenarios (often under-exploited) and the need to institutionalize insights into organizational memory (Reference ChermackChermack, 2011).
As scenario work becomes more data- and digitally assisted, governance requirements such as input privacy/IP, output justification, and decision auditability increase. Therefore assumptions, rationales, and signposts must be documented for accountability and learning (Reference ChermackChermack, 2011).
4. State of the art
This section reviews the state of the art in scenario planning and digital foresight, providing a conceptual and empirical foundation. It highlights the methodological continuity of scenario work and the recent shifts enabled by AI and digital technologies. Based on this, we identify gaps specific to SMEs and distill the corresponding requirements and implications that shape the design of the solution.
4.1. Conceptual foundations of scenario work
Scenario methods emerged in response to the limitations of forecasting under deep uncertainty. Classic texts define scenarios as structured lenses for exploring plausible, alternative futures rather than as predictions. These methods emphasize shared frameworks, disciplined variation, and decision relevance (Reference SchwartzSchwartz, 2007) (van der Heijden, 2009). The literature converges on a process logic that links environmental drivers to projections, combines projections into consistent sets, and translates sets into decision premises (Reference Lindgren and BandholdLindgren & Bandhold, 2009). Morphological analysis and formal consistency checks are treated as quality anchors because they preserve system structure and identify incompatibilities early on (Reference Siebe, Michl, Frank, Echterhoff and SiebeSiebe, Michl, et al., 2018). Historical reviews map out the different schools of thought and caution against drifting into ad hoc storytelling without structure (Reference Bradfield, Wright, Burt, Cairns and van der HeijdenBradfield et al., 2005). Overall, scenario planning is presented as a structured discussion practice that produces tangible artifacts for strategy development.
4.2. Current practice and tool landscape
The following market scan analyzes existing tools against four criteria derived from the SME-specific gaps: end-to-end process continuity, structural fidelity (morphology and consistency), traceability and provenance, and support for cross-project reuse. In practice, the state of the art reflected in the literature is only partial, as available IT support often consists of point solutions without end-to-end continuity across the entire early-foresight process (Reference Siebe, Korsmeier, Schmidt and SiebeSiebe, Korsmeier, & Schmidt, 2018). Many tools on the market only cover single phases or sub-processes. There is no system that supports the entire process holistically (Reference Korsmeier, Schmidt and SiebeKorsmeier & Schmidt, 2018). Using a different tool for each activity increases process complexity, coordination overhead, and inefficiency, especially for SMEs (Reference Siebe, Korsmeier, Schmidt and SiebeSiebe, Korsmeier, & Schmidt, 2018). Survey evidence indicates that a systematic approach has not been established, and that current IT offerings often fail to deliver sufficient benefits to many companies. At the same time, the increased demand for forecasts means that many instruments can no longer be handled manually, underscoring the importance of suitable software (Reference Korsmeier, Schmidt and SiebeKorsmeier & Schmidt, 2018). Practice studies also indicate a shift toward short-term thinking under digitalization pressure, though satisfaction with scenario outcomes remains high when applied. At the same time, the data environment raises challenges to systematic practice. Growing data volumes, heterogeneous sources, varying speeds, quality concerns, and value discernment all increase the effort required to maintain current and well-grounded scenarios (the “5Vs” challenge) (Reference Ködding, Koldewey, Dumitrescu, Shajek and HartmannKödding et al., 2023). As a result, organizations underutilize formal methods and rely on simpler workshop alternatives that risk oversimplification. In summary, there is a clear need for an integrated set of tools that enables companies to conduct foresight effectively and efficiently.
