What Decision Intelligence Actually Is

On a Tuesday afternoon in the third quarter, a delivery director makes a call that will decide whether a client is still a client next year.

The engagement is a transformation programme for a large retail group. It is eleven weeks behind, the burn rate has crept past plan, and that morning the client's programme sponsor sent an email with the word "concerned" in the subject line, which everyone in the room understands to mean something closer to "furious." The current team lead is competent but stretched thin across two accounts and has, fairly or not, lost the client's confidence. The delivery director weighs the options for the better part of an hour. She could bring in an external hire, but that takes six weeks she does not have. She could leave the current lead in place and add support around him, but the client has stopped returning his calls. She could go back to the client, admit the plan is broken, and renegotiate the scope, but the relationship is too fragile to survive that conversation right now.

So she makes the call she thinks is right. She pulls her strongest senior delivery lead off a healthy account and moves her onto the struggling one, betting that the healthy account can coast for a quarter and the struggling one cannot wait. She decides. She sends two messages, updates a resourcing sheet, and moves on to the next fire.

That decision was a genuine act of judgement. It rested on evidence, it considered and rejected real alternatives, and it carried a real risk that a lesser manager might not have taken. It was, in the truest sense, the work. And within a year, it will have vanished.

The resourcing sheet will show that the senior lead was on the retail account from the third quarter. The finance system will show what the account made. The CRM will show that the client renewed, or did not. But nowhere, in any of the nine systems this firm runs, will there be a record of the decision: that on a specific Tuesday, a specific person, facing a specific set of bad options, chose this one, for these reasons, at this cost. The record of what happened will survive. The reasoning that produced it will not.

The question this chapter answers

The ledger taught software to record. Business Intelligence taught it to report. Data science taught it to predict. The next capability, the one the whole estate is missing, is the ability to hold the decision itself.

A discipline that cannot be defined precisely is just a mood, a way of gesturing at the future while selling the present. So this chapter does the unglamorous, essential work of definition. By the end of it you should be able to do three things: state what Decision Intelligence is in a sentence you actually believe, tell the difference between Decision Intelligence and the two things it is most often confused with (Business Intelligence and artificial intelligence), and explain to a colleague the single idea the whole discipline is built on.

Decision Intelligence has an atomic unit, the way chemistry has the atom and accounting has the transaction. Get the unit right and the rest of the discipline follows. Get it wrong and you have another dashboard.

The decision object: the atom of the discipline

Here is the definition, stated plainly and then unpacked.

Decision Intelligence is the practice of treating decisions as first-class objects that can be captured, connected to evidence, governed, and learned from.

The load-bearing phrase is first-class object. Software people will recognise it. It means a thing the system knows about directly, that it can name, store, retrieve, relate to other things, and reason over. In your CRM, a customer is a first-class object. So is a deal. In your finance system, an invoice is a first-class object, an entity with a defined shape, a place to live, and a set of things you can do to it and with it. A decision, in almost every firm, is not a first-class object anywhere. It is a ghost. It happens, it moves the business, and it leaves no body.

To make a decision a first-class object, you have to say precisely what one contains. This is the decision object, and it is the single most important idea in this chapter. A properly captured decision holds six things.

The first is the choice. Not the topic, the choice. Not "resourcing on the retail account," but "move the senior delivery lead from the healthy account to the retail account, effective this week." A decision is a fork in the road with a direction taken, and the object records the direction, specifically enough that someone reading it later knows exactly what was done.

The second is the decider, the owner. A named person (or, sometimes, a named group) who is accountable for the choice. Not a department, not a system, a person. A decision without an owner is a decision no one can be asked about, and a decision no one can be asked about is one no one learns from. The owner is the thread you pull when you want to understand.

The third is the evidence. What the choice rested on. The burn rate, the client's "concerned" email, the current lead's other commitments, the senior lead's track record on a similar recovery the year before. Evidence is what connects the decision to the rest of the firm's reality, and it is the difference between a judgement and a whim. Crucially, the evidence is captured as it stood at the moment of the decision, not as it looks in hindsight. The point is to preserve what the decider actually knew, and what she did not.

