AI Coaching in Esports: GIANTX's Exclusive Deal and the Grey Zone the Rulebook Has Not Defined

Three in the morning in Seoul. I rewind game four of an LEC best-of-five, not to watch the fight at minute 28, but to count the break between games: four minutes and forty seconds. Inside those four minutes, five players sit in a private room, one coach stands behind them, one monitor glows, and a pre-built pick-ban priority list sits on the desk. I rewound that segment eleven times over three weeks. Each time I logged a single line: which team changed its ban order, which team held, and which team left the break with a choice nobody in the arena saw coming.

That is how I read a match. When the numbers stop lying, my heart starts listening.

This week a single line of news changed how I keep notes. Jack Williams, the man behind iTero, confirmed in an interview that his tool is being used by GIANTX under an exclusive arrangement, and he said plainly that he expects the product to be copied. A software product. An EMEA esports organisation. One word: exclusive.

I care little about the software. The word "exclusive" is what made me reopen my notebook.

In a closed league, where ten members own permanent slots and nobody is relegated, every structural advantage lives longer than a single season. A tool reserved for one team never gets competed away through a promotion tournament. It simply sits there, season after season, quietly bending the equation. Switzerland did not beat France; it only skewed my equation. The same logic applies here. The question is not who wins one match, but who has been handed a variable their rivals cannot access.

I have covered esports since 2026, starting as a competitor, then a tournament organiser, then a media worker, and finally a betting analyst. Twelve years of watching this industry taught me one thing: the biggest changes never come from a patch. They come from infrastructure. A patch changes the draft. Infrastructure changes who gets to know which draft will work.

To place this story correctly, I need to rebuild a timeline.

In 2026, at Gamescom in Cologne, Natus Vincere beat EHOME 3-1 and lifted the first Aegis of Champions in The International's history. That roster included Danylo "Dendi" Ishutin and Clement "Puppey" Ivanov. Back then, "analysis" in esports meant a notebook and a captain's memory. No public database, no stats vendor, no model.

By 2026, top teams began hiring people to watch VODs. In 2026, when Riot Games converted the EU LCS into the franchised LEC with ten permanent partners, coaching staffs swelled: head coach, strategy coach, two or three analysts, a psychologist. Excel Esports joined the franchise system in 2026. Giants Gaming had been present in European competition since well before that. In September 2026, the two organisations announced a merger into GIANTX.

By 2026 and 2026, the next infrastructure layer arrived: machine-learning models trained on historical match data, generating ban recommendations, forecasting win probability by team composition, and optimising draft priority order. iTero sits in this layer. And GIANTX is its exclusive customer.

I want to be precise about the limits of what I know. The public record on this deal sits at headline level: one section covers the exclusive partnership with GIANTX and the likelihood of being copied, another covers AI-assisted cheating. There is no dataset, no sample size, no evaluation methodology, no contract terms. Every calculation below is inference from industry structure, and I label it as such.

The crux sits in patch cadence, and patch cadence varies so widely that the two biggest titles create two entirely different markets for the same product.

Riot Games patches League of Legends roughly every two weeks. That is about twenty-six patches a year. A major patch shifts a champion's win rate, opens a new draft direction, and closes an old one. Meta stability in League lasts roughly two to three patches, meaning four to six weeks.

Valve patches Dota 2 in large, infrequent strokes. Systemic patches can sit four to eight months apart. Patch 7.33, titled New Frontiers, launched in April 2026, enlarged the map substantially and rewrote the jungle's role. After a shock like that, the community takes months to stabilise, and throughout that window, models built on historical data keep their value.

The commercial consequence is obvious. For Dota 2, a machine-learning model has a long half-life. Its value lies in depth: you train on years of data, you surface correlations the human eye misses, and you exploit them for months.

For League of Legends, the half-life of a learned pattern shrinks to weeks. Model value shifts from "solving the meta" to "detecting the meta's delta faster than your opponent". That is a speed advantage, not a knowledge advantage. And a speed advantage in a league playing two matches a week is only worth something if it arrives before the match begins, meaning inside the between-game break or inside the forty-eight hours between fixtures.

One product marketed identically across both titles is a warning sign. Correct positioning must differ. If iTero sells one message to Dota 2 organisations and League organisations alike, my confidence in its efficacy claims drops sharply.

This leads to the central equation: the value of an exclusive coaching tool does not lie in the model's quality, but in how much time it removes between a meta shift and a team catching up.

