Learn

Decision intelligence is the discipline of engineering decisions themselves: connecting data, models, and human judgment so a specific decision can be framed, supported, traced back to its evidence, and improved from its outcomes.
Decision intelligence is the discipline of engineering decisions themselves: connecting data, models, and human judgment so a specific decision can be framed, supported, traced back to its evidence, and improved from its outcomes. It is also called decision engineering, and in software terms a decision intelligence platform is the system that carries a decision from raw data to an accountable choice.
Last updated: November 16, 2026
Why it matters to your mission. Most defense organizations do not lack analysis. They lack a dependable path from analysis to a decision someone can defend. A watch floor can hold a dozen dashboards and still leave a commander asking the only question that matters: given what we know right now, what should we do, and why? Decision intelligence starts from that question and works backward to the data, instead of starting from the data and hoping a decision falls out.
The distinction is practical. Business-style analytics answers "what happened?" Decision intelligence answers "what should we do next, what will it cost, and how will we know if we were right?" For defense and intelligence missions, that last part carries the weight. A recommendation that cannot be traced to its sources, or checked against what actually happened, does not survive contact with a commander, an oversight review, or an adversary who is deliberately feeding you bad data.
Key takeaways
Decision intelligence by the numbers - As much as 80 percent: the share of their time Army leaders have historically spent on data and information processing rather than higher-order thinking, according to a Modern War Institute analysis (Mingus and Daker, MWI, 2026). - Ten-fold: the increase U.S. Central Command reported in the volume of targets it can identify and act on within 24 hours after data-integration efforts, as recorded by the Government Accountability Office (GAO-25-106454). - Seven-fold: the decrease in average decision-making time from machine decision aids in the Air Force's first DASH human-machine teaming sprint, per the initial analysis reported by the Air Force Research Laboratory (AFRL, 2025). This is a single experiment, not an operational result.
Decision intelligence matters in defense because command is, at its core, a sequence of decisions made under time pressure with incomplete information, and oversight reporting shows manual, swivel-chair data handling still limits that tempo. The Government Accountability Office reports that "maintaining information and decision advantage is one of DOD's top priorities," and that "DOD is trying to pivot toward more automated, scalable decision-making tools, according to its command and control modernization strategy" (GAO-25-106454, Defense Command and Control, 2025).
The same report names the obstacle. Under the "swivel chair" model, an analyst receives inputs and manually enters data from different systems, and GAO finds that this approach "makes it more difficult for warfighters to make decisions at the scale and tempo needed to effectively accomplish their missions." The data exists. The decision path from that data to an action does not.
Congress has framed the opportunity the same way. The Congressional Research Service notes that AI "could, for example, facilitate autonomous operations, lead to more informed military decisionmaking, and increase the speed and scale of military action," and that AI may be able to give commanders "a menu of viable courses of action based on real-time analysis of the battle-space, potentially improving the quality and speed of wartime decisionmaking" (CRS R45178, Artificial Intelligence and National Security, updated 2020).
Put those together and you have a documented government need: the ability to turn fragmented data into decision-ready options, quickly, with a human still deciding. Decision intelligence is the name for designing that path deliberately. It is also the practical engine behind decision advantage, the competitive condition defense strategy is chasing: decision advantage is the outcome, and decision intelligence is one disciplined way of producing it.
Decision intelligence works by modeling the decision first and then wiring data and analytics to it, rather than collecting data and searching it for insight. A useful way to see it is as five linked elements that every well-engineered decision has, whether the decision is a retasking order or a logistics reallocation:
The ordering is the point. Starting from the decision tells you which data matters, which is the cure for the familiar problem of drowning in feeds that never change an outcome. It also forces a confidence statement onto every link, so the decision-maker sees not only a recommendation but how strongly the evidence supports it.
In a defense setting the fourth element, the evidence, is usually the expensive part. Mission data arrives from many systems, in many formats, under different classifications. Before any model can score an option, those fragments have to be connected into a coherent, traceable picture. That is the same make-sense problem that sits underneath multi-INT fusion, and it is why decision intelligence in defense is as much a data-infrastructure problem as an analytics one.
A concrete defense example makes the abstraction useful. Consider a maritime operations center that receives three signals in the same hour: a commercial vessel's Automatic Identification System (AIS) transmissions stop near a strait, a partner report mentions an unscheduled ship-to-ship transfer in the area, and overhead imagery from earlier in the day shows two hulls close together. The commander has one patrol aircraft available and a competing tasking elsewhere.
Without decision intelligence, each signal sits in its own system. An analyst correlates them by hand, writes a summary, and the commander decides on the basis of a briefing that may already be stale.
