Decision IntelThe feedback arrives five years after the decision
Every other input into a buyout gets better with repetition. This one cannot. By the time a vintage resolves, the regime that produced it is gone and the result cannot be attached to the assumption that drove it, so the process never learns which assumption that was. Decision Intel is built for decisions with that shape — and in this asset class, they are the ones carrying the largest checks.
Where the capital is, and the feedback is not
Two properties of an environment decide whether experience turns into judgement: whether the situation repeats in a form you can recognise, and whether the result comes back fast enough to attach to what caused it. Plot them against each other and the domains separate cleanly. Large-cap buyout underwriting sits in the corner where neither holds, and that corner is what this product is built for.
Why that corner is hard, stated precisely
Neither property is about the people in the room, and that is the whole point. Both are facts about the field. They are the reason fifteen years in this asset class does not produce the kind of calibrated instinct that fifteen years in anaesthesiology does, and the reason no amount of additional deal flow would change it.
An environment regular enough to contain cues
The situation has to repeat in ways that carry signal. Not identically — but enough that a pattern seen before means something about the pattern in front of you.
A chess position, a ward round, a structure fire. The pieces are finite, the physics is stable, and the same configuration means the same thing this year as last.
Every large-cap deal is assembled from a different capital structure, a different counterparty set, a different regulatory posture and a different point in the cycle. The parts recur; the configuration almost never does.
Feedback fast enough and clean enough to learn from
You have to find out whether you were right, soon enough to connect the answer to the judgement, and cleanly enough to know which judgement it was answering.
The anaesthetist knows within minutes. The firefighter knows within seconds. The chess player knows by the end of the game, and plays another that evening.
A buyout resolves over five to seven years, through a macro regime that has since changed, with the outcome attributable to a dozen causes and the original team partly dispersed. A partner may complete twenty of these in a career.
What this argument does not say
Every one of these is a conclusion a reader could reasonably draw, and none of them follows. Worth being explicit, because the useful version of the argument survives them and the careless version does not.
It does not say the partner is bad at their job. It says the environment cannot teach one specific thing, however good they are.
It does not say experience is worthless. Experience in this field builds real, transferable skill in running a process, reading a counterparty, structuring a negotiation and managing a committee — all tasks with tight enough feedback to learn.
It does not say more deals would fix it. Twenty repetitions of an irregular task with delayed feedback produce confidence, which is a different thing from calibration, and the two are indistinguishable from the inside.
It does not say a machine judges better. It says the failure is a missing feedback loop and a missing outside view, and both of those are addressable without anyone deferring to anything.
An error that repeats because nothing corrects it
A missing feedback loop is abstract until you price it. At a hundred million and up per equity check, with the structure fixed at signing, these are the three consequences a deal partner will recognise from their own firm — and will struggle to name the process that addresses any of them.
The lesson arrives with no return address
A 2019 vintage resolves around 2025. By then the credit environment has turned over twice, the sponsor who underwrote it holds a different seat, and the result is the sum of an entry price, a financing structure, an operating plan and a macro regime. Nothing in that sequence tells the next committee which assumption was the load-bearing one. The outcome is known and the lesson is not.
Conviction compounds, calibration does not
Repetition still does its work — pattern recognition sharpens, and the tenth deal genuinely reads faster than the first. What repetition cannot supply here is the error signal, because the error arrives too late and too confounded to attach to anything. Twenty deals of practice with no correction produces conviction rather than calibration, and from the inside those two feel identical.
The structure is fixed the moment you sign
Operating partners can improve an asset. They cannot reprice the entry, reopen the covenant package, restore a customer concentration the model assumed away, or add an indemnity that was never negotiated. Whatever protection the structure does not contain on day one is not available later, which is why the reasoning has to be examined before commitment rather than during the hold.
Manufacture the feedback the environment will not give you
The two properties are facts about the field and cannot be argued with. What can be changed is whether the decision gets an independent structural read before commitment, and whether that read leaves a record the outcome can later be graded against. Four mechanisms, each of which the artifact shows rather than asserts.
Read the packet as of the decision date
Retrieval is off, entity names are redacted before the reasoning layer sees them, and no account history is in scope. The read cannot be flattered by knowing how the deal turned out, because the run has no route to that information — and the artifact states which of those three toggles were set, so the claim is checkable rather than asserted.
Supply the outside view the firm cannot supply itself
One firm sees its own deal flow. The engine reads each decision against a register of 1,307 merger registrations that have already resolved, matched on structure rather than on sector or story. That is the reference class a partner has no way to hold in their head, and it is where "this shape of deal has failed before, and here is the mechanism" comes from.
Make the structural read reproducible
The layer that finds the fragilities is deterministic: the same document produces the same read every time it is run. That matters more than it sounds, because it is the difference between an opinion and an instrument. The language model phrases the finding; it does not decide what the finding is.
Close the five-year loop in an afternoon
Submit a deal that has already resolved. The engine reads it blind, then the day-one read is graded against what actually happened, with misses recorded at the same weight as hits. That is the feedback the environment refuses to produce, manufactured on demand — and because you already hold the outcome, you are grading us rather than trusting us.
Time to signal is one of seven variables, and this is why
Whether an audit is worth running on a given decision is decided by seven things multiplied together: how structurally complex it is, how much capital rides on it, how formal the process is, how much of the reasoning is written down, how much of the outcome is locked at signing, how often the decision recurs, and how long it takes to find out. The seventh is the one this page is about. A decision whose result arrives in a week teaches its own lesson; a decision whose result arrives in six years never does, and that is precisely where an external record earns its place.
The seven variablesWhere the two properties come from
Kahneman, D. & Klein, G. (2009). Conditions for Intuitive Expertise: A Failure to Disagree. American Psychologist, 64(6), 515-526.
The source of the two conditions. Written jointly by researchers from opposing traditions, which is what makes it hard to dismiss as one camp’s position.
Klein, G. (1998). Sources of Power: How People Make Decisions. MIT Press.
The recognition-primed decision model. The case that expert intuition is real, load-bearing, and worth building on.
Kahneman, D. & Lovallo, D. (1993). Timid Choices and Bold Forecasts. Management Science, 39(1), 17-31.
The inside view and the outside view, and why forecasters who have the reference class available still do not use it.
Kahneman, D., Sibony, O. & Sunstein, C. (2021). Noise: A Flaw in Human Judgment.
The distinction between noise and bias, and why the same reader scores the same document differently on different days.
Test it on a deal you already know the answer to
Give us one closed transaction. We read the packet as of the decision date, with the outcome withheld from the engine, and you compare what it named against what actually happened. You already hold the ground truth, so you do not have to take our word for any of it.
Request a blind audit