The Arbitrage Window

Capital is allocating to sport with the structural conviction it brought to real estate after the REIT cycle of the 1990s — when changes to US tax and pension rules turned real estate from a private, relationship-driven asset class into a publicly tradable institutional one — and to defense after 2001, when post-9/11 budget expansion and later procurement reform turned the Pentagon into one of the world's most consequential software buyers.⁰ The decision infrastructure that capital relies on in those domains — auditable, repeatable reasoning chains that turn data into defensible institutional recommendations — has not yet arrived in sport at the same speed. At the same time, current-generation AI has collapsed the cost and time of building such infrastructure for a small team led by an operator with the right domain judgment. A nimble, ambitious institution can attach itself to the new capital influx with enough force to win a meaningful slice of it before the incumbents rebuild. The window is open because most organizations resist rebuilding from definitions up; it will close when the median operator inside the vertical has rebuilt, when the well-capitalized buyer builds the layer in-house, or when a rollup assembles coherent method at scale. This essay is about why the window is open, what occupies it, and what would prove the claim wrong.

Read as PDF

That gap is the window.

Capital Has Outrun Its Software

Sport is now an institutional asset class — not in the magazine-cover sense, but in the sense that the firms managing pension obligations and sovereign wealth allocate to it. Apollo Sports Capital completed its investment to become majority shareholder of Atletico Madrid in March 2026, with Apollo describing Atletico as the platform's flagship majority equity investment and committing up to EUR 100 million of additional strategic capital to support the club's long-term plans.¹

This is not one firm's bet. Ares Management raised $3.7 billion of dedicated sports, media, and entertainment capital.² CVC Capital Partners established Global Sport Group, bringing together investments across properties including La Liga, Ligue 1, the WTA, Volleyball World, Six Nations, Premiership Rugby, and United Rugby Championship.³ Arctos Partners has raised dedicated institutional capital for sports franchise stakes, and Sixth Street has become an active sports investor across franchises, clubs, and structured partnerships.⁴ The pattern is systemic.

The numbers around football make the shift harder to dismiss. PitchBook reported in late 2025 that European sports private-equity activity had already exceeded EUR 10.6 billion for the year, roughly triple 2024's full-year level, and separate PitchBook analysis put private-equity, venture-capital, or private-debt participation at roughly 36.5% of clubs in Europe's Big Five leagues.⁵ UEFA has tracked hundreds of clubs worldwide in multi-club investment structures.⁶ UEFA's 70% squad-cost rule is in full effect, and the Premier League's squad-cost rule comes into effect from 2026/27.⁷ Every transfer is becoming a constrained optimization problem under explicit financial rules.

The women's-sports signal confirms that institutionalization is not confined to men's football: Morgan Lewis has described new investor opportunities in the sector, and McKinsey projects the US women's sports market could reach at least $2.5 billion by 2030.⁸

The shape should be recognizable to anyone who has watched capital enter a domain before. When real estate institutionalized through REITs in the 1990s, the demand surface produced Yardi and CoStar — the operational and data-platform companies that emerged to serve institutionalized real estate. When defense procurement institutionalized after 2001, it produced Palantir and later Anduril — the defense software companies that emerged to serve institutionalized post-9/11 procurement. When legal procurement met current-generation AI, it produced Harvey — the legal-AI company co-founded by a former Sullivan & Cromwell associate and a DeepMind researcher, used inside firms including A&O Shearman, Ashurst, and PwC — which announced an $11 billion valuation in March 2026 after press reports put ARR at $190 million by the end of 2025 and later annualized revenue above $200 million.⁹ The pattern holds: institutional capital arrives in a domain, demands the operational software it cannot operate without, and the companies that build that software become very valuable.