A scan of leading market offerings highlights this fragmentation. Shaping Tomorrow (https://www.shapingtomorrow.com) is an AI-driven horizon-scanning and reporting service with modules for signal aggregation and visualization. However, based on public materials, the emphasis remains upstream. There is no evidence of an end-to-end, methodical coupling into morphology/consistency checking, evaluative bundling of scenario sets, or a linked model spanning factors, projections, scenarios, and indicators. The Futures Platform (https://www.futuresplatform.com) offers a rich, curated content base (e.g. trend cards, radars, and signals) and a two-axis scenario tool with collaboration features. However, traceability across projections, consistency matrices, and scenario portfolios, as well as a reusable provenance backbone, remains user orchestrated. ITONICS (https://www.itonics-innovation.de) is excellent as an innovation operating system with scouting, radars, workflows, and portfolios, but the scenario methodology seems secondary. Formal morphology and consistency, as well as scenario-based evaluation pipelines, are not articulated. The Foresight Strategy Cockpit from 4strat (https://www.4strat.de/foresight-strategy-cockpit) offers collaborative trend and signal management, analytics, and reporting. However, there are no publicly documented modules for morphology, consistency matrices, projection catalogs, or a provenance chain. Parmenides EIDOS (https://parmenides-eidos.com) is powerful for visual reasoning and structuring complex problems. However, it does not present a standardized, automated scenario workflow with consistency checks, project libraries, and evaluative follow-through. This implies a high level of facilitation effort. In short, these tools are strong in scanning, content, and collaboration, but they do not deliver a continuous, auditable scenario pipeline. This is precisely the gap that our solution addresses with a wizard + expert environment on a knowledge backbone.
4.3. Communication and learning
Research consistently frames scenarios as learning and decision architectures rather than one-off analytical products. Participation is important because diverse perspectives surface assumptions and broaden the hypothesis space. The quality of communication matters because stakeholders need to understand why a set of scenarios is coherent and how it relates to decisions (Reference ChermackChermack, 2022). Practice experiments with visual futures and narrative formats aim to raise engagement, but many outputs remain monoperspectival or disconnected from decisions in use (Reference Siebe, Michl, Frank, Echterhoff and SiebeSiebe, Michl, et al., 2018) (Reference Siebe, Michl and SiebeSiebe & Michl, 2018). A second gap concerns organizational memory. Drivers, projections, and narratives are often reinvented across projects because knowledge is not encoded in reusable structures with provenance (Reference ChermackChermack, 2011). This limits cumulative capability building.
4.4. Digital and AI support
Recent studies have explored how digital and AI techniques have changed the economics and quality of scenario practice. Evidence from a broad range of knowledge-intensive work reports efficiency and quality improvements with GenAI assistance, which encourages its adoption in related domains (Reference Dell’Acqua, McFowland, Mollick, Lifshitz-Assaf, Kellogg, Rajendran, Krayer, Candelon and LakhaniDell’Acqua et al., 2023).
Specifically for scenario-based foresight, a recent literature review consolidates 14 distinct use cases for digital technologies across the stages of scenario-based foresight. These tasks include identifying key influence factors, semi-automated consistency assessment, and algorithm-supported scenario generation. Overall, the focus is on data work. The tools primarily assist in collecting, analyzing, and extracting knowledge from large volumes of text and signals. Very few reach into decision support directly. The review also categorizes technologies based on explainability, ranging from procedural algorithms and symbolic AI to subsymbolic approaches. Deep learning remains largely absent, and humans are still central to interpreting and elaborating scenarios to preserve transparency and judgment (Reference Ködding, Koldewey, Dumitrescu, Shajek and HartmannKödding et al., 2023). In parallel, practice-oriented work indicates a hybrid division of labor. Digital technologies expand coverage and speed up the collection, analysis, and structuring of data. However, framing, plausibility judgment, and translation into strategy remain the responsibility of practitioners (Reference Ködding, Koldewey, Dumitrescu, Shajek and HartmannKödding et al., 2023). Specifically, it demonstrates how integrating digital support can reduce the effort required for data gathering and improve usability. However, it maintains the central role of humans in interpretation and decision-making processes. Methodologically, this integration enables dynamic analyses, such as real-time sensitivity checks within scenario work, while still requiring an explicit role allocation between humans and digital tools. To sustain trust and ensure that conversations are decision-oriented, the studies also emphasize the need for digital sovereignty—primarily the robustness and explainability of the results and their provenance (Reference Ködding, Dumitrescu and HartmannKödding & Dumitrescu, 2022) (Reference Ködding, Koldewey, Dumitrescu, Shajek and HartmannKödding et al., 2023). The guidelines focus on pragmatic strategies that clarify prompting patterns, establish guardrails for explainability and bias, and integrate outputs into existing governance systems instead of operating AI independently (Reference Spaniol and RowlandSpaniol & Rowland, 2023).
