AI Coaching in Esports: The GIANTX-iTero Exclusive Deal and the Rulebook Gap Nobody Wants to Fill

An exclusive contract. An AI tool. A closed league with no relegation. Three pieces scattered across an interview with Jack Williams about iTero and GIANTX, and none of them called by its proper name.

I counted the payload before writing a single word. Thirteen information points. Ten of them describe the person who wrote the piece: Ollie, recalling Natus Vincere lifting the Aegis of Champions at Gamescom "fourteen years ago." Na'Vi won The International 2026 at Gamescom. Add fourteen years, and the interview sits somewhere around 2026.

The remaining three points touch what matters: Jack Williams, iTero, GIANTX. Two section headings are disclosed. One covers working exclusively with GIANTX and the likelihood of being copied. One covers AI-assisted cheating.

No patch. No scoreboard. No player names. Not a single contract term expressed as a number. What is left to analyse is structure, and in this case structure is more alarming than any number.

Context: esports learned its own rulebook late

I entered esports in 2026, competing and working behind the scenes running tournaments before moving into writing. The job taught me something audiences never see: the rules do not live on the stage. They live in a PDF.

In League of Legends, coaches were once barred entirely from the competitive area, then permitted, then restricted by communication windows. Every loosening and tightening reopened the same question: where is the line between preparation and competition. Dota 2 took a different road, leaning on Valve's rules and community culture rather than a granular code. Two governance philosophies, one shared gap.

GIANTX is the name merged from Excel Esports and Giants Gaming, formed in 2026, competing inside Riot Games' LEC ecosystem in EMEA. The LEC has run as a closed franchise since 2026. No relegation. Ten organisations, season after season, the same names.

iTero is the AI tool the interview references. Jack Williams sits between those two ends. And what is being sold in that conversation is not a feature. It is a seat in the room.

One scope note matters: the "AI coaching" debate in esports almost certainly concerns pre-match, between-game and post-match analytics. Real-time in-game assistance is already unambiguously prohibited in every major title, leaving nothing to debate. The genuine grey zone is the between-game window of a BO3 or BO5 — five to ten minutes that the existing rules were never written to describe.

The star system, esports edition

I have written before that in football the star is only the tip of the iceberg. In esports the iceberg is more visible because the org chart is shorter and the median age of decision-makers is lower. A modern LEC team carries a head coach, a performance coach, a data specialist, a psychologist, a team manager, and now a tool vendor.

When an organisation signs an exclusive tooling deal, it is not buying software. It is buying a chair in the meeting. And that chair changes how decisions get made. An argument presented with model output outweighs an argument built on instinct. After two seasons, the person who speaks without a model behind them starts to shrink. That is a power mechanism, not a technical one.

In traditional sport, expected-goals models took roughly a decade to move from curiosity to orthodoxy, and when they arrived, the analyst got a seat at the table. Esports will move faster, because the data pipeline was digital from day one rather than hand-notated as football's was in the 2000s.

There is a paradox here. More data means more error, not less. Esports generates data at every level, including unofficial games. A model trained on that learns the noise. And the largest source of noise is scrim behaviour, where teams deliberately play badly to hide their hand.

Patch cadence inverts the value of AI across titles

This is the part of the interview that does not exist, and it decides what iTero is actually worth.

Valve ships Dota 2 updates rarely and heavily. Long stable stretches sit between large patches. Models trained on historical match data hold their accuracy for longer windows. In that title, depth of modelling wins.

Riot ships League of Legends on a rapid cadence, typically every two weeks. The half-life of any learned pattern is compressed. There, the tool's value shifts from "solve the meta" to "detect the meta delta several days before your opponent." That is a tempo advantage, not a knowledge advantage.

The practical consequence: a product marketed with an identical pitch across both titles is a red flag. If the seller does not distinguish, the buyer should, before signing.

In a closed league, exclusivity never gets competed away

In an open system, a structural advantage decays because weak teams are removed and knowledge disperses. In the LEC, nobody is removed. Ten organisations stay. A small edge compounding across seasons becomes a gap that effort alone cannot close.

Picture a mid-table European club with a superior analytics department. It can still be relegated, and on relegation that department sells itself elsewhere. The capability diffuses. The LEC has no such release valve.

The new meta lives where people are afraid of losing something, not in the tactics. What the other nine LEC organisations fear losing is their international qualification slot. That fear, not the patch, is what pushes them to sign tooling deals.

The moat is not the algorithm

The source interview contains a section on the likelihood of being copied. The honest answer: the model is copied in a week. The model itself is not the asset.

The asset is three other things. First, deep data access — scrim data, practice data, operational data that no public scoreboard carries. Second, integration depth: a tool already embedded in a team's decision loop is more expensive to replace than to keep. Third, the legal shell of the exclusivity clause, which protects the seller from imitation and simultaneously becomes the wall the buyer hits when they want out.