The pattern is not universal, and it is worth being precise about why. Maritime shipping, mining, and private credit each received institutional capital without producing a single consolidated operational-software layer in the same way. The pattern requires three conditions: genuinely absent operational software, regulatory complexity that rewards data layering, and durable contractual cash flows. Sport now has all three. The cash flows do not come from transfer fees, which are speculative and one-off; they come from the long-duration structure of media rights, league distributions, sponsorship, and matchday revenue — the revenue base that gives a club its institutional weight in the first place. The regulatory complexity is tightening, not loosening. And the operational software is, for the most consequential decisions, missing.

Apollo's own white paper, published in December 2025, calls sport a $2.5 trillion global ecosystem and emphasizes media-rights cash flows, under-levered franchises, early institutional development, and opportunity across the capital stack.¹⁰ Apollo is not the protagonist. It is the vocabulary lesson. The relevant buyer does not talk primarily about dashboards, scouting vibes, or startup TAM. It talks about underwriting, origination, cost of capital, capital-stack flexibility, cash-flow durability, and platform capability.

The capital is arriving. The decision layer it requires — the layer that turns a number on a screen into a defensible recommendation — has not arrived with it.

Who Builds Decision Layers Now

Before turning to what is missing in football specifically, it is worth naming what has changed about who builds decision layers at all. Without that change, the rest of the argument does not hold; with it, the football-capital gap becomes a category of opportunity, not a one-off.

Three years ago, a category like sporting-decision diligence for institutional capital would have been a B2B software problem of the familiar shape: a credentialed founder, a sales-led organization, an enterprise contract negotiated bespokely with each buyer — the football-data incumbents do not publish their pricing¹¹ — an eighteen-month implementation cycle, a buyer relationship mediated by procurement and legal. The economics ruled out anyone who had not already built a previous category. The pace ruled out anyone outside the existing software commercial machine.

That shape has come apart. The cost of producing institution-grade software has collapsed for teams that pair domain judgment with current-generation AI agents. A small group with the right operator at the front can now produce, in weeks, what previously required quarters and a sales force. Categories that used to support a single bespokely-priced incumbent are now being entered by AI-native teams selling at four to five figures, with deeper specificity and faster iteration. Harvey is the most legible example in legal services; the same dynamic is visible across compliance, healthcare administration, regulated procurement, vertical accounting, and adjacent professional verticals. The pattern is not a curiosity. It is the live market dynamic of 2026.

What makes this work is not the AI in the abstract. It is a particular kind of operator. The teams that win these verticals are led, almost without exception, by people who felt the problem first — practitioners who carry tacit knowledge about how the work is actually done, what the current tools get wrong, and which judgments inside the work are load-bearing. AI agents systematize that judgment into product, methodology, and artifact at speed. Without the domain operator, agents produce plausible output that misses what matters. Without agents, the domain operator produces correct judgment slowly. Together, they produce institution-grade artifacts at a pace incumbents cannot match.

A latency in the system makes the dynamic durable for some period. Incumbents in these verticals are slow to adopt current-generation AI in their working practice — not because the technology is unavailable, but because adopting it requires accepting that the frame within which their organization was built may no longer hold. That hesitation is human, not technical. It is also expensive. While they hesitate, the AI-native operator-led team produces the artifact, ships it, learns from buyers, and iterates again.

The window opens, in practice, where three conditions overlap: a vertical is institutionalizing fast enough to demand decision infrastructure; the incumbents inside it are too slow to rebuild from definitions up; and a domain operator with the relevant judgment is willing to take the rebuild on. Legal and compliance had those conditions in 2023 and 2024; their leaders are now largely visible. Sport has them now.

Sport is, in fact, especially favorable. The asset transfer happens at a visible decision point — buying a player for tens of millions of pounds is a capital allocation about the asset itself, not about an abstracted financial instrument. The buyer can evaluate the methodology directly. The regulatory layer is tightening rather than loosening. The data layer exists and is sophisticated. The decision layer is where the gap has opened. And the incumbents inside the vertical have, with limited exceptions, built around enhancing existing metric definitions rather than rebuilding the primitives.

The argument of this essay is that an operator-led institution can attach itself to the sport-capital influx with enough force to win a meaningful slice of the new buying — and that the window is open now because the incumbent layer has not yet rebuilt.