In summary, these strands of literature are directly associated with the five problem clusters identified in Section 3. Two boundaries are apparent. First, end-to-end automation is neither realistic nor desirable for high-stakes strategy. The value lies in augmenting analysis and increasing coverage while retaining human judgment. Second, explainability is non-negotiable. Teams require comprehensible artifacts and navigable rationales to trust and reuse AI-assisted outputs.
4.5. SME-specific gaps and design implications
According to Sections 3 and 4, the gap between methodological ambition and organizational reality widens for SMEs. Current digital offerings are still point solutions with little end-to-end continuity or governance for reuse. In practice, tools offer individual process steps but rarely holistic solutions or operational connections. This impairs daily consistency and reusability. These issues converge into five SME-specific challenges that collectively define the design space for the solution.
Gap 1 — Integration and traceability: Without a shared representation linking factors, projections, scenarios, and indicators, teams struggle to reliably move from outputs to inputs or justify choices after the fact. Implications: Introduce a knowledge backbone that explicitly encodes the provenance and “why” rationales of factors, projections, scenarios, and indicators. Artifacts must make causal logics and key uncertainties visible and explicit to users (Reference Lindgren and BandholdLindgren & Bandhold, 2009). To stabilize conversations and reduce drift, outputs should support shared mental models through traceable links and rationale cues (Reference SchwartzSchwartz, 2007). Gap 2 — Intelligibility and participation: Value creation hinges on understandable artifacts and meaningful participation. However, many outputs are unclear and require continuous facilitation. Implications: Provide understandable materials, such as one-page documents with explanations, to help participants grasp the logic even when a moderator is not present. Gap 3 — Effort and cost under SME conditions: Tight budgets, small teams and heterogeneous data make multi-tool workflows expensive and unreliable. Implications: Reduce non-value-adding effort via defaults and templates. Design pause and resume flows and apply assistive automation where it demonstrably reduces time and cost without compromising quality (Reference ChermackChermack, 2022). Human expert assessment remains central to key decision-making points. Gap 4 — Structural fidelity: “Quick methods” that omit morphology and consistency checks endanger internal coherence and limit the space of plausible futures. Implications: Retain morphology-like structuring and lightweight, AI-supported consistency checks to guard against incompatible projection bundles. Make the trade-offs between simplicity and fidelity explicit and traceable in the artifacts and their decision trails Reference Siebe, Michl, Frank, Echterhoff and Siebe(Siebe, Michl, et al., 2018) (Reference Siebe, Michl and SiebeSiebe & Michl, 2018). Gap 5 — Governance for reuse: Without provenance and historical project information, organizations repeatedly reinvent material and cannot credibly reuse or audit it across projects. This implies implementing provenance alongside project directories that include factors, projections, scenarios, and indicators, which would enable reuse and adaptation (Reference Lindgren and BandholdLindgren & Bandhold, 2009). Ensuring traceability from stakeholder inputs to the results of each step and the final scenario outputs is essential (Reference ChermackChermack, 2011).
The identified gaps refine the earlier problem clusters for SME conditions. Together, these deltas focus the design on traceability, intelligibility, effort, structural fidelity, and reuse. These implications are derived from the gaps in Sections 3 and 4 that were synthesized from the literature. Each design direction transforms a deficiency in traceability, intelligibility, effort structure, methodological rigor, or governance into an architectural requirement. These requirements are operationalized in the following solution concept.
5. Solution approach
Based on the identified gaps and requirements, this section translates the conceptual and empirical findings into a concrete system design. The solution approach outlines how we want to implement the objectives of intelligibility, efficiency, structural fidelity and reuse through an integrated architecture and modular user flow. The overall workflow of the proposed system is structured into five phases, as illustrated in Figure 1.
Five-phase scenario management process
5.1. Design goals and guiding principles
The solution is conceived as an end-to-end, modular and AI-enabled system designed for scenario management in small and medium-sized enterprises. Its design targets four key objectives derived from Section 3/4: intelligible outputs, significantly reduced effort per step, structural quality in terms of morphology and consistency, and adaptability to SME conditions.