With decision intelligence, the problem is framed as a decision from the start:
The commander still decides. What changes is that the options, the trade-off, and the evidence arrive together, already connected, with a clear line back to every source. This is the same pattern that underpins maritime domain awareness work generally: the signal lives in the correlation, and the value lives in turning that correlation into an action someone can defend.
Business intelligence describes and diagnoses what has happened; decision intelligence models what to do next and holds itself accountable for the result. The two are complementary rather than competing: decision intelligence typically consumes the same authoritative data that business intelligence reports on, then adds options, consequences, confidence, and a feedback loop.
| Dimension | Business intelligence | Decision intelligence |
|---|---|---|
| Core question | What happened, and why? | What should we do next, and why? |
| Starting point | The available data | The decision to be made |
| Primary output | Dashboards, reports, metrics | Options with expected consequences and a recommended action |
| Treatment of uncertainty | Often implicit or omitted | Explicit confidence attached to each link and recommendation |
| Role of the human | Interprets the report and decides | Decides with options, trade-offs, and evidence laid out together |
| Accountability | Reports on outcomes after the fact | Records what was decided, why, and what happened, then learns from it |
| Defense analogue | A readiness dashboard or intelligence summary | A course-of-action recommendation tied to its supporting evidence |
For a defense staff, the practical difference shows up in how long it takes to get from the screen to the order. A dashboard tells the commander the situation. A decision intelligence capability tells the commander what the options are, what each is likely to cost, and how sure the evidence is, which is the input the AI course of action generation process is built to produce.
No. Decision intelligence is a discipline for structuring decisions, and AI is one set of tools it can use. A decision intelligence capability might use machine learning to estimate a likelihood, natural language processing to pull entities out of reporting, graph analysis to expose relationships, or simple deterministic rules to check a recommendation against doctrine and constraints. It might also use no AI at all for some links, where a lookup table or a human estimate is more reliable.
"AI decision intelligence" usually means a decision intelligence capability in which AI does most of the evidence processing and option scoring at machine speed. That is where the payoff is in defense, because the volume of mission data outruns any staff working by hand. It is also where the risk concentrates. A probabilistic model can produce a fluent, confident answer that is wrong, and if the decision layer cannot show where that answer came from, the error flows straight into an order.
The safer pattern separates the roles. Probabilistic AI handles the work it is good at, such as extraction, correlation, and pattern detection, while the logic that a commander relies on stays inspectable. Where a decision must conform to doctrine or rules of engagement, deterministic evaluation that returns a traceable verdict is often the better fit, which is the subject of deterministic decision support for defense. The broader trade-off is covered in deterministic vs. probabilistic AI.
A decision intelligence platform is software that carries a decision from raw, multi-source data to a recommended action, with the evidence, logic, and confidence visible at every step. In practice it has to do four jobs, and most tools on the market do only one or two of them:
The first three jobs are the evidence layer, and they are where defense programs most often stall. A platform that offers polished recommendations on top of unresolved, untraceable data is not decision intelligence; it is a confident interface over the same fragmentation. The fourth job is what separates decision intelligence from analytics, and it is why the platform must preserve human judgment rather than replace it. The relationship between the human and the machine in that loop is explored in human-machine teaming and human-in-the-loop vs. human-on-the-loop.
Decision intelligence overlaps with several established terms. The differences are about scope and starting point.
| Term | What it is | How it relates to decision intelligence |
|---|---|---|
| Decision support system | Software that helps a human reach a decision by gathering, filtering, or scoring information | A component; decision intelligence adds explicit decision modeling, confidence, and an outcome feedback loop |
| Decision advantage | The competitive condition of understanding, deciding, and acting faster and better than an adversary | The outcome; decision intelligence is one disciplined way to produce it |
| Business intelligence | Descriptive and diagnostic reporting on what has happened | An input; decision intelligence builds on the same data to recommend what to do next |
| Operational intelligence | Real-time visibility into ongoing activity and operations | A source of current evidence that feeds the decision model |
| Data analytics | The broad practice of examining data to find patterns and answers | The toolkit; decision intelligence directs analytics at a named decision |
| Course of action generation | Producing candidate plans that accomplish a mission | A specific decision intelligence application in military planning |
The DIKW progression (data, information, knowledge, wisdom) is a useful way to place decision intelligence: it operates at the top of that hierarchy, where knowledge becomes a judgment about action. See the DIKW pyramid for that framing.
Trust is the hardest part because a recommendation is only as useful as the decision-maker's ability to check it, and most AI outputs are difficult to check. The intelligence community has been blunt about this. The CRS report on AI and national security quotes a senior CIA science and technology official: "Until AI can show me its homework, it's not a decision quality product" (CRS R45178). The same report concludes that increasing explainability "will thus be key to humans building appropriate levels of trust in AI systems."
