The Rise of Decision Intelligence
Until recent years, decision intelligence was a little-known term. Google Trends data for the United States starting in 2004 shows only scattered interest until the late 2010s. Interest began to rise consistently after 2020, nearly doubled from 2024 to 2025, and has surged to its highest level so far in 2026.
Decision intelligence points to a gap that, I believe, should have been central all along in traditional analytics. While descriptive and business intelligence show what happened and predictive analytics projects what will happen, neither answers: What should we do about it? Insights only create value when they inform effective action. DI emerged to bridge the gap between accumulating data and taking informed action.
The field is entering the mainstream. In 2019, Lorien Pratt identified DI as a discipline combining analytics, behavioral science, and AI, while Cassie Kozyrkov, then Google’s Chief Decision Scientist, described it as a blend of data science, psychology, and leadership. By 2021, Gartner named DI a top strategic trend, exposing it to global corporate leaders.
Rising interest in DI coincides with generative AI adoption. As AI agents begin to recommend and execute actions, the quality of underlying decisions is paramount. DI provides the framework to represent causal relationships, account for uncertainty, and incorporate human knowledge, ensuring that rapid, AI-driven decisions remain transparent and trustworthy.
Various definitions of DI exist, differing in their emphasis on automation, judgment, and modeling. This article describes the most salient perspectives and their commonalities
1. Decision analysis as a foundation for decision intelligence
Decision analysis (DA) provides the methodological foundation for DI. The INFORMS Decision Analysis Society [1] defines it as a prescriptive approach grounded in probability and utility theory. DA practitioners use a toolkit including decision trees and influence diagrams [2] to structure problems, identify uncertainties, and calculate decisions with the greatest expected value.
They employ expert elicitation for probability distributions and multi-criteria methods for stakeholder objectives. By applying psychological insights, DA reduces cognitive biases like overconfidence, ensuring a more rigorous decision-making process.
Pioneered in the 1960s by professors Howard Raiffa at Harvard, Ronald A Howard at Stanford, DA is typically applied to high-stakes strategic problems, especially in energy, pharmaceuticals, and public policy, where conflicting objectives and deep uncertainty are common.
In recent years, the decision analysis community has developed the notion of Decision Quality. This provides a set of guidelines and tools to ensure that all six elements of the process are properly addressed, including selecting an appropriate frame, creating doable decision options, obtaining relevant and reliable information, defining clear values and trade-offs, using sound reasoning, and finally committing to the recommended actions.
2. Decision science and causal modeling: Lorien Pratt
Lorien Pratt treats DI as an engineering discipline. Her work, notably in Link (2019) [3], shifts focus from analyzing data to designing decisions as rigorous, repeatable processes. Her approach emphasizes the architecture of the decision over predictive models alone.
Pratt and Mark Zangari co-developed Causal Decision Diagrams (CDDs) around 2010. Similar to influence diagrams, CDDs facilitate problem structuring and align incentives by making assumptions explicit, treating decision-making as an improvable organizational asset.
CDDs identify levers (decisions), externals (uncertainties), intermediates, and outcomes (objectives), using specific icons to distinguish variable types.
CDD arrows depict causal flow, unlike in influence diagrams where they denote probabilistic dependence, which may or may not be causal. This helps users visualize how specific actions directly impact final outcomes.
3. Data science and decision architecture: Cassie Kozyrkov
Cassie Kozyrkov, formerly Chief Decision Scientist at Google, defines DI as the discipline of turning information into better actions at any scale [4]. Drawing from data science and psychology, she emphasizes defining the decision process before analyzing data, ensuring analytics serve a specific purpose.
Kozyrkov stresses clarifying objectives, options, and relevant evidence. She distinguishes between calculation and judgment; while AI improves information, humans must establish the values and trade-offs that guide choices.
4. Enterprise Decision Intelligence platforms: Gartner
Gartner [5] defines DI as a discipline to improve decision-making by engineering how they are made and evaluated. They identify Decision Intelligence Platforms (DIPs) as software that augments or automates decisions through data, analytics, and AI.
In 2026, Gartner formalized DIPs as an official software category. This recognition establishes DI as a distinct enterprise requirement, influencing how organizations evaluate vendors and allocate technology budgets.
Gartner’s [5] platform view of DI covers the organizational decision lifecycle, including:
- collaboratively designing and explicitly modeling decisions
- composing decision services
- executing decision flows at scale
- governing decision quality
- learning from actions and outcomes
The platform view treats decisions as managed organizational assets—using rules, machine learning, and simulation to design, deploy, and audit decision workflows at scale.