The architecture and module design explicitly respond to the five problem clusters identified in Section 3. Perceived complexity (P1) is addressed through guided workflows and graspable artifacts such as one-pagers. Effort and cost barriers (P2) are mitigated via modular defaults and assistive automation. Structural fidelity risks (P3) are handled through explicit factor–projection modeling and consistency-aware algorithms. Participation challenges (P4) are supported by intelligible artifacts and human-in-the-loop review modes. Finally, reuse deficits (P5) are addressed through the knowledge backbone and explicit provenance modeling.
To achieve these goals, the solution adheres to the following guiding principles. Firstly, the system is designed to be human-in-the-loop. AI modules increase coverage and speed while practitioners retain responsibility for framing, plausibility, and decisions. Secondly, explainability and traceability are treated as primary requirements. Every AI-assisted output carries visible rationales, provenance, and uncertainty signals, enabling teams to analyze ‘what drives what’ and retrace results to their inputs. Thirdly, it offers a reuse-first logic. A shared knowledge backbone links factors, projections, scenarios and indicators across projects and versions, meaning that organizations do not ‘reset to zero’ in each cycle. Fourthly, the user experience employs progressive disclosure. A wizard guides novices step-by-step, while experts can adjust parameters without leaving the flow. Finally, secure data paths and IP protection are embedded from the outset, including authentication, role-based access and auditable processing. These guiding principles entail deliberate trade-offs rather than universally compatible goals. For instance, increasing structural fidelity and traceability inevitably raises complexity, which is counterbalanced through progressive disclosure and modularization. Similarly, while automation has the potential to enhance efficiency, it is deliberately constrained in areas where it might compromise interpretability, governance control, or human sensemaking.
The solution architecture incorporates the deltas identified in Sections 3 and 4. Graspable artifacts and visible rationales address perceived complexity. Defaults, templates and assistive automation reduce effort without compromising quality. Lightweight consistency checks and morphological analysis preserve structural accuracy. An AI review mode, expert review roles, and transparent results strengthen participation. Finally, the knowledge backbone and project libraries institutionalize memory and enable auditable reuse. Beyond its application as a specific method, the proposed system can be interpreted as a modular sensemaking and decision-support artifact for knowledge-intensive design under uncertainty. Scenario processes are structured forms of collective sensemaking that link weak signals, drivers, assumptions, and strategic options into coherent decision premises. In this respect, the architecture aligns with research on decision-support systems and organizational knowledge reuse.
5.2. Solution concept and user flow
Functionally, the AI solution combines two modes on one backbone: a Scenario Wizard for low-barrier, step-by-step use and an expert environment for more advanced settings. Both operate on the same objects and services, producing the same artifacts for communication. The knowledge backbone (KB) is the system’s central data structure. It defines the core entities such as Factor, Projection, Scenario, and the auxiliary entities like Evidence, Rationale and Decision Link. The KB stores their relationships, versions and provenance and maintains coherence across modules. When scenarios are updated, the KB records the changes made. Over time, organizations can build libraries of reusable factors or projection patterns with clear checks to ensure they are fit for use in new contexts. Based on this foundation, we implement the user flow with a set of AI-enabled components that align with the standard phases of scenario work. Each component is designed to minimize unnecessary effort while preserving methodological accuracy and human oversight.
Phase 1 – Scenario Field Definition: During the scoping phase, three assistants reduce ambiguity and produce a traceable brief. The Assignment Clarifier identifies and standardizes the focal question, decision horizon, and boundary conditions. These outputs become KB objects (problem statement, scope notes, and initial assumptions). The Layer Detector proposes a multi-level structuring of the field (e.g., macro, industry, firm, product). A Structure Checker validates internal coherence, consistency of scope with objectives, duplication across levels, and missing linkages. It flags issues and records rationale snippets for later auditing. The separation into three assistants reflects the intention to distinguish between framing, structural layering, and internal coherence checks. These represent cognitively distinct tasks in the early stages of scenario scoping, which should be usable independently of each other in a later software.
