Jack Williams, iTero and GIANTX: As AI Walks Into the Analytics Room, Esports Still Hasn't Finished Writing the Rules

In the summer of 2026, I sat in a rented apartment in Busan, rewinding the tape of Italy vs Belgium at the Euros. In the 79th minute, Leonardo Spinazzola went down. His Achilles tendon had ruptured. Roberto Mancini was forced to switch to a back three for the rest of the tournament — and that unplanned shift helped carry Italy to the trophy at Wembley. I wrote at the time: "Spinazzola left the Euros on a stretcher but runs forever in memory — injury sometimes echoes louder than a trophy."

Four years later, I read an interview about Jack Williams, about iTero, about GIANTX, and realised I had overlooked something all along. A player's injury leaves measurable traces: ultrasound scans, days out, matches missed, measurable drops in sprint metrics. But the software a coaching staff uses to prepare for the next match — the thing that sits between the coach and the player, between data and decision — has no clinic to diagnose it. Nobody audits it. Nobody writes rules for it.

The real subject of this story lives there. Patches pass. Meta shifts. The preparation apparatus stays.

Context: one interview, three names, and a legal vacuum

Jack Williams is the person answering questions. iTero is the AI-driven coaching tool attached to him. GIANTX is the EMEA-based esports organisation working with iTero under an exclusivity arrangement. The interview is sectioned clearly, and the two most notable sections deal with working with GIANTX exclusively and the likelihood of being copied, alongside a section on AI-assisted cheating.

Skimmed, it looks like a product introduction. Read closely, it is an undiagnosed case file.

Let me be blunt from the start: the publicly available substance of this exchange is far thinner than people would hope. No patch is named. No game version. No standings, no win rates, no pick-ban metrics. No rosters, no specific players. No brackets, no format, no schedule. If someone set out to write a pure tactical analysis from this material, they would have to invent it. I refuse to do that.

The only real signal in hand is structural: an esports organisation signing an exclusivity deal with an AI tooling vendor, the head of that tool worrying about being copied, and an industry arguing over where the line on cheating falls once AI enters the frame.

Those three pieces, assembled, form a far larger question than "will AI replace coaches." The real question is this: when a preparation tool becomes a proprietary asset, who exactly is playing fair against whom?

I have tracked esports and football in parallel for years — as a competitor, a tournament organiser, and a media professional. That experience taught me one thing: every advantage in sport was once a legal argument before it became part of the rulebook. Carbon-plated running shoes once triggered a dispute. Full-body swimsuits were once banned. Heart-rate analytics rigs were once the exclusive property of a handful of national laboratories. Esports is walking the same road, only ten times faster.

The real boundary sits in the between-game window

Most major titles have banned real-time in-game assistance in explicit terms. Nobody may read live data off the tournament server to inform decisions while a match is running. That zone is clean, at least on paper.

But the gap between Game 1 and Game 2 of a BO3, between Game 3 and Game 4 of a BO5, is a grey zone. There, the coaching staff gets minutes. In those minutes they must answer life-or-death questions: which weakness is the opponent avoiding, should the draft pivot, who is actually controlling the tempo.

An AI tool designed to answer exactly those questions, in exactly that window, breaks no clause that has been written — because the clause does not exist.

The between-game window is the real battlefield in the AI-in-esports debate, and it has no gatekeeper.

This is why I distrust the popular framing. People fear AI intervening inside matches. But AI does not need to intervene inside matches to generate meaningful advantage. It only needs to read the right historical data, synthesise it correctly during the break, and present it correctly to the decision-maker. The gap between a good coach and a good coach with AI is the gap between judgement built on memory and judgement built on an entire modelled dataset.

Put another way: this is a cognitive optimisation problem, not a cheating problem.

Patch cadence determines the value of every AI model

Here I must lean on background knowledge, and I will say so openly.

Two major titles run on completely different update cadences. One is Dota 2, where large patches are infrequent and systemically disruptive, with long stable stretches between them. The other is League of Legends, where the patch loop runs densely all year.

That cadence determines the value of a machine-learning model.

In a slow-patch title, historical data retains usable value for a long window. A model trained on thousands of old matches is still valid months later. The advantage lies in depth of historical modelling.

In a fast-patch title, the half-life of any learned pattern is sharply shortened. The value of an AI tool shifts from "solving the meta" to "detecting the meta delta faster than the opponent." That is a speed advantage, not a knowledge advantage.

A single AI product marketed identically across both types of title deserves scrutiny, because its core value proposition inverts entirely between the two patch environments.

I tested this against a familiar comparison. In the 100m sprint, coaches analyse stride length and step frequency. Those numbers vary by individual, but the physiology underneath is stable across decades. That is a slow-patch environment. By contrast, in a combat sport with constantly rewritten competition rules, people stop analysing stride length — they analyse which stride length the new rule just made advantageous.

In the public materials about iTero, I found no data on sample size, evaluation methodology, or independent validation for any performance claim. That renders every statement about the tool's effectiveness unverifiable.

Exclusivity in a closed league is an advantage that cannot be relegated

This is the part I consider most important, and the most overlooked.