"Showing the homework" has a concrete meaning for decision intelligence. For every recommendation, the decision-maker should be able to see which evidence supported it, where that evidence came from, how old it is, how confident each link is, and what rule or model connected evidence to option. Without that, two failure modes appear. Operators over-trust a fluent system and miss its errors, or they under-trust it and fall back to manual correlation, which erases the speed the system was bought for.
Federal standards point the same direction. The National Institute of Standards and Technology's AI Risk Management Framework exists "to improve the ability to incorporate trustworthiness considerations into the design, development, use, and evaluation of AI products, services, and systems" (NIST AI RMF). For decision intelligence, that translates into provenance on every input, inspectable logic on every recommendation, and a recorded rationale on every decision. Two related pieces go deeper: data provenance for AI and auditable AI for defense.
If you are evaluating decision intelligence for a defense or intelligence mission, the questions that separate a capability from a dashboard are specific. Require:
Evaluating a capability? Bring these eight requirements to a scoping call and we will walk each one against your environment: talk to Torch.AI.
Torch.AI builds government-owned reasoning infrastructure for the part of decision intelligence where defense programs most often stall: the evidence layer, where fragmented, multi-source data has to become coherent understanding before any recommendation can be trusted. The reasoning layer connects to mission data where it already lives and operates on top of existing systems of record rather than replacing them, the system of reasoning vs. system of record distinction at the center of the approach.
The products map to the jobs a decision intelligence platform has to do. ORCUS is the data-availability layer, connecting mission data across systems and preserving where each element came from. NEXUS is the analytical layer above it: it transforms data into high-fidelity semantic representations that encode meaning, time, and location together, and packages that work into reusable Knowledge Domains, so each new decision starts from structured meaning rather than raw files. HALO applies graph-based reasoning so the relationships among units, assets, locations, and events are maintained continuously instead of rebuilt by hand, which is the input indications and warnings, targeting, and course-of-action work depend on. Where a recommendation has to conform to doctrine or rules, deterministic evaluation can return a traceable verdict instead of a generated guess.
The human stays in command. The infrastructure is built to put options, evidence, and confidence in front of the person accountable for the decision, with every element traceable to its source, not to decide for them. Ownership varies by deployment and purchase; government-owned deployment is available when the mission requires the customer to control the reasoning layer and its provenance. See how this is packaged on the Torch.AI software page.
Because command is a sequence of time-pressured decisions, and oversight reporting shows manual, swivel-chair data handling still limits the needed tempo. GAO reports that DOD is pivoting toward more automated, scalable decision-making tools and that manual swivel-chair analysis makes it harder to decide at the scale and tempo missions require. Decision intelligence designs the path from fragmented data to a defensible decision.
It models the decision first, then wires data and analytics to it. A well-engineered decision names the choice and its owner, the available options, the expected outcomes and the cause-and-effect links between them, the evidence behind each link with its source and confidence, and a feedback loop that records what was decided and what happened.
A maritime operations center deciding whether to retask its only patrol aircraft after a vessel's AIS goes dark near a strait. Decision intelligence frames the options, resolves the AIS gap, a partner report, and overhead imagery to the same vessel with their sources, estimates the likely payoff of each option against the cost of leaving the original tasking, and records the outcome to calibrate the next decision.
Business intelligence describes and diagnoses what has happened through dashboards and reports. Decision intelligence starts from the decision to be made and models the options, their likely consequences, and the confidence of the evidence behind a recommended action, then records the outcome. The two are complementary: decision intelligence builds on the same authoritative data.
No. Decision intelligence is a discipline for structuring decisions, and AI is one set of tools it can use for extraction, correlation, likelihood estimation, and option scoring. Some links are better served by deterministic rules or human estimates. In defense, the safer pattern keeps the logic a commander relies on inspectable, even when AI does most of the evidence processing.
A decision intelligence platform is software that carries a decision from raw, multi-source data to a recommended action with evidence, logic, and confidence visible at every step. It must make data available across systems, make it meaningful through enrichment, make relationships understandable through entity resolution and graph analysis, and make the decision accountable by recording the choice, its rationale, and its outcome.
Because a recommendation is only useful if the decision-maker can check it. A senior CIA official, quoted by the Congressional Research Service, put it as "Until AI can show me its homework, it's not a decision quality product." Decision intelligence earns trust through provenance on every input, explicit confidence, inspectable logic, and a recorded rationale for every decision.
Describe the problem.
Define the constraints. We will structure the session.
No demo.
No slide deck.
Real work.
Describe a decision your staff makes under time pressure. We will scope decision intelligence around your data. No dashboard tour. Your people deciding. Real work.