Phase 2 – Scenario Field Analysis: This phase combines evidence gathering with systematic curation on the KB. A Factor Scanner (Web Search/File Search) retrieves and ranks candidate drivers from permitted web and corpus sources. The findings arrive as traceable “evidence” items attached to provisional factors. The Factor Catalog Checker enforces naming, granularity, and non-redundancy rules. When organizations want to reuse prior assets, the Factor Integrator merges legacy lists and project libraries, while preserving provenance and version history. Quality is improved through lightweight quantification. An Influence-Matrix Appraiser assists teams in scoring mutual influences. In parallel, a Relevance Appraiser assesses the importance, uncertainty, and relevance. Together, these diagnostics narrow the field to a transparent and defensible set of drivers. To prepare for projection work, two selectors operate on the curated set. A Key Factor Selector proposes three subsets of key factors: reliable, optional and other. Additionally, a Key Factor Optimizer explores alternative selections under different constraints.
Phase 3 – Scenario Prognostics: For each key factor, the Portfolio Generator recommends a structured projection portfolio. This is done by selecting two core dimensions based on the factor definition and existing data. The generator then defines their ranges (min/max) and generates up to five alternative future projections with brief descriptions. A Projections Catalog Checker checks the catalog for differences, completeness, and compliance with restrictions. The selection of the two core dimensions is based on established scenario heuristics that balance tractability and variation. Limiting the number of projections to a manageable amount supports cognitive usability and prevents a combinatorial explosion in SMEs.
Phase 4 – Scenario Building: A Consistency Appraiser evaluates the logical coherence between all projections in a consistency matrix. It rates the pairwise compatibility of each projection on a defined scale. The algorithm then uses these scores to identify and rank the most consistent projection bundles according to overall plausibility and coverage. A pairwise consistency matrix was chosen because it provides more transparency and explainability than opaque optimization approaches. This aligns with the design goal of achieving traceable structural fidelity.
Phase 5 – Scenario Interpretation: Three authoring and analysis tools support communication and consequence analysis. A Scenario Writer converts each structured scenario into an executive one-pager while preserving links to assumptions, uncertainties, and cited evidence.
For impact reasoning, an Impact Analyst guides users through consequence chains, grounding the discussion in previously recorded drivers and indicators. The Dimensions Oracle evaluates scenarios based on user-defined dimensions (e.g., robustness, sustainability, skills) using transparent scoring formulas and records justifications. When organizations adopt formal scoring rubrics, the optional Assessment Designer assists in defining and versioning them, ensuring that assessments remain comparable across projects.
5.3. Technical architecture and operations
The system is designed as a cloud-native, service-oriented setup that reflects the modular logic of the solution concept. A web-based user interface provides access to the guided wizard and expert environments at the frontend. A unified API layer mediates all data exchange between client applications and backend services. This ensures consistent data and shields users from internal complexity. A content distribution layer delivers static resources and user-facing artifacts to support scalable, low-latency access. These architectural decisions are directly aligned with the previously defined design goals. For instance, modular services enable reuse (Gap 5) and controlled storage supports traceability (Gap 1). The core application logic is implemented through modular backend services that handle tasks such as managing scenario objects, validating data, and coordinating workflows. Long-running or computationally intensive processes, such as projection generation, consistency assessment, and document compilation, are executed via asynchronous job services. The system state, metadata, and the operational representation of knowledge-backbone entities are stored in a relational database with explicit version control. This enables auditability and cross-project reuse. AI-enabled components operate within a dedicated runtime environment that hosts the pipelines based on LLMs and optional retrieval. This environment manages model execution, prompt libraries, and evaluation utilities. It also integrates a semantic index that supports retrieval from internal artifacts and permitted external sources. The orchestration is model-agnostic, enabling organizations to integrate different LLM providers or on-premises models based on governance requirements. A separate service handles document rendering and media generation. This service produces scenario one-pagers and reports in common formats. Authentication and authorization are implemented through an identity service that supports role-based access and the separation of tenant data. Cross-cutting concerns, such as encryption, error monitoring, and compliance with data protection requirements, are embedded at the platform level rather than being delegated to individual modules. From an operational standpoint, the architecture emphasizes robustness, transparency, and manageable cost profiles. Usage and performance metrics can be continuously monitored to support cost control for SMEs and allow more intensive analyses for larger users. The system can be embedded into existing foresight or governance infrastructures through integration points, such as APIs and webhooks.