The fourth is the one almost everyone forgets: the alternatives considered and rejected. The external hire that was too slow. The added support that the client would not accept. The scope renegotiation that the relationship could not survive. The roads not taken are not clutter. They are the shape of the judgement. A choice only means something against the options it beat, and a firm that records only the option chosen has thrown away the reasoning and kept the conclusion. When the retail account is reviewed next year, the question will not be "did moving the lead work?" It will be "was moving the lead better than the things we could have done instead?" You cannot answer that if the alternatives were never written down.

The fifth is the timestamp and context. When the decision was made, and the circumstances around it. A Tuesday in the third quarter, mid-contract, with the client threatening not to renew and the senior lead herself halfway through another delivery. Context is what stops a later reviewer from judging a good decision harshly because they have forgotten how the world looked at the time. Decisions are made forward, in fog, and reviewed backward, in daylight. The timestamp and context keep the daylight honest.

The sixth completes the object, and it arrives late: the outcome. What actually happened. Not what was hoped, what occurred. The retail account stabilised and renewed. The healthy account, starved of its best lead, slipped a little and came in under margin. Was it the right call? The honest answer is that it depends on how you weigh a client saved against a client dented, and reasonable people could disagree. But you can only have that argument, and only learn from it, if the outcome is connected back to the decision that produced it. A decision object without an outcome is a question with no answer. A decision object with one is a lesson.

Of the six, the outcome is the one that matters most, because it is the only one that teaches. The first five capture the decision: they turn a vanished judgement into something the firm can hold. But capture on its own improves nothing. It is the outcome, joined back to the choice that caused it, that lets the firm ask whether the judgement was sound, and it is that question, asked across a thousand decisions, that makes judgement compound instead of merely accumulate. Without it, a decision object is a diary; with it, the diary becomes a teacher.

A record tells you what the organisation did. A decision tells you what it was thinking. Almost every system you own keeps the first and throws the second away.

Put those six things together, choice, decider, evidence, alternatives, context, outcome, and you have a decision object.

It is not exotic. You could capture one in a well-designed spreadsheet row, which is a point we will return to, because it matters enormously. What makes it powerful is not the technology that holds it but the decision, by the firm, that this is a thing worth holding at all.

Notice what the object does. It takes the most valuable and most perishable thing a people business produces, the judgement of its best people under pressure, and it gives that judgement a body. And once judgement is an entity, you can do to it all the things you can do to any first-class object. You can retrieve it. You can connect it to others. You can review it. You can govern it. You can learn from it. You can, eventually, teach a machine to reason over it. None of that is possible while the decision remains a thing that happens in a meeting and dies in an inbox.

This is why the decision object, not the dashboard and not the algorithm, is the atom of the discipline. Everything else in Decision Intelligence is a way of creating these objects, connecting them, and putting them to work.

Decision Intelligence is not Business Intelligence

The fastest way to be misunderstood in this field is to let people hear "Decision Intelligence" and think "Business Intelligence, but newer." It is the difference between a mirror and a memory.

Let us be fair to Business Intelligence first, because it earned its place and the book has no quarrel with it. Business Intelligence took data that was trapped in transactional systems, the ledger, the CRM, the operational databases, and made it visible. It gave leaders reports, dashboards, and the ability to ask "what happened?" and get an answer drawn from records rather than from the loudest person in the room. That was a genuine advance. A great deal of good management rests on the humble competence Business Intelligence provides: knowing your revenue, your utilisation, your margin by client, your pipeline by stage. Do not let anyone tell you that is worthless. It is the floor.

But look closely at what Business Intelligence is, mechanically. It reads the records the other systems produced, aggregates them, and presents the past in a form a human can absorb. It is, by its nature, a rear-view mirror: a very good one, cleaned and angled and lit, but a mirror all the same. It tells you what happened. It is silent on two things that matter enormously.

The first thing it is silent on is what to do next. A dashboard can show you that the retail account's margin is sliding. It cannot tell you whether to move a lead, add support, or renegotiate the scope, because that is a judgement about an uncertain future, and Business Intelligence only knows the certain past. The deciding, the actual work, happens in the space the dashboard cannot reach, which is exactly the space this book is about.