5. Operational decision automation: IBM and other platform vendors
IBM Decision Intelligence [6] is an AI-powered platform that transforms corporate policies into governed workflows. Vendors like FICO, SAS, and Pega offer similar platforms to automate operational decisions with high consistency and auditability.
Typical applications for these platforms include:
- approving or routing a transaction
- determining eligibility
- detecting fraud
- applying a pricing or underwriting policy
- recommending a product or action
- allocating inventory
- responding to an operational event
Automation is ideal for repeatable, high-frequency decisions with stable objectives. However, people remain responsible for setting policies, monitoring performance, and handling exceptions.
6. AI and Transparent Decision Intelligence
Generative AI often provides persuasive but opaque recommendations for complex decisions. To ensure trust, organizations must move beyond simple chatbot prompts toward explicit decision modeling. Like professional decision analysts who make assumptions, probabilities, and objectives transparent for stakeholder verification, AI should function as a collaborator that builds inspectable models.
This approach—transparent decision intelligence—shifts AI from a ‘black box’ advisor to a shared representation of the decision problem. Tools like Assista in Analytica demonstrate this by enabling users to collaboratively develop models where alternatives, uncertainties, and objectives are clearly defined. In this framework, AI contributes speed, modeling knowledge, and explanation, while people retain responsibility for framing the problem, clarifying values, and making final decisions. Ultimately, the most valuable role of AI is not delivering a mysterious ‘best answer,’ but helping people create, challenge, and trust the models that guide their choices.
7. The Decision Continuum: Carl Spetzler
A challenge in making sense of “decision intelligence” is the wide range of types of decisions, in terms of scale and frequency. While decision analysts typically address high-stakes strategic choices, platforms like IBM Decision Intelligence focus on routine operational tasks. Carl Spetzler (2026) clarifies this via a Decision-Making Continuum [7]. The left end comprises frequent, data-rich, repeatable decisions suitable for automation. The right end involves unique, high-impact strategic decisions dependent on human judgment, creativity, and values.
| Position on the continuum | Frequent operational decisions | The “muddled middle” | Nonroutine strategic decisions |
|---|---|---|---|
| Examples | Credit screening, transaction fraud checks, inventory replenishment | Pricing, scheduling, underwriting, capacity planning, and supply-chain responses | Market entry, technology strategy, major investments, public policy, and infrastructure planning |
| Typical frequency | Thousands or millions of times per day or year | Weekly, monthly, quarterly, or yearly | Infrequent or unique |
| Repeatability | Highly standardized | Partly repeatable, with exceptions | Each decision has distinctive features |
| Available data | Abundant data and rapid feedback | Useful data, but often incomplete or unstable | Limited precedent and long feedback cycles |
| Uncertainty | Often bounded and measurable | Changing conditions and significant exceptions | Deep uncertainty, structural change, and novel risks |
| Objectives | Usually clear and stable | Multiple goals requiring tradeoffs | Strategic, conflicting, and sometimes contested |
| Role for technology | Prediction, rules, optimization, and automation | Orchestration, decision support, exception handling, and selective automation | Structuring, modeling, scenario analysis, and human–AI collaboration |
| Role for people | Set policies, constraints, and governance | Review exceptions, supply context, and exercise judgment | Frame the problem, develop alternatives, assess uncertainty, clarify values, and make the decision |
Algorithms excel where data is abundant. Strategic decisions, however, require human judgment to assess novel alternatives. Spetzler identifies a “muddled middle” between these extremes where the optimal approach is less certain. The goal is to determine which aspects to automate, augment, or leave to human judgment. This spectrum requires matching the technology to the specific decision context. Hybrid approaches—combining automation, simulation, and domain expertise—are best for the “muddled middle,” ensuring rigorous framing and commitment to action.
Six perspectives on decision intelligence
This table summarizes the six DI perspectives. The last column identifies their primary focus on Spetzler’s continuum.