Together, these elements form a coherent technical foundation that translates the methodological requirements of traceability, structural fidelity, intelligibility, and reuse. The modular setup supports approachable, guided scenario work while retaining expert-level depth. The knowledge backbone ensures continuity and provenance across projects, and the AI components accelerate value-adding steps without replacing human judgment. The result is a scalable, explainable, and reusable platform designed for practical SME conditions.
6. Conclusion
This paper introduces an AI-based, modular approach to end-to-end scenario management tailored to SME conditions. Rather than automating scenario work entirely, the system reorganizes it. A cloud-based expert environment should enable deeper parameterization and a shared knowledge backbone links factors, projections, scenarios and their provenance. This combination addresses the practical deltas identified earlier—complexity, effort, structural fidelity, participation and reuse—without diluting the method’s core logic. Explainability and traceability enable teams to inspect assumptions, understand causal links, and reuse materials across cycles. The expected contribution to the field is twofold. First, the solution is designed to reduce effort in scanning, curation, projection drafting and consistency checks while keeping framing and judgment with practitioners. Second, it is intended to preserve structural quality via morphology- and consistency-aware workflows, making results communicable through one-pagers and visible rationales. For SMEs, the approach is aimed to support processes that fit short time windows, can be paused and resumed, and still produces decision-ready outputs. Our systems design pattern is a human-in-the-loop scenario pipeline on a knowledge backbone that is adaptable across domains.
The work has boundaries. The solution does not aim for full automation or replace expert facilitation. Its value depends on informed users who understand the rationales and make context-sensitive choices. Furthermore, AI assistance introduces governance demands, such as model choice, bias control, and data protection, that must be managed as part of normal operations. Technical performance and cost can vary based on data, tasks and model configurations, therefore careful monitoring and prompt governance are essential. Additionally, the proposed architectural framework imposes constraints at the design stage. Although the distribution of complexity is a benefit of modularization, the overall system may still appear challenging at the beginning in terms of conceptual familiarization, usage, and process discipline. SMEs with limited digital maturity or no previous scenario experience may encounter challenges during the introduction phase. The development of a knowledge base necessitates organizational commitment and must be compatible with data protection laws in the respective SME contexts. These aspects represent important constraints that must be empirically investigated in future evaluations.
The next steps will focus on further implementation of the system, conducting an empirical evaluation, and generalizing the results. In DS-II, we will evaluate the system in a mechanical engineering use case. We will track efficiency (lead time and time per step), as well as the intelligibility, usability, structural quality, coverage, and robustness of strategy options. Comparative baselines against conventional toolchains, ablation studies of AI modules and longitudinal reuse effects will clarify where the approach delivers the most value. Further work will expand the knowledge backbone for richer provenance, broaden domain libraries, and explore integration pathways with existing foresight tools and enterprise systems.
Overall, the presented solution aims to address current shortcomings of scenario practice by integrating fragmented steps, offering a holistic solution for scenario management and transforming one-off projects into reusable organizational memory while maintaining a focus on human judgment. This approach outlines a potentially viable path toward professional, AI-supported scenario management for SMEs, which will need to be examined empirically with respect to cost, effort, and adoption.
Declaration of generative AI and AI-assisted technologies in the writing process
During the preparation of this work the authors used tools to enhance language and readability. After using this tool, the authors reviewed and edited the content as needed and takes full responsibility for the content of the publication.
Acknowledgement
These research results were developed as part of the “AISceMa” project, which was selected in the innovation competition “NEXT.IN.NRW” and is funded as part of the European regional funding program (EFRE/JTF Program NRW 2021-2027) by the Ministry of Economic Affairs, Industry, Climate Protection, and Energy of the State of North Rhine-Westphalia (MWIKE) in cooperation with the Ministry of Culture and Science of the State of North Rhine-Westphalia (MKW).
Open access
