Decision intelligence (DI) is the discipline of engineering how decisions get made: it combines data, analytics, AI models, business rules and human judgment into an explicit decision model, connects the chosen action to the systems that execute it, and measures the outcome so the next decision is better.
A precise definition
Most organizations are rich in data and poor in decisions. Dashboards describe the past, models predict the future, and yet the moment that matters — what should we do now, and who gets to decide? — is still handled in email threads and meetings. Decision intelligence treats that moment as something that can be designed, measured and improved.
Decision Intelligence (DI) is a practical discipline that advances decision making by explicitly understanding and engineering how decisions are made, and how outcomes are evaluated, managed and improved via feedback.
Gartner, as quoted in coverage of the 2025 Hype Cycle for AI
In practice, a decision intelligence system answers five questions for every important decision: what changed, what it means, what we could do, why one option is better, and who has the authority to act. It then records what happened, so the organization learns.
Where the term came from
The idea grew out of decision analysis, management science and data science. Lorien Pratt and Mark Zangari published early work on decision modeling for complex business problems in the late 2000s. In 2018 Google renamed parts of its applied data science practice “decision intelligence” to emphasize that data only matters when it changes an action — a framing popularized by Cassie Kozyrkov, then Google’s chief decision scientist.
Analyst attention followed. Gartner named decision intelligence one of its top strategic technology trends for 2022, rated it “transformational” on the 2025 Hype Cycle for AI, and published its first Magic Quadrant for Decision Intelligence Platforms in January 2026 — a sign that DI has moved from concept to a software category buyers evaluate.
How decision intelligence works
Every decision intelligence implementation, whatever the industry, follows the same loop:
↺ Every outcome feeds the context of the next decision.
- Signal. Something changes: a supplier slips, a machine vibrates, a claim is denied, a price moves.
- Context. The system assembles only the context this decision needs — orders, schedules, policies, history — and says explicitly what is missing. (See decision context.)
- Options. Feasible actions are generated and simulated against cost, risk and time. (See scenario intelligence.)
- Evidence. Each recommendation carries its sources, confidence and the disagreements it found.
- Decide. Policy determines whether a human approves, a human monitors, or the system acts on its own. (See decision autonomy.)
- Act. The decision is written back into the systems that do the work — ERP, MES, CRM, a robot fleet manager. (See decision execution.)
- Learn. The outcome is measured against the expectation and fed into the next decision. (See decision observability.)
A worked example
At 3:14 a.m. a component supplier confirms a four-day delay. Two customer orders due Thursday depend on the part.
| Step | What the DI system does |
|---|---|
| Signal | Detects the slipped delivery date in the supplier portal feed. |
| Context | Links it to open orders in ERP, the line schedule in MES and inventory from IoT sensors; flags that Supplier B’s capacity is unknown and requests it. |
| Options | Wait (−$310K at risk), move 2,400 units to Supplier B (+$216K protected), or split production (+$120K). |
| Evidence | Eight sources support Supplier B at 87% confidence, including lead-time history and contract terms. |
| Decide | Policy says reallocations above $100K need procurement approval — a named buyer approves in one click. |
| Act | A purchase order is drafted in SAP and the MES schedule updates. |
| Learn | The actual delivery and margin are compared with the forecast and improve the next recommendation. |
Nothing in this example requires exotic AI. What is new is that the decision itself is a first-class, observable object rather than an undocumented judgment call. The same pattern appears in manufacturing, healthcare, logistics and robotics.
Decision intelligence vs. BI, analytics and AI
| Business intelligence | AI / machine learning | Decision intelligence | |
|---|---|---|---|
| Core question | What happened? | What is likely to happen? | What should we do, and who decides? |
| Primary output | Dashboards and reports | Predictions, classifications, generated content | A recommended action with evidence, authority and a feedback loop |
| Unit of work | The metric | The model | The decision |
| Closes the loop to action? | No — a human interprets | Rarely — output is an input | Yes — writes back to operational systems |
| Measured by | Report usage, data freshness | Accuracy, precision, recall | Decision quality, latency and business outcome |
Decision intelligence does not replace BI or AI; it sits on top of them and uses both. The full comparison is in Decision Intelligence vs Business Intelligence vs AI.
Why decision intelligence matters now
Three shifts make DI urgent rather than academic:
- AI agents can now act. Large language models and agents are being wired into operational systems. Without an explicit decision layer, nobody can say which decisions an agent may take, on what evidence, or how to undo them. Gartner warns that by 2027, 25% of ungoverned decisions made with LLMs will cause financial or reputational loss.
- Physical AI raises the stakes. When software moves robots, trucks and valves, a bad decision is a collision or an outage, not a wrong chart. (See decision intelligence for robotics and physical AI.)
- Speed has become a competitive variable. Gartner predicts that by 2030, explicitly modeled business decisions will be five times more trusted and 80% faster than ungoverned ones, and that by 2027 half of business decisions will be augmented or automated by AI agents for decision intelligence.
Where to start
Start with one recurring, high-value decision — not a platform migration. Good first candidates are decided often, cost real money when wrong, already have data behind them and have a clear owner. The 90-day implementation guide walks through choosing, modeling, piloting and scaling that first decision, and the platform guide explains what capabilities to look for when you buy or build.
- Decision intelligence makes the decision — not the dashboard or the model — the unit of work.
- It combines data, AI, rules and human judgment in an explicit, auditable decision model.
- It closes the loop: recommend → approve → act → measure → learn.
- Autonomy is granted per decision by policy, never assumed.
- Start with one high-value, recurring decision and prove the outcome.
Frequently asked questions
What is decision intelligence in simple terms?
Is decision intelligence the same as AI?
What is an example of decision intelligence?
Who coined the term decision intelligence?
What industries use decision intelligence?
Sources
- Gartner definition of Decision Intelligence, as quoted in coverage of the 2025 Gartner Hype Cycle for AI (Cloverpop)
- Gartner, Magic Quadrant for Decision Intelligence Platforms (inaugural edition, January 2026)
- SD Times — Gartner acknowledges growth of Decision Intelligence Platforms with inaugural Magic Quadrant (March 2026)
- Wikipedia — Decision intelligence