| Perspective | Person or organization most associated | Central question | Main emphasis | Typical methods and technologies | Primary focus on the continuum |
|---|---|---|---|---|---|
| 1.Decision analysis | Howard Raiffa and Ronald A. Howard | How can we make logically consistent choices under uncertainty? | Prescriptive methods, probability, and utility theory | Decision trees, influence diagrams, and expected value | Strategic end |
| 2.Decision science and causal modeling | Lorien Pratt | How will actions lead to outcomes? | Causal structure | Causal diagrams, influence diagrams, simulation, systems models, and optimization | Middle through strategic |
| 3.Data science and decision architecture | Cassie Kozyrkov | How can information improve action? | Designing the decision before analyzing data | Statistics, experiments, forecasting, machine learning, metrics, and behavioral science | Entire continuum |
| 4.Decision intelligence platforms | Gartner | How can an organization design, execute, govern, and improve decisions? | Integrated decision lifecycle | Models, rules, analytics, knowledge graphs, workflows, agents, monitoring, and governance | Automated and middle |
| 5.Operational decision automation | IBM and business-rules platform vendors | How can repeatable decisions be made quickly and consistently? | Execution at scale | Rules, predictive models, optimization, decision services, and audit trails | Automated end |
| 6.Transparent decision intelligence | Lonnie Chrisman and Max Henrion | How can people and AI reason through an explicit, inspectable model? | Transparency, accountability, and shared understanding | Influence diagrams, quantitative models, uncertainty analysis, sensitivity analysis, and natural-language AI | Middle through strategic |
Unifying the Perspectives on Decision Intelligence
Despite diverse origins, these perspectives share a unified core. Recent work by Pratt and Kozyrkov has popularized and reframed mature decision-analysis principles for the age of data science and AI.
Core commonalities include:
- Ensuring intelligence only creates value when it informs action.
- Distinguishing controllable decisions from external uncertainties.
- Clarifying objectives and prioritizing human purposes over technology.
- Making reasoning visible and learning through feedback.
Key Differences and the Decision Continuum
The primary differences among these perspectives lie in their positioning along Carl Spetzler’s decision continuum—the spectrum from frequent, automated operational tasks to unique, high-stakes strategic judgments.
- Operational Focus: Perspectives such as Enterprise DI Platforms and Operational Automation emphasize speed, scale, and consistency, often leveraging rules and predictive models to automate repetitive decisions.
- Strategic Focus: Decision Analysis and Decision Quality frameworks prioritize framing, utility, and human judgment, focusing on complex decisions where historical data is scarce and risks are high.
- Bridging the Gap: Other approaches, like Data Science/Architecture and Transparent DI, operate across the continuum, focusing on the decision-making lifecycle and the integration of human judgment with AI assistance.
Implications for Practitioners
For practitioners of decision intelligence, these frameworks provide a toolkit rather than a single “correct” approach. Key implications are:
- Match the Tool to the Task: Don’t default to a single methodology. Use the decision continuum to determine the decision type. Is it a high-volume, predictable operation suitable for automation, or a high-stakes, novel strategic problem requiring rigorous human modeling and judgment?
- Cultivate Decision Quality: Regardless of the technology, the fundamental requirements for a “quality” decision remain: an appropriate frame, creative alternatives, reliable information, and clear values. AI can augment these components, but it cannot replace the human responsibility for objectives, tradeoffs, and final accountability.
- Prioritize Transparency and Trust: In an era of black-box AI, practitioners must insist on “transparent DI.” If AI recommends an action, ensure there is an explicit model, clear assumptions, and a rationale that stakeholders can inspect, challenge, and verify.
- Design for Governance: Decisions are organizational assets. Whether automated or human-led, they should be designed as repeatable processes that are auditable, governable, and continuously improved through feedback.
Further reading
- INFORMS Decision Analysis Society
- Howard, R.A. and J.E. Matheson (1981) Influence diagrams, In Readings on the principles and applications of decision analysis, Vol II, R.A. Howard and J.E. Matheson (Eds.), Strategic Decision Group.
- Lorien Pratt, Link: How Decision Intelligence Connects Data, Actions, and Outcomes for a Better World, 2019:Â
- Cassie Kozyrkov, Introduction to Decision Intelligence: https://medium.com/data-science/introduction-to-decision-intelligence-5d147ddab767
- Gartner, Magic Quadrant for Decision Intelligence Platforms: https://www.gartner.com/en/documents/7363830
- IBM Decision Intelligence: https://www.ibm.com/products/decision-optimization
- Carl Spetzler, 2026, Decision-Making Continuum on Decidewise,com:Â
Author
Max Henrion, PhD,
 is the CEO and Founder of Lumina and the creator of Analytica. With a background as a researcher, software designer, decision consultant, and entrepreneur he has worked across energy, healthcare, aerospace, and more. A former professor at Carnegie Mellon, he also led decision technology teams at Ask Jeeves. Max received the 2014 Decision Analysis Practice Award and the 2018 Frank Ramsey Medal from the Decision Analysis Society.Â
See also
Embrace risk & uncertainty
What is Analytica software?
Beyond the spreadsheet