The second thing Business Intelligence is silent on is more subtle and more damaging. It does not remember that you decided anything. Return to the retail account. A year later, Business Intelligence can show you, in full colour, that the account renewed and the other account came in under margin. What it cannot show you, because it never held it, is that these two facts share a cause: a decision, made on a Tuesday, to move a lead from one to the other. The records know the effects. Nobody kept the cause. Business Intelligence reports the wake the ship left. It knows nothing of the hand on the wheel.

This is the loop that Business Intelligence leaves open, and that Decision Intelligence exists to close. In a Business Intelligence world, decisions flow one way: evidence goes in, a human decides, action is taken, and the results eventually show up as new records for the dashboard to aggregate, with the decision itself never captured at any point. The loop is broken precisely where the learning would happen.

Decision Intelligence closes the loop by inserting the missing step. Before the action, it captures the decision as an object: the choice, the reasoning, the alternatives, the evidence. After the outcome, it connects the result back to the object that caused it. Now the firm has something it has never had before, a body of its own decisions and how they turned out, which is the only raw material from which better decisions can actually be made. Business Intelligence answers "what happened?" Decision Intelligence answers "what should we do, what did we choose, and was it right?" The first is a report. The second is a practice.

So the clean distinction, the one to carry: Business Intelligence is rear-view. Decision Intelligence is round-trip. They are not rivals. Decision Intelligence needs good Business Intelligence to feed it evidence. But no quantity of Business Intelligence adds up to Decision Intelligence, any more than a very complete set of photographs of a journey adds up to knowing why you chose the road.

Business Intelligence is a mirror. Decision Intelligence is a memory.

Decision Intelligence is not artificial intelligence

The second confusion is newer and, right now, louder. Since capable language models arrived, every capability in enterprise software has been rebranded with the letters "AI," and it is tempting to assume that Decision Intelligence is just the current fashionable name for "AI that helps you decide." It is not. To see why, you have to notice that they sit on different axes entirely.

Artificial intelligence, stripped of the noise, is a capability. It is a family of techniques for doing certain cognitive tasks with machines: generating text, predicting a number, classifying a case, ranking options, spotting a pattern in more data than a person could hold in mind. It is a way of producing things, answers, forecasts, drafts, scores. Like any capability, it can be present or absent, strong or weak, well applied or badly applied.

Decision Intelligence is not a capability. It is a discipline. It is a way of organising something: a set of practices for capturing decisions, connecting them to evidence, governing who decides what and how, and learning from outcomes over time. Disciplines are not tools. Double-entry bookkeeping is a discipline, and it does not stop being one whether you keep the books in a leather ledger or in cloud accounting software. Decision Intelligence is a discipline in exactly that sense. It is the practice of treating decisions as objects worth capturing and learning from, and it is defined by that practice, not by any particular technology that supports it.

Because they sit on different axes, a capability and a discipline, you can have either one without the other, and both real-world combinations are instructive.

Consider a firm with excellent Decision Intelligence and no artificial intelligence at all. Picture a mid-sized consultancy that has simply decided, as a matter of how it runs, that significant decisions get captured. When the delivery director moves the lead, she spends four minutes logging it: the choice, the alternatives she rejected, the evidence, her reasoning, the risk she is taking. A quarter later, when the outcomes are in, someone connects them back. There is no model anywhere in this. It is a spreadsheet, a shared template, and a discipline that the leadership actually enforces. And this firm has something precious: a growing library of its own real decisions and their consequences, which its partners read before they make similar calls, and which makes the whole firm's judgement compound instead of evaporate. That is Decision Intelligence, practised well, with no AI whatsoever. The rigour is doing the work.

Now consider the opposite firm: every tool it owns has AI in it. The CRM drafts follow-up emails. The support desk auto-summarises tickets. There is a chatbot that answers questions about the handbook and a model that scores leads. By the measure of "how much AI do you have," this firm is well ahead of the consultancy. And yet it captures not a single decision. When its delivery director moves a lead, the reasoning dies in the same inbox it always did. Its AI is fluent and its memory is empty. This is a firm that has bought the tool and skipped the practice.

Hold those two firms side by side, because together they prove the point. If the firm with spreadsheets and rigour has more Decision Intelligence than the firm stuffed with models, then Decision Intelligence cannot be AI. One axis asks "how capable are your machines?" The other asks "how well do you capture, govern, and learn from your decisions?" Confusing the axes is how organisations end up spending a fortune on capability while the discipline that would make the capability worth anything goes unbuilt.