Data Access Is Solved; Decision Integrity Is Not

Football's data layer is more sophisticated than it has ever been. StatsBomb covers clubs and federations with granular event data. SkillCorner builds physical tracking models from broadcast footage and has raised significant capital. Hudl/Wyscout sells scouting video and data at global scale across professional football.¹² Front offices have access to more structured football information than at any point in the sport's history.

The decision layer is emerging. Well-capitalized, credentialed teams have entered with serious acquisition strategies and significant analytical depth. The gap between what those entrants currently provide and what institutional capital actually requires remains large. Stating the gap precisely takes a distinction that is easy to write down and harder to internalize.

Current solutions provide decision support — better inputs for human judgment. A dashboard, data feed, scouting video layer, model score, and search tool can all improve the information available to a decision-maker. That is meaningful. It is not the same as decision infrastructure.

Decision support improves the inputs to judgment. Decision infrastructure makes the reasoning process itself repeatable, auditable, and defensible.

Consider the buyer. When a private equity firm acquires a football club for hundreds of millions of pounds and needs sporting due diligence, the analyst preparing the investment-committee memo does not need a better dashboard. They need a sporting-diligence memo that can be attached to an investment memo: visible methodology, comparable cases, stated uncertainty, and a recommendation the institution can defend. Decision support gives that analyst better inputs. Decision infrastructure gives the institution a defensible process. The first reduces individual error; the second creates institutional accountability. The price point, the buyer profile, the deliverable format, and the contractual relationship around each are different.

This is not an anti-consulting argument. Consulting can be the beachhead because trust often begins with bespoke work. The difference is what compounds afterward. A consulting engagement compounds into relationship, reputation, and case experience. Decision infrastructure compounds into method, audit trail, product surface, and institutional memory.

The reason this distinction matters in practice — not just in definition — is that the buyer's failure mode is different in each case. A scout using decision support and getting it wrong has a bad day. An investment committee approving a transfer based on decision infrastructure and getting it wrong has a paper trail. The paper trail is what allows the institution to learn, apportion responsibility, and defend its decisions to its own capital partners. Buyers who can absorb individual error differently from institutional error will pay differently for the two products.

Sport has a strange skin-in-the-game problem. The asset is a human performer inside a role, a tactical context, a development curve, and a market. The allocator cannot directly inspect the asset by becoming it. Even ex-player expertise does not automatically explain whether another player will translate across role, league, team model, coach, injury history, and price. That makes the observer's method unusually important. The buyer is not only buying a view on whether a player is good. The buyer is buying a reason to trust the chain that produced the view.

What does decision infrastructure look like in practice? The clearest demonstration in football is not in a dashboard. It is in the audit of a primitive that practitioners have been using without examining. Here is one such audit, presented as a worked example of the distinction.

Set-pieces are football's closest thing to a playbook: bounded, rehearsed, role-assigned moments inside a fluid game. Corners, throw-ins, long balls, second balls, height, and duel competence are not decorative details; they reveal a squad's hidden beliefs about who is trusted to head, mark, recover, strike, protect the counter, and execute under rehearsed constraint. As set-pieces and margins became more important to how teams create advantages, the measurement primitive underneath aerial evaluation became consequential.

The standard aerial-duel definition records a contest only when both players leave the ground. In STATSWING's single-match study, that definition excluded approximately 80% of contested aerial situations.¹³ The metric does not measure what many practitioners think it measures. It misses exactly the wrestling, pinning, first-contact, and second-ball situations that set-piece and long-ball football depend on. Decision support built on the flawed definition inherits the flaw without surfacing it. Decision infrastructure that examines the definition — that asks whether the metric measures the football question — is a different product entirely.

The pattern is not limited to aerials. Aerials audit the event definition. Possession adjustment and its reanalysis audit the normalization. Mechanics audits the inference from action output to transferable skill. The work repeatedly asks what is collected, what is normalized, what is inferred, and whether the inference carries the warrant practitioners assign to it.

