Field Notes 01 · Methodology
Decision Intelligence became a software category in January. Almost every vendor in it is built for decisions that repeat thousands of times a day.
Jayven Rappa · 9 min read
In short
Decision Intelligence is a class of software that makes the decision itself the object of the architecture — modeled explicitly, executed with governance, and measured against its outcome. Gartner formalized it as a market in January 2026 with a first Magic Quadrant covering 17 vendors. Almost all of those vendors serve high-frequency operational decisions, because repetition is what justifies modeling a decision and what proves the model was right. That leaves infrequent, high-consequence strategic decisions without a system.
Most companies have never been better instrumented. Warehouses are consolidated, BI is deployed, forecasting models run, and every function has a copilot inside its workflow. And then the quarter misses, and the leadership team sits down to work out why, and the answer comes out of a discussion rather than a system.
of organizations report AI having a measurable effect at the enterprise EBIT line. Nearly two-thirds have not scaled it enterprise-wide at all.
McKinsey global AI survey, 2025
The tooling worked. The enterprise result did not follow.
The reason is structural, and it is not a shortage of intelligence. It is that almost every enterprise system stops one step before the thing that creates value. Value does not appear when someone receives an insight. It appears when the insight changes a decision, an action follows, and the outcome improves.
Look at what the enterprise stack actually produces, and where each part of it hands off to a human.
Business intelligence answers what happened. Its object is the metric. It stops at a chart. Predictive analytics answers what is likely to happen. Its object is the prediction. It stops at a forecast. Process intelligence answers how work is flowing and where it breaks; its object is the process. A digital twin answers what is happening in a modeled system and what might happen if it changed; the system representation remains the object.
An AI copilot answers help me with this task. Its object is the user’s request. It is genuinely useful, and it does not make the underlying enterprise decision into anything that can be inspected, owned or audited afterward.
Every one of those hands the actual decision back to a person, unrecorded. What was decided, on what basis, by whom, with what alternatives considered, and whether it worked — none of that becomes an asset. It stays in a meeting, a deck, and someone’s memory.
Scroll to see the full diagram →
Gartner has been describing Decision Intelligence as a practical discipline since 2024: understand and engineer how decisions are made, and improve them through outcome feedback. In 2026 it stopped being only a discipline and became a software market.
A timeline of three Gartner publications in 2026: the first Magic Quadrant for Decision Intelligence Platforms on January 26, companion Critical Capabilities research on January 27, and a market overview on August 7. Jan 26, 2026 First Magic Quadrant for Decision Intelligence Platforms 17 vendors Jan 27, 2026 Companion Critical Capabilities research 15 capabilities · 4 use cases Aug 7, 2026 Market overview for the categoryGartner published its first Magic Quadrant for Decision Intelligence Platforms on January 26, 2026, evaluating 17 vendors. It published companion Critical Capabilities research the following day, decomposing the market across 15 evaluated capabilities and four principal use cases: Decision Analysis, Decision Engineering, Decision Science, and Decision Governance and Stewardship.
The publicly stated market definition is unambiguous. A Decision Intelligence Platform combines decision modeling, analytics and AI to support, augment and automate decisions, and to drive business outcomes. Gartner’s public mandatory feature set for the category resolves that into six things a product has to do.
Model the decision explicitly
Not the data around it, not the process around it. What triggers the decision, what it consumes, what alternatives and constraints apply, what it emits.
Execute human decisions
Route a decision to its rightful owner with the evidence attached, capture the approval or the override, and persist what was decided.
Execute automated decisions
Run a bounded class of decisions without a human in every instance, under explicit escalation rules and human oversight.
Compose decision services
Break decisions into modular, reusable components that other systems can invoke without the logic being rebuilt each time.
Monitor outcomes
Track the decision, the action and the result, so the next decision is informed by whether the last one worked.
Govern the decision
Ownership, authority, policy, version, transparency, repeatability and audit history — attached to the decision, not only to the model or the data.
That last pair is where the category earns its name. Everything else in the enterprise stack governs assets. This governs choices.
The temptation is to read Decision Intelligence as analytics with a stronger engine. It isn’t. The change is in what the architecture is built around.
The unit of architecture moves. For thirty years that unit has been the data, then the model, then the dashboard, then the workflow, and most recently the agent. Decision Intelligence makes it the decision. A decision becomes a named, versioned, inspectable, testable, improvable object — the same treatment code got when version control arrived.
A vertical progression showing the unit enterprise architecture has been built around: data, then the model, then the dashboard, then the agent, and now the decision. Data Model Dashboard Agent DecisionGovernance moves with it. Conventional AI governance asks whether a model is accurate, secure, unbiased and compliant. Decision governance asks the questions a board actually cares about: what was decided, which evidence and logic informed it, who had authority, what action followed, could a human intervene, and what happened. As more decisions get delegated to agents, that is the only place governance can usefully sit.
Prediction becomes intervention. A forecast tells you something is likely. A decision requires knowing what you should change and what changing it would do. Those are different questions needing different mathematics — the prescriptive analytics literature has been formal about this distinction for years. Predicting a metric accurately does not establish that moving a correlated lever will move it.