Artificial intelligence is a way of generating answers. Decision Intelligence is the discipline of remembering which answers you chose, and finding out whether they were right.

This is not an argument against AI. It is an argument for knowing what kind of thing you are buying. AI is a genuinely important capability, and the last section of this chapter is about how much more valuable it becomes in the presence of the discipline. But you cannot compose two things well if you think they are the same thing. A firm needs both, but it needs to know which is which.

How the two compose: grounding, revisited

The discipline and the capability are not just compatible. Each is worth far more in the presence of the other, and the reason takes us straight back to an idea from Chapter 2.

Recall the blind cockpit. A large language model is, at its core, a spectacularly capable producer of plausible text. It will answer almost any question fluently, in complete sentences, with the confident cadence of expertise. What it cannot do, on its own, is know whether what it is saying is true of your firm, because on its own it has no access to your firm's reality. Ask a bare model "should we move a lead onto the struggling retail account or renegotiate the scope?" and it will give you a thoughtful, well-structured, generic answer, an answer that would be equally at home in any consultancy on earth, which is precisely the problem. It sounds like judgement and it is actually improvisation, and the danger is that the fluency disguises the emptiness. That is the blind cockpit: an instrument that reads out beautifully and is connected to nothing.

The cure, we said, is grounding: giving the machine something true to stand on. And here is the quiet payoff of this whole chapter. Decision Intelligence is what grounding is made of. The decision objects a firm captures, thousands of real choices, with their real evidence, their real alternatives, and their real outcomes, are exactly the true ground a model needs. They are the firm's own reasoning, made into data.

Watch what changes. Ask the grounded system the same question, and it is no longer improvising. It can reach into the firm's own history and find the twelve times this firm has faced a slipping engagement, what it chose each time, what it rejected, and how each choice turned out. It can note that in the two cases most like this one, moving the lead worked, but that each time the account the lead was pulled from later slipped, which is a cost worth naming out loud. It is reasoning over the firm's real, recorded judgement, and it can show its work, because every claim it makes points back to a specific decision object a human can open and check. That is not a co-pilot flying blind. That is a co-pilot with the instruments finally wired to something real.

So the composition runs in one clear direction, and the direction matters. Decision Intelligence does not need artificial intelligence to exist; the consultancy with the spreadsheets proved that. But artificial intelligence, if you want it to be trustworthy rather than merely fluent inside a specific business, needs Decision Intelligence. Without the record of decisions, even the most advanced model is a better guesser, and a better guesser dressed in the language of certainty is more dangerous than an honest one, because it launders a guess into what sounds like counsel. With the record of decisions, the same model becomes a way of putting the firm's entire accumulated judgement at the fingertips of whoever is deciding next.

The counterpoint: "decisions are too messy to systematise"

Every honest chapter should meet its strongest objection head on, and this idea has a strong one. It usually arrives from the most experienced person in the room, and it goes like this:

Decisions in a real business are not clean. They are messy, human, and political. Half of what actually drove that Tuesday call was things no one will ever write down: a hunch built from twenty years of scars, a worry about a partner's ego, a read on the client's mood, a quiet knowledge of which people work well together and which do not. You cannot reduce that to six fields in an object. The moment you try, you will capture the sterile, defensible version and lose the real reasoning, or worse, people will start deciding for the record instead of for the business. Judgement cannot be systematised. It should not be. This is the sort of thing consultants love and operators know better than to trust.

It is right about almost everything except its conclusion. Take its claims one at a time, because each of them, correctly understood, turns out to be an argument for the discipline rather than against it.

Start with the central worry: that systematising decisions means automating or sterilising judgement. It does not, and this is the crux of the whole misunderstanding. Decision Intelligence is not an attempt to take the human out of the decision. It is an attempt to keep the human's decision, which the firm currently throws away. The delivery director still makes the call, still uses her twenty years of scars, still reads the client's mood. All it asks is that the judgement she already exercised leave a trace, so the firm can hold it, connect it, and learn from it, instead of losing it the moment she moves on to the next fire. The goal is not to automate the decision. It is to make it visible, connected, and improvable.

