In a tournament year, everyone in every industry gets asked the same question by someone at dinner: is artificial intelligence going to change this too. Football is a convenient thing to point at when answering, because most people already have an intuition for how the sport works, and that intuition turns out to be enough to make the useful point.
The useful point is not about models. It is that sport is an unusually good environment for this kind of technology, and the reasons have almost nothing to do with the sophistication of the software. They have to do with the condition the data is in before anyone builds anything. That is the part worth borrowing.
What this piece is, and is not: this is an analogy about data conditions, not a report on sports technology. We are not describing what any tournament, league or governing body has deployed, and we name no systems, vendors or officials. Everything said about football below is general enough that you can check it against your own understanding of the game. If it does not match, discard the analogy and keep the argument about funds.
1. A bounded domain where everyone agrees what happened
A pitch is a bounded space. The rules are fixed before play begins and do not change during the match. The things that happen in the match are countable and consistently defined: a pass is a pass, a substitution is a substitution, and two people watching the same event will label it the same way. That agreement is not a technical achievement. It is a definitional one, arrived at long before anyone thought to compute on it.
Now ask two private-market funds what "ARR" includes. One counts contracted revenue not yet invoiced; the other does not. One annualises the last month; the other takes a trailing average. Ask what "committed capital" means and you will find honest disagreement about recycling, about defaulted commitments, about capital committed to a parallel vehicle. Neither fund is wrong. They are using different definitions of the same word.
The consequence is mundane and severe: the same question, asked twice, returns two different answers, and no amount of model quality repairs that. Definition is a prerequisite, not a nicety. A firm that has never written down what its own terms mean is not ready for automation of any kind, because the automation will faithfully reproduce the ambiguity at speed.
2. Instrumentation at the source
The second condition is that in sport, the record of an event is created as the event happens, by whatever is observing it. The same match is watched by many instruments at once, all of them pointed at the same bounded space, all of them producing their own version of the same few minutes. Nobody reconstructs the match afterwards from a written summary.
Compare that with how a fund learns what its portfolio did. The update arrives as prose in an email, six weeks after quarter end, written by a founder from memory and a spreadsheet, then retyped by an analyst into a different spreadsheet. Three transcriptions separate the event from the record. Each one is a chance to lose precision, and none of them is recoverable later, because the original was never captured.
This is the least glamorous and most decisive gap. Capture at the source means a portfolio company reports into a structured form with defined fields, and a bank or custodian delivers a transaction through a connection rather than a PDF. It is plumbing. It is also the reason one organisation can compute on its own history and another cannot.
Sport did not get good at analysis because it got clever. It got good at analysis because it had already agreed what a pass was, and had already arranged to record every one of them.
3. A record nobody disputes
Officiating decisions in football are largely binary — the ball crossed the line or it did not, the player was offside or was not — and some of them can be reviewed after the fact. Review only works because there is an authoritative record of what happened that both sides accept in advance. Where the record itself is contested, review is theatre: it produces a decision without producing agreement.
Funds get this wrong more often than they admit. A firm where the distribution waterfall lives in a file one person maintains has no authoritative record. It has one person's working. When an investor queries an allocation, the review consists of that person re-explaining their own spreadsheet, which is not a review at all. The same is true of a valuation nobody can trace, or a capital account that disagrees with the last statement sent out.
The requirement is not sophistication. It is that the waterfall is a computation the system performs, on a record with a history, so that the question "why is my number this" has an answer that does not depend on who is in the room.
4. A person at the end of it
The interesting uses of analysis in sport inform a decision that a person still makes and owns. Someone chooses the substitution. Someone signs the player. Someone accepts or overrules what the review indicates. The analysis narrows the question; it does not answer it, and no one pretends the accountability has moved.
Hold to the same standard when money moves or when something reaches an investor. An agent can assemble a capital call, reconcile receipts against expectations, chase a missing tax declaration, or draft a quarterly letter from the underlying data. Every one of those should arrive at a named human who approves it, and the approval should be recorded. We build agents on that assumption, because the alternative — automation with no owner — is the version that fails audit and, eventually, fails the investor.
The honest asymmetry
Here is the part the dinner-table version leaves out. Sport had years of unglamorous work behind it before any of this produced value: agreeing definitions, standardising how events are described, arranging to capture them as they happen, and building records that competing parties would accept. The analysis arrived last and got the credit.
A fund that wants comparable leverage has to do comparable work, and it does not get to skip to the interesting part. That means writing down what your terms mean and enforcing them in one system rather than four. It means moving portfolio reporting from prose to fields, and transaction data from attachments to connections. It means one authoritative record of capital, with history, that survives the departure of whoever built it.
None of that is a technology problem in any deep sense. It is a sequencing problem, and the sequence is not negotiable. Firms that put the model first get a faster version of their existing confusion. Firms that fix the data conditions first find that the clever part turns out to be the easy part — which is, roughly, what the sport analogy has been saying all along.