There is a point outsiders miss. The moment a vendor sells to a second organisation, the exclusivity story dies. So most of the value of the exclusive phase sits in the narrative told to the next buyer, not in the league table.

The grey zone is between games, not inside them

Major publishers prohibit outside assistance during a live game. That part is uncontested. The contested part is the four to ten minutes between games.

Make it concrete. A team loses game one. During the break, the coach opens the tool. The tool proposes three ban options based on the opponent's last seventeen games. The team wins game two on the second option. Is that updated pre-match preparation, or in-game coaching by proxy? No publisher answers that in writing.

This is the largest governance hole in esports for 2026 through 2027. And a hole that goes unfilled will be filled with money.

Hidden-data hunting: what never reaches the scoreboard

Football's transfer market has inverted its signal-to-noise ratio. Most published transfer news is planted by agents, not confirmed by clubs. The same mechanics run through esports vendor deals.

A press release about an exclusive deal is a sales asset. It is not a disclosure. It says exclusivity, vision, strategic partnership, and not one line about how much win rate the tool contributes.

I have tracked how leagues buy ageing stars as tourism ambassadors rather than competitors. The same logic applies here: plenty of esports technology deals exist to tell a story to investors, not to climb a table. That claim can be wrong. But it explains the situation better than the assumption that every organisation has already quantified a benefit it never states as a number.

Data models misprice the thing that decides a BO5

Valuation models for young talent consistently overweight production metrics and underweight dressing-room chemistry. In esports the same error wears different clothes: the model rates historical data very highly and the human variable very low.

Historical data knows the win rate on a specific composition. It does not know that the mid laner has lost three straight days of scrims and has started doubting the coach's voice. Game five of a final is not decided by a model. It is decided by who clicks first, and by who in the room is shaking.

I do not trust head-to-head history. I trust the way a team trembles in the 85th minute. In esports, the 85th minute is game five, minute thirty-five, when every resource is spent and one wrong decision is enough.

AI tooling will be excellent in draft, good in review, and close to useless at the psychological moment of decision. Buyers should ask vendors exactly one question: how do you measure that moment.

Where Vietnam and Southeast Asia stand

Based on my experience tracking matches in the region, the budget gap between a LEC team and a VCS team is a multiple, not a percentage. An enterprise-grade analytics licence sits outside the reach of most Vietnamese organisations.

That leads to two futures. One: the gap widens, because compounding advantage belongs to whoever can buy the tool, hire the person who can read it, and retain that person long enough to turn output into process. Two: the gap narrows, if free or educational tiers are genuinely good.

I lean toward the first, and not because of licence pricing. The binding constraint is people. A tool only answers well when the person asking knows how to ask. The analyst position is the first cut when a budget thins, because it does not score. Vietnam produces excellent individual players and exports them abroad. We export fewer analysts, because we have never sustained many people in that seat.

There is a sharper tactical risk. The signature style of Southeast Asian teams is chaos, high pressure, high variance. Models trained on historical data systematically underrate low-probability, high-variance situations because they appear rarely in the sample. If a regional team puts an AI tool at the centre of its process, the tool will pull it toward the safe, predictable European standard — and erase the only edge it has.

Exclusive tooling does not just sell software. It sells a playing style.

Where I could be wrong

First, this exclusive deal may be purely commercial and contribute nothing to win rate. If the tool is worth three thousandths of a win probability, everything I built above is a house on sand.

Second, the between-game grey zone may already be closed by existing rules, merely not yet written in AI language. Prohibitions on in-game coaching are broad and could cover tooling.

Third, and this weighs on me most: the interview's absence of any metric may simply mean there is nothing yet to state. That would mean AI coaching in esports in 2026 is still a pitch deck, not a validated product.

And an honest admission: I am reasoning from structure, not evidence. I have three data points from a thin interview. My dating of the source to roughly 2026 comes from arithmetic inside its own sentences, and could be off by a year.

Silence is never a win. It is extra time before the collapse. Publishers not yet writing rules does not mean rules will not exist. It means the problem is currently being solved by whoever pays first, and will be rewritten by whoever holds authority later.

A verifiable prediction

The grey zone gets filled within eighteen months. A major publisher will publish explicit rules on third-party AI tooling in official competition. If no such document exists by the end of 2027, the reason will not be that the industry failed to notice. It will be that the industry chose silence.

Second, a franchised league will be forced to choose between two paths: mandate equal access for all members, or ban exclusive tooling deals. If neither path is taken within three seasons, that will be evidence the league has quietly accepted preparation inequality.

In the first half people laugh at me, in the second half I laugh at the whole match. But there is no second half belonging to the writer here. It belongs to whichever organisation dares to ask its vendor the simplest question, ahead of everyone else: what percentage of win rate does your tool move, on what sample, across how many games.

If nobody can answer that, the only thing actually exclusive here is the silence.