Now take the claim that the real reasoning is tacit, a hunch built from scars, and cannot be fully written down. This is true. And it is the most powerful argument for capture, not against it, once you see it clearly. Ask yourself what happens to that tacit judgement today. Today, it lives entirely in the head of one experienced person. When she is on leave, it is unavailable. When she leaves, it walks out of the door with her, and the firm relearns, expensively, what she already knew, exactly the amnesia this book diagnosed earlier. The messiness of judgement is not a reason to leave it uncaptured. It is a description of what the firm is currently losing. You will never capture all of it, and no serious version of the discipline pretends you can. But capturing the choice, the alternatives she weighed, the evidence she leaned on, and how it turned out captures far more than the zero the firm captures now. The objection assumes the alternative to imperfect capture is perfect tacit wisdom, safely retained. It is not. The alternative to imperfect capture is total loss.

Then the sharpest version of the worry: that capture will corrupt the very thing it captures, that people will decide for the record, defensively, or game it. This risk is real, and it is a design problem, not a reason to abandon the project. It is real in every system of record humans have ever built. People can dress up their reasons in a decision log exactly as they can dress up their hours on a timesheet or their pipeline in a forecast, and the answer is the same answer good organisations already use: capture lightly, at the moment of deciding rather than in a later audit, govern with judgement rather than bureaucracy, and make the record something that helps the decider next time rather than something used to punish her this time. A decision log that is used to hang people will be gamed within a month and deserve to be. A decision log that visibly makes the firm's next hard call easier, because it holds the last twelve, earns honest entries because honesty is what makes it useful to the very people filling it in. Culture decides which one you get, and culture is buildable.

The messiness of decisions is not a bug that the discipline has to work around. It is the entire reason the discipline is needed. Clean, mechanical decisions, the ones with an obvious right answer, do not need capturing, because anyone can reconstruct them and no one disagrees about them. It is precisely the messy ones, the judgement calls made in fog under pressure with no clean answer, that are worth their weight in gold and that vanish most completely. The mess is not the argument against Decision Intelligence. The mess is the whole case for it.

A decision you cannot revisit is a lesson you cannot learn. And the messiest decisions, the ones hardest to revisit, are the ones you most need to learn from.

So the objection is right that decisions are messy, human, and political, and right that you can never fully reduce judgement to fields in an object. It is wrong only in what it concludes. The aim was never a sterile firm that decides by formula. The aim is a firm whose messy, hard-won human judgement stops evaporating, and starts to compound.

What you can now do

This chapter set out to leave you able to do three specific things.

You can define Decision Intelligence: the practice of treating decisions as first-class objects that can be captured, connected to evidence, governed, and learned from.

You can distinguish it from its two neighbours. Business Intelligence reports the past from records and points only backward; Decision Intelligence captures the decision and closes the loop by connecting it to what happened. And artificial intelligence is a capability for producing answers, while Decision Intelligence is the discipline that surrounds decisions, so a firm can have excellent Decision Intelligence with nothing but spreadsheets and rigour, or the most advanced AI with no Decision Intelligence at all, and the second firm has bought a capability with no ground to stand on.

And you can explain the decision object to a colleague, which is the test that matters most, because the object is the atom the whole discipline is built from. A captured decision holds six things: the choice that was made, the person who made it, the evidence it rested on, the alternatives that were considered and rejected, the timestamp and context, and, arriving later and completing the object, the outcome, what actually happened. Give judgement a body, and you can do with it all the things you can do with anything real: connect it, govern it, learn from it, and one day ground a machine on it. Leave it a ghost, and you lose it every time, forever.

The bridge

All of which raises an obvious and fair question. This is a fine idea. Is anyone actually doing it?

It is one thing to define a discipline cleanly and quite another to show that it is real, alive, and taking hold rather than a tidy abstraction that sounds good in a book and dissolves on contact with a Monday. So the next chapter turns from definition to reality. It looks at the state of Decision Intelligence as it actually stands: where the idea came from, why now and not ten years ago, and, most usefully, where you can already see the discipline emerging in the firms that are practising it before they have the word for it, in the decision logs and the operating reviews and the quietly renamed functions that are really decision functions in disguise. It will be honest about how early and uneven the picture is, because the honest picture is the persuasive one. The definition is settled. Now let us see who is living it.