Integrating Decision Intelligence with Other Decision-Making Frameworks

Every organization faces the same fundamental problem: decisions involve multiple competing values, uncertain information, and consequences that ripple far into the future. Consequential choices in public policy in particular affect people's lives in ways that poor frameworks make worse. For decades, two broad traditions have tried to address this problem. The first, rooted in operations research and management science, gave us formal frameworks for systematically weighing competing criteria. The second, emerging from artificial intelligence and data science, extracts decision-relevant insight from large datasets. For most of their histories, these traditions have operated largely independently, but that is changing. In particular, the integration of Decision Intelligence (DI) with other established decision-making methodologies holds promise for leveraging the strengths of all approaches while addressing their shortcomings. Some of this integration is already underway in published research and some remains a frontier of possibility.

Multiple Decision Frameworks

Decision Intelligence is a discipline formalized over the past two decades, associated prominently with the work of computer scientist Lorien Pratt, who co-developed the concept around 2010 and authored The Decision Intelligence Handbook with co-author N. E. Malcolm.¹ At its core, DI focuses on making explicit the causal chains that connect actions to desired outcomes within a specific objective question, and then using data, AI, and structured modeling to navigate that chain more reliably. Pratt's framework draws on causal decision diagrams, machine learning, and simulation, with a strong emphasis on keeping human judgment in the loop rather than replacing it.

Multi-Criteria Decision Analysis (MCDA) is a family of techniques developed to help decision-makers evaluate options across multiple, often conflicting, objectives. When a city must choose between two bridge designs and must weigh cost, environmental impact, aesthetic value, and construction time simultaneously, MCDA provides structured methods for doing so. Tools within the MCDA family include the Analytic Hierarchy Process (AHP), TOPSIS, VIKOR, and multi-attribute utility theory (MAUT), among others. As one review describes it, MCDA provides "a rigorous framework" for navigating these multi-dimensional trade-offs in a transparent, reproducible way.²

Structured Decision Making (SDM), as described by Hammond, Keeney, and Raiffa in their work Smart Choices³, is built around the PrOACT process: Problems, Objectives, Alternatives, Consequences, and Trade-offs. SDM is a philosophy as much as a toolkit, emphasizing deliberate, step-by-step reasoning over intuitive or reactive judgment. It has been widely applied in natural resource management, public policy, and, more recently, clinical medicine, where the term "shared decision making" refers to the collaborative process between clinicians and patients.

Cost-Benefit Analysis (CBA) and its close cousin, Regulatory Impact Assessment (RIA), are the workhorses of government decision-making. CBA converts the consequences of a proposed policy into monetary terms so that benefits can be compared against costs. RIA extends CBA to assess regulatory effects before enactment. Both are standard practice across OECD member countries. Their strength is discipline and comparability, but their weakness is that important consequences including fairness, dignity, and community cohesion, typically resist conversion into dollar figures.

In contrast, Deliberative Democracy approaches rest on a different premise: the legitimacy of a policy decision depends not just on its analytic quality, but on the quality of the process by which it was reached. Deliberative methods bring together representative groups of citizens to reason through complex public issues, often with expert briefings and professional facilitation. The goal is consensus or mutual understanding, not mathematical optimization.

Incrementalism, developed by political scientist Charles Lindblom in the 1950s and 1960s, observes that comprehensive rational analysis is rarely possible in practice. Decision-makers face information limits, competing interests, and time pressures. Policy tends to change in small, successive steps from the status quo rather than through analytically optimized leaps.⁴ Incrementalism is less a normative ideal than a descriptive reality, but one that any decision support framework must consider to be genuinely useful.

What Has Already Been Built

Neural networks meet multi-criteria analysis. The Neural Network-based Multiple Criteria Decision Aiding (NN-MCDA) method, developed by Guo and colleagues⁵, combines a standard MCDA value model with a deep neural network. Traditional MCDA assumes that decision-makers' preferences follow predictable patterns and that criteria can be evaluated independently, assumptions that often fail in practice. NN-MCDA handles both explicit relationships between attributes and complex nonlinear interactions that conventional MCDA cannot capture, producing a model that is more accurate and more interpretable than either approach alone.

Large language models as virtual AHP experts. A framework developed by Svoboda and Lande² uses GPT-4 as a panel of "virtual experts" within the Analytic Hierarchy Process, one of the most widely used MCDA tools. Normally, AHP requires human experts to make pairwise comparisons between criteria, a time-consuming and bias-prone process. Automating it with a large language model makes MCDA substantially faster and more scalable while preserving the structure of expert judgment.

AI augments regulatory impact assessment. Governments are beginning to use AI to strengthen what has long been a laborious manual process. A 2025 OECD report surveying 200 real-world government AI deployments found that AI is now being used to streamline policy analysis, RIA, and legal drafting, identifying gaps in regulatory frameworks, predicting compliance cost impacts of proposed legal text changes, and enabling counterfactual policy evaluation.⁶ In one case, machine learning tools were combined with economic theory to analyze the emissions and cost impacts of the United Kingdom's Carbon Price Support policy, a case where no experimental control group existed and traditional evaluation was effectively impossible.