Notice what this implies for intelligence platforms assembled through acquisition. A platform that buys multiple analytics companies inherits the metric definitions of each acquisition. It assembles intelligence from parts that may not share foundational assumptions. Platforms assembled through acquisition can become powerful. They also inherit the assumptions of the products they acquire. If the foundational definitions are wrong, integration can scale the error faster than it corrects it.

Dr. Ian Graham, former Director of Research at Liverpool FC, observed at a StatsBomb conference that more than 50% of "big transfers" — those at GBP 10 million or above — fail by a minutes-played criterion.¹⁴

Football has solved data access. It has not solved decision integrity.

Why the Window Is Open

The interesting question is not whether AI changes things; everyone already believes that. The interesting question is what specifically changed, what gap opened, and why sport is the arena where the gap has become most visible.

Begin with the operator. AI agents, given a clear specification, are competent. They retrieve, structure, compare, analyze, and write. The remaining bottleneck — which used to be execution time and is now nothing of the kind — has migrated up the stack. It sits in the operator's abstraction: the willingness and capacity to hold complexity over multi-hour stretches, the focus required when the agent is offering tempting work that is not load-bearing, the patience for the dialectic that refines the spec. None of these are technical capacities. All of them are human.

Ethan Mollick, the Wharton professor whose research has become one of the more widely read sources on AI's workplace effects, quantified one version of this with BCG in 2023: GPT-4 users completed 12.2% more tasks, 25.1% faster, at 40% higher quality than colleagues without access — but the capability was uneven across tasks.¹⁵ Mollick called it the jagged frontier. The operator's judgment about where to deploy the tool became the remaining scarce resource.

There is a diffusion lag underneath this. Anthropic's Economic Index, published in March 2026, shows that even the most exposed occupational category — Computer & Math (a broad knowledge-work category in the US occupational taxonomy) — exhibits only about a third of its theoretical ceiling in observed usage on Anthropic's platform.¹⁶ The March 2026 learning-curves update adds a complication: high-tenure users show persistent usage differences, and geographic convergence appears slower than earlier diffusion curves might have suggested.¹⁷ The right-tail compounding interpretation is plausible, but it is an inference from one provider's platform data, not independent labor-market fact.

The deeper source of the arbitrage, however, is not technology. It is human inertia. The gap exists because most organizations — and most individuals inside them — resist the possibility that the entire frame within which they have been working may need to be rebuilt from the definitions up. Not tweaked. Not enhanced with a copilot. Rebuilt. AI is the mechanism that makes the right-tail operator's advantage possible. The source is the difficulty of accepting that one's prior model of how work should be done may no longer hold.

Jack Dorsey and Roelof Botha articulated a version of this in Block's From hierarchy to intelligence in March 2026. Their thesis: organizational hierarchy is an information-routing protocol built around human span-of-control limitations, and AI can replace the routing function itself.¹⁸ In sport, the version is precise. Most analytics groups in football are using AI to enhance existing metric definitions and existing recruitment workflows. The arbitrage belongs to whoever is willing to rebuild the definitions.

The throughput of an institution is a function of how epistemically sound its primitives are. Institutions built on AI propagate epistemic unsoundness at speed. Rebuilding from sound foundations is not optional; it is the precondition for the intelligence layer to be trustworthy. The aerial example matters for this reason. Once the market starts caring about set-pieces, height, duels, and second balls, an inherited aerial-duel primitive stops being a harmless metric quirk. It becomes a capital-allocation risk.

Ajibola Lawal called one configuration of this earlier in 2026 the Young King: the operator who enters a vertical they were not formally credentialed for and competes because they have judgment, an LLM, and the willingness to be told they are wrong.¹⁹ The window stays open as long as the median operator inside the vertical has not adopted the same posture.

The arbitrage is not "we have AI and they don't." It is not "we work faster." It is the fit between domain judgment, production speed, primitive quality, and an institutionalizing market that has not rebuilt its decision layer. The window opens because the specificity of the problem — auditable decision infrastructure for institutional buyers of sporting assets — sits adjacent to the integration overhead of acquisition-assembled intelligence stacks, and the operating model fits the specificity unusually well. The resistance to rebuilding from foundations is the window. AI is the mechanism.