The league GIANTX competes in runs on a closed franchise model. Put simply: member teams are permanent members, with no relegation threat hanging over them.

In an open system, structural advantage gets eroded by competition. Weaker teams either copy it or get removed from the league, and the reward goes to whoever adapts fastest. In a closed system, structural advantage persists across seasons, compounds over time, and no mechanism automatically takes it away.

An exclusive analytics-tooling deal, placed inside a closed system, is exactly that kind of advantage.

In a closed franchise league, an exclusive tooling deal is not relegated with the season — it compounds. That is why this story belongs to the governance frame, not the commercial frame.

What is striking is that the interview appears to mine only two frames. The first is exclusivity and the risk of being copied — a commercial frame. The second is AI-assisted cheating — an integrity frame. The third, league fairness, sits precisely between them and is touched by nobody.

As a tournament operator, I have seen this before. Sponsors and technology partners always prefer to talk about the first two frames, because they sound compelling and pose no hard questions to organisers. The third frame forces organisers to admit that the playing field they built is not level.

On intellectual property: the moat is data, not algorithms

Jack Williams talks about being copied. This is the part I want to dissect through a sports-business lens.

A software tool carries three layers of value. The first is the algorithm. The second is the data. The third is the exclusive distribution relationship.

The algorithm layer has the weakest moat. Current language-model and predictive-model architectures are widely published. A competent engineering team, a few months, and a decent dataset can rebuild most core features of any sports analytics tool.

The data layer has a medium moat. Professional-grade match data is purchasable, but labelled data — data annotated by humans, with situations classified and tactical intent tagged — is expensive and slow to build. This is where real advantage forms.

The exclusive relationship layer has the strongest moat. An exclusivity contract with an organisation inside a closed league cannot be copied technically. It can only be broken legally.

I have written about something similar in football. Shirt sponsorship has gradually eroded the bond between clubs and local communities, because global sponsors care only about brand-exposure metrics. I see a similar dynamic here. When an esports organisation's value is tightly bound to an exclusive tooling vendor, that organisation's story stops being written in match results and starts being written in contract clauses.

Fans do not read contract clauses.

A lesson from the track: wasted running still produces pretty numbers

I graduated in exercise science, and I carry an occupational obsession: effort metrics.

Distance covered. Sprint count. Pressing actions. These numbers get packaged and sold as proof of dedication. They appear in broadcast graphics, get read aloud in solemn tones, and are always missing half the story.

There is one match from 2026 I will never forget. One team held the ball for most of the game and lost. The other defended deep and won. That night I sat down and argued that worshipping possession metrics was outdated, and I used one Asian player's sprint count to argue that speed-based counter-attacking was the evolutionary model. The post got a few hundred views. The first person to share it was my professor, who then made the whole class rewatch the tape to debate it.

That lesson applies directly here.

An AI tool can generate numbers that sound highly persuasive. It can tell you a player moved ten percent more than his opponent. It cannot tell you whether that ten percent meant map control or simply running in pointless circles.

Effort metrics are always easier to measure than effectiveness metrics. An AI tool sells better when it measures what is easy to measure, and that is the single biggest cognitive risk it brings to a team.

On the track, I learned that people endure pain for their own limits, not for medals. In an analytics room, the same holds in a different way: people believe numbers because numbers give them a sense of certainty, not because the numbers are right.

The counter-intuitive angle: AI makes coaches more replaceable, not better

This is where I want to break the familiar framing.

Everyone asks whether AI makes coaches better. I think that question points the wrong way.

When a tool becomes ubiquitous, it does not lift the floor evenly. It compresses the gap between those in the middle of the table, and simultaneously lowers the threshold required to do the job at an acceptable level. Meaning: the market value of an average coach declines, while the value of a genuinely exceptional coach — one who asks questions the tool was never programmed to answer — rises.

This is an effect I have observed across industries. When drawing tablets got cheap, the number of competently adequate illustrators exploded. The number of genuinely exceptional artists did not.

In esports, the consequence is sharper stratification of coaching staffs. There will be a small group of analysts who can read data at the level of first principles, and a large group of coaches who operate through a tool's interface.

The biggest blind spot in this entire debate lies elsewhere. People argue over whether AI may be used. They do not argue over who gets to use it.

A free tool and a proprietary tool create two entirely different worlds, even when they run the same algorithm. The cheating debate asks questions about the conduct of players and coaches. The exclusivity debate asks questions about the conduct of organisers.

And organisers, in my experience, rarely enjoy answering that kind of question.

What I take away

I once stood in a stadium and learned a trade: listening to the noise to know when to stay silent. The empty stadiums of 2026 taught me that football does not lack an audience — the audience lacks football. Esports is now in a moment much like that.

It does not lack tools. It lacks rules for tools.

If every team ends up with the same AI model, predictive quality flattens out, and competitive advantage returns to where it has always belonged: to the human behind the screen, the one who knows which question to ask when every answer is already sitting inside the machine.

Do not ask who controls the match. Ask who is writing the rules for the software the match-controller uses to control it.