The organization gets a memory. Companies retain enormous quantities of data and almost none of the reasoning behind past decisions. Which alternatives were on the table, what assumptions mattered, what the decision was expected to produce, whether it did. That loss is so normal it is rarely named as a loss.
Gartner’s own forecasts give a sense of how consequential it thinks this is. It has predicted that by 2027, half of business decisions will be augmented or automated by AI agents using Decision Intelligence, and that by 2029 explicitly modeled decisions will be five times more trusted and 80 percent faster than ungoverned ones. Those are forecasts rather than observed results. What they indicate is the direction the analyst consensus expects — speed and trust together, from making decisions explicit rather than from making models bigger.
Seventeen vendors is a category, but it is not seventeen versions of one product. The market converged from several previously separate software traditions: business rules and decision management, credit and risk decisioning, customer next-best-action, supply chain and enterprise planning, process orchestration, and graph and contextual intelligence.
Read their public positioning and the center of gravity is unmistakable. Approve the credit application. Detect the fraudulent transaction. Choose the next-best action for this customer. Adjust the inventory position. Route the case. Enforce the policy. Allocate the supply.
There is a good structural reason for that concentration. Those decisions repeat — thousands or millions of times — which is exactly what justifies the investment in modeling and automating them, and exactly what generates enough instances to measure whether the model is any good. High frequency makes the economics work.
Now consider the other kind of decision.
Where should the next tranche of transformation capacity go. Which execution constraint is actually suppressing financial performance. Which of six initiatives has to happen before the other five can produce anything. Which capability must change before a platform investment can return. Which apparent revenue problem is really a delivery problem wearing a revenue costume.
Scroll to see the full diagram →
These decisions have the opposite profile. They are infrequent. They are enormous. They cross every function. They are made under real uncertainty. And the feedback loop runs eighteen months, which means you cannot learn your way to competence by repetition — by the time you know whether the sequence was right, the conditions have changed and the people have moved.
That combination is why the strategic layer has no system. You cannot A/B test a transformation sequence. The only way to reason rigorously about a decision you will make once is to have a model of the business that says how a change in one place propagates to everywhere else — and then to be honest about how much confidence that model actually has.
Every function in a large company has rigor proportional to how often it acts. Finance closes the books monthly and has controls. Supply chain plans weekly and has systems. The decisions that determine whether the strategy happens at all get made a handful of times a year, and they get made in a room.
Rejoyce is built for that corner. We call it Strategic Execution Intelligence, and it is worth being precise about what that name is and isn’t: it is our term for our domain specialization inside Decision Intelligence. It is not a Gartner category and we don’t present it as one.
What it means concretely: a Digital Mirror that models a specific enterprise across its organizational layers and journey phases; a sector substrate of benchmarks and distributions that establishes what should be true for a company of this type, stage and size; and Execution Physics — our model of how execution capability, constraint and intervention propagate into operating and financial outcomes. Joyce sits on top as the interface an executive can interrogate, with the arithmetic walled off from the language layer so that identical inputs return identical numbers and missing information is reported as missing rather than filled with something plausible.
We are not trying to be a horizontal platform. The companies with two decades of installed base in high-volume operational decisioning are very good at high-volume operational decisioning, and building a competing rules engine would be a poor use of anyone’s time. We built for one class of decision, deliberately, and the depth of the model is the point.
The most useful thing about a formalized category is that it gives buyers a common test. If you are evaluating anything positioned as Decision Intelligence, take one decision that genuinely matters in your business and ask six questions.
Business intelligence tells you what happened. A copilot helps you interpret it. Neither of those is a system that decides, records why, and finds out whether it worked. If you already have the first two, you do not have the third — and the third is the one the value was always waiting on.
Decision Intelligence is a class of enterprise software that makes the decision itself the object of the architecture. A Decision Intelligence Platform combines decision modeling, analytics and AI to support, augment and automate decisions and to drive business outcomes. In practice that means modeling a decision explicitly, executing it with a human or without one, composing it as a reusable service, monitoring its outcome, and governing it with ownership, authority, version and audit history.
Business intelligence answers what happened and stops at a chart, leaving the decision to a person and leaving no record of it. Decision Intelligence continues past that point: it carries the decision through to a governed choice, an executed action and a measured outcome, and it feeds that outcome back into the next decision. The difference is not analytical power but where the system stops.
Gartner published its first Magic Quadrant for Decision Intelligence Platforms on January 26, 2026, evaluating 17 vendors, followed by companion Critical Capabilities research covering 15 capabilities and four use cases, and a market overview on August 7, 2026. Gartner had described Decision Intelligence as a discipline since 2024; 2026 is when it became a named software market.
Because the economics of the category favor repetition. Modeling a decision is only worth the investment when the decision recurs often enough to justify it, and only frequent decisions generate enough instances to prove the model works. Strategic decisions such as transformation sequencing, capital allocation and operating-model change happen a handful of times a year and take roughly eighteen months to grade, so they fall outside the pattern almost every vendor in the category was built around.