STATSWING as First Field Site

If the argument is right, somewhere a first attempt at building this institution would already be underway: an operator-led team publishing primitive audits, building decision-infrastructure artifacts in the open, and testing whether the buyer category exists. I have made such an attempt. STATSWING is where this thesis has first been tested.

STATSWING is a sports intelligence institution — research-first, product-led — built to serve the decision-infrastructure gap. In early 2026, the first public version went live. The launch-state surface covered roughly 14,800 players across 21 leagues by our internal count, and the public product has since crossed 100,000 organic visits by STATSWING's public/operator-reported figures. The research published around the first surface covered contested aerial measurement, epistemic certainty in recruitment, and the assumptions buried inside possession-adjusted statistics. Underneath both, an internal research-orchestration substrate connects proprietary research, agent-assisted analysis, and the public surface.

What the operator brought to the agents was not technical sophistication. It was the kind of judgment that asks whether the standard aerial-duel definition is doing what practitioners think it does, and that holds the question long enough to design a study around it.²⁰ That instinct is the load-bearing input. The agents did not invent it. They could not have. What they did — and what most organizations have not yet absorbed in their working practice — is collapse the time between having a clear sense of what something should be and the artifact existing in the world. A research note that previously took two months to draft, review, defend, and publish now takes ten days. A grade system that previously required a team of analysts can be specified, tested, audited, and revised by one operator working in tight loop with agents. The compression is not in the thinking. It is in the distance between the thought and the public artifact.

That collapse is the arbitrage. AI did not supply the judgment; it compressed the distance between judgment and an institution-shaped artifact. An operator with domain knowledge, training in evaluating the trustworthiness of claims, and meaningful proficiency with current-generation AI agents can now produce public, testable, institutional-grade work at a speed and scope the market has not yet seen from a single team, let alone a single operator.

That is as far as the evidence currently goes. The stronger claim — that this is a transferable institutional pattern, not a one-off football project — has to be earned through buyer-grade artifacts, inspected methodology, decision memory, and recurring institutional work. If those do not materialize, the claim should shrink: STATSWING may be an unusually serious football research project, not evidence of a generalizable institution-building pattern.

What Would Prove This Wrong

The thesis can fail in four ways.

First, the sample may stay at one. STATSWING may show a public instance of primitive audit, AI-enabled production, and decision-infrastructure framing in football without proving that the operator-led model generalizes across buyers, sport, or verticals.

Second, the best-capitalized buyers may build the layer in-house. Apollo-like firms may treat specialized sports underwriting as a core platform capability, not a function to buy from outside specialists.¹⁰ The external-vendor case then depends on speed, independence, domain specificity, published methodology, and auditability.

Third, the diffusion lag may close faster than expected. The March 2026 Anthropic update supports caution more than certainty: it shows tenure-linked differences and slower geographic convergence, but it does not prove that early adopters will keep pulling away across all knowledge work.¹⁷

Fourth, a well-capitalized rollup may already have assembled the missing layer. The relevant question is whether breadth has produced coherent method. If a rollup publishes compatible definitions, traceable recommendations, and uncertainty practices at scale, the native-coherence advantage narrows quickly.

None of those objections kills the thesis today. Each can kill it later. They define what to watch, what to build, and what would count as being wrong.

What To Watch Next

The first market test is purchasing behavior. By Q4 2027, at least one club, multi-club ownership group, PE-backed sports asset, or sports investor should hold a recurring external contract for structured, auditable sporting-decision diligence, distinct from data feeds, scouting video, dashboards, or ordinary advisory work. The thesis strengthens if that purchasing category becomes visible. It weakens or time-shifts if diligence remains bespoke, hidden, or bundled inside familiar consulting and data relationships.