AI and deliberative democracy. The platform Pol.is, which uses machine learning to cluster citizen arguments and surface areas of emerging agreement, has supported deliberations in Taiwan that led to actual legislative changes on ride-sharing regulation and privacy law.⁷ More recent research has explored the "Habermas Machine", which uses large language models to serve as AI mediators to help participants find common ground in structured deliberative settings.⁸

AI-supported shared decision making in healthcare. The AI-SDM framework, described in a 2025 paper in JMIR AI,⁹ addresses a persistent gap: AI systems designed to support clinical decisions have focused on technical transparency rather than clinical reasoning. AI-SDM synthesizes predictive modeling, evidence-based recommendations, and generative AI to produce adaptive, context-sensitive explanations that support the conversation between clinician and patient rather than bypassing it.

The gaps and what could fill them

Causal models as a source of CBA and MCDA weights. One of the weaknesses of both CBA and MCDA is determining how to value different consequences relative to one another. DI's Causal Decision Diagrams, which map how actions produce outcomes through chains of intermediate variables, could be used to derive these valuations empirically, thereby grounding them in observed system behavior rather than analyst assumptions or political negotiation.

Feeding outcome data back into SDM. The PrOACT framework is structured but not inherently adaptive. Once a decision is made, there is no built-in mechanism for learning whether the consequence model was correct. DI's emphasis on closing the loop between decisions and outcomes could make SDM genuinely iterative, so that each decision cycle refines the models used in the next.

DI tools to manage the pace of incrementalism. Incrementalism is realistic about what is politically achievable, but policy changes made in small, disconnected steps rarely add up to a coherent strategy, and their cumulative effects are often invisible until something goes wrong. DI's scenario modeling and outcome-tracking capabilities could give policy teams a running view of how a succession of incremental decisions is moving (or failing to move) toward stated objectives, providing an evidence base for course correction before problems compound.

AI-mediated deliberation at scale. Traditional deliberative processes are necessarily small: a citizens' assembly of 150 people is considered ambitious. AI tools offer the possibility of scaling genuine deliberation to much larger populations, though researchers caution that "scaling democratic deliberation is not susceptible to a technological fix alone; it requires careful technological integration alongside broader processes of social and political change".¹⁰

Simulation to resolve stakeholder conflict. A well-documented challenge in multi-stakeholder MCDA is that different groups, using the same method with different criteria weights, arrive at completely different rankings, leading to protracted disputes.¹¹ DI's simulation capabilities could allow stakeholders to explore the consequences of different weighting schemes interactively and in real time, making trade-offs visible before positions harden into conflict.

The integration already underway among these various approaches suggests that combinations of approaches can be more capable than any one alone. The work ahead involves not just technical development, but careful institutional design: building processes and governance structures that keep human judgment genuinely central, that make trade-offs visible rather than hidden in model parameters, and that treat the people affected by decisions as stakeholders rather than as data points. Decision intelligence can have particular power in these efforts as a way of diagramming causal relationships, integrating technology while keeping human expertise at the center, and explicitly modeling feedback loops in time and causality that make addressing wicked problems in public policy more tractable.

References and Further Reading:

1. Pratt, L. Y. & Malcolm, N. E. The Decision Intelligence Handbook: Practical Steps for Evidence-Based Decisions in a Complex World. (O’Reilly Media, Sebastopol, CA, 2023).
2. Svoboda, I. & Lande, D. Enhancing Multi-Criteria Decision Analysis with AI: Integrating Analytic Hierarchy Process and GPT-4 for Automated Decision Support. Preprint at https://doi.org/10.48550/arXiv.2402.07404 (2024).
3. Hammond, J. S., Keeney, R. L. & Raiffa, H. Smart Choices. (Harvard Business School Press, Boston, MA, 1999).
4. Incrementalism | Social Sciences and Humanities | Research Starters | EBSCO Research. EBSCO https://www.ebsco.com/research-starters/social-sciences-and-humanities/incrementalism.
5. Guo, M., Zhang, Q., Liao, X., Chen, F. Y. & Zeng, D. D. A hybrid machine learning framework for analyzing human decision-making through learning preferences. Omega 101, 102263 (2021).
6. OECD. Governing with Artificial Intelligence: The State of Play and Way Forward in Core Government Functions. OECD https://www.oecd.org/en/publications/governing-with-artificial-intelligence_795de142-en.html (2025) doi:10.1787/795de142-en.
7. Megill, C. pol.is in Taiwan. Medium https://blog.pol.is/pol-is-in-taiwan-da7570d372b5 (2016).
8. Tessler, M. H. et al. AI can help humans find common ground in democratic deliberation. Science 386, eadq2852 (2024). https://doi.org/10.1126/science.adq2852
9. As’ad, M., Faran, N. & Joharji, H. AI-Supported Shared Decision-Making (AI-SDM): Conceptual Framework. JMIR AI 4, e75866 (2025).
10. Rymon, Y. Of the people, by the algorithm: how AI transforms the role of democratic representatives? AI Soc. https://doi.org/10.1007/s00146-026-02852-x (2026) doi:10.1007/s00146-026-02852-x.
11. Tsakalerou, M., Efthymiadis, D. & Abilez, A. An intelligent methodology for the use of multi-criteria decision analysis in impact assessment: the case of real-world offshore construction. Sci. Rep. 12, 15137 (2022).

Feature image by Gerd Altmann from Pixabay