The second test is whether sporting judgment attaches to capital process. By the end of 2028, at least one sports acquisition, governance process, or major capital allocation should publicly reference external sporting-decision diligence, methodology, or an independent intelligence provider. The constraint is visibility: this work may already happen privately around multi-club ownership and sports-capital underwriting without becoming part of the public institutional apparatus.

The third test is provenance demand. By the end of 2028, at least one major sports analytics or decision-intelligence provider should publish a methodology audit trail: versioned definitions, traceable recommendation chains, uncertainty disclosure, or comparable governance artifacts. If audit trails become a buying criterion, the category is forming. If buyers reward outcomes without asking how the reasoning chain was produced, provenance may be philosophically right and commercially premature.

STATSWING has its own burden. It should produce a transfer or sporting-diligence artifact that an institutional reader would plausibly attach to an investment memo. Insight is not enough. The artifact has to carry methodology, comparable cases, uncertainty, and recommendation in a form legible to a sporting director, investment team, owner, or board.

It also has to become visibly inspectable. An institution arguing for auditability has to publish a methodology trail for its own research primitives before the market demands one by default: definitions, revisions, uncertainty, retrodiction tests, and known limitations. It should build decision logs so past judgments can be revisited and scored against the assumptions that produced them. Without memory, decision infrastructure becomes one-off persuasion.

The commercial test is whether the method survives revenue. Consulting, research, diligence, or bespoke work may be the beachhead, but the durable asset is method plus audit trail. The thesis strengthens if STATSWING becomes one of the first external decision-infrastructure vendors with a recurring institutional contract in sport. It weakens if commercial work requires abandoning the decision-infrastructure thesis and dissolving into generic advisory labor.

The broader test is whether the operating model travels. By April 2029, at least three operator-led institutions should exist in newly institutionalizing verticals — legal diligence, regulated procurement, public-sector audit, healthcare decision support, or similar categories — with paying institutional clients, published methodology, and recurring revenue. If this remains idiosyncratic to one operator and one football market, the claim narrows.

The watchlist is equally important. Apollo-like investors may build sports decision infrastructure internally, acquire it, partner with specialists, or outsource narrow diligence. Well-capitalized rollups may close the primitive-audit gap if they publish coherent definitions, traceable recommendations, and uncertainty practices at scale. AI diffusion may also close the window if usage rises symmetrically and median operators catch up quickly. Regulation may strengthen the thesis if squad-cost rules, multi-club governance, LP scrutiny, and acquisition diligence make sporting method explicitly inspectable; it weakens the thesis if buyers satisfy oversight through lawyers, bankers, and existing advisors without changing the decision layer.

Provenance and Judgment

The deepest version of the argument is that the buyer is not paying for the recommendation alone. The buyer is paying for the ability to evaluate why the recommendation was made, by what method, on what evidence, with what acknowledged limitations.

In an era where producing words is trivial, what becomes scarce is the audit trail behind them. The institutions of the previous information economy — journals, diligence reports, investigative units, underwriters, club analytics teams — were structures for managing the scarcity of information and the cost of producing it. The institutions of the next economy will manage something different: the trustworthiness of testimony.^{21 22} The forensic evidence is already accumulating. A January 2026 audit examining 5,514 citations across fifty AI-era survey papers found that for nearly one in five citations, the digital chain of custody was severed: the referenced paper could not be verified through the citation chain.²³ Other domains are responding. The C2PA standard for media provenance is the most visible attempt at literal infrastructure for chain-of-custody on documents and images.²⁴

A company that produces structured, auditable reasoning for a major player acquisition decision — one that shows its evidence, assumptions, metric definitions, comparable cases, uncertainty, and revisions — is selling an audit trail for testimony about a player's value. Sport is the first arena where such an institution can be built and tested in public. It is not the last.

When production is commodified, what remains is judgment. The institutions that certify judgment — that show their work, version their definitions, and stand behind their recommendations — are the ones that will matter. The window is open now.

For correspondence: joel@statswing.com.

— Joel A. Adejola, Lawrence, Kansas, April 2026.

Notes