Picture it in numbers. An LEC team employs two analysts. Each week they prepare for two matches, each requiring a draft plan across roughly thirty potential matchup pairs. In a stable patch, priority order shifts slowly and two analysts keep pace. In the first week of a major patch, priority order shifts daily and two people cannot keep up. That is the window where an automated model produces marginal value.

How wide is that window? With a two-week patch cycle, the disruption window spans roughly three to five days per cycle. That is around twenty-five percent of the season. If an exclusive tool buys half an advantage across twenty-five percent of matches, and that advantage converts into win rate, then across an eighteen-match LEC regular season the difference is one to two games. One to two games is the gap between fourth place and seventh place.

I have counted every empty space on the pitch when the crowds vanished. In 2026, when K League 1 resumed inside empty stadiums, I collected data from forty-two matches without spectators and found the home win rate falling from 42.3 percent to 29.8 percent, with draws rising to 31.5 percent. One environmental variable, removed, skewed the entire old model. The lesson I carried into esports: before debating a tool's quality, identify the environmental variable it acts upon.

Here, that variable is the interval between patches. The second variable is the publisher's data-access rules.

Riot Games has a history of tightly regulating third-party software. Its competitive rules restrict tools that intervene during live play. Coaches may speak to players before and after games and during designated breaks, but software delivering real-time assistance is explicitly barred. Valve takes a different approach, intervening less in the third-party tool ecosystem.

If the two publishers diverge in policy, a tool vendor faces two markets with different sizes, different legal risk, and different renewal rates. That is why I want to see iTero positioned by title, not merely by function.

There is one more region few people read closely: the break between games in a best-of-three or best-of-five. In-game assistance is banned. Pre-game assistance is permitted. The interval between games sits between those two zones, and I have yet to see a rulebook define decisively which zone it belongs to. The four minutes and forty seconds I counted sit precisely inside that gap. The most interesting grey zone in esports is not inside the match. It is in the four minutes between games.

That is why the interview's "AI-assisted cheating" section caught my attention. But I think that frame misses the centre of gravity.

AI-assisted cheating is a technical problem. Software runs on a tournament machine, gets caught by process inspection, and gets handled through device checks. It is a solved category, much like detecting third-party software at LAN events.

The unsolved problem is resource asymmetry inside a closed league. One member holds exclusive access to a tool, and no mechanism exists to erode that advantage, because there is no relegation. In an open circuit, weak teams drop out, strong teams promote, and structural advantages dilute over time. In a franchise system, structural advantages compound.

I am not saying this arrangement breaks the rules. I am saying the rules have not defined it, and that definitional gap will be filled by precedent, not by text.

In my world, luck is only the residual I have not yet explained. If GIANTX improves sharply next season, I will not call it luck, and I will not call it talent either. I will go looking for the environmental variable, and the first one I check is this exclusive deal.

This is where I must be blunt about the evidentiary ceiling.

No sample size has been published. No evaluation methodology has been published. No control group has been published. No claim has been made about win-rate improvement, ban accuracy, or draft-phase reaction time. In my daily work, a model without a control group is a model I cannot price. I can read it, I can understand it, but I cannot stake money on it.

There is a notable commercial paradox here. Jack Williams says he knows his product will be copied. If the product is easy to copy, exclusivity is merely a marketing claim, because any rival with enough engineers rebuilds it within months. If the product is hard to copy, exclusivity is a genuine competitive distortion. Those two propositions cannot both hold strongly. And a vendor has every incentive to claim both, because the first sells product and the second retains accounts.

I do not blame them. I simply note that incentive and evidence do not point the same direction.

So which metrics can actually be measured over the coming months?

What I track is draft-phase efficiency, not win rate. Win rate is contaminated by too many variables: individual form, schedule density, opponent quality, the variance of a single teamfight. Draft efficiency is cleaner. Specifically, I track three metrics.

The first is priority-ban accuracy: how often a team bans the champion its opponent intended to pick. This measures intent-reading directly, and it is precisely where a machine model outperforms humans once the data is dense enough.

The second is draft drift across games within a single series. If a team reshuffles its priority order in ways opponents cannot anticipate, drift rises. If the team repeats the same pattern, drift falls, and that is a signal the model is being reverse-engineered.

The third is the elapsed time between the end of one game and the team's first ban decision in the next. This is a metric I count myself; no data vendor sells it. If a tool genuinely operates during the break, either decision time must fall or decision quality must rise, or both. One of the two must move. If both sit still across an entire season, the tool produces no marginal value, whatever its name.

I do not believe in inspiration. I believe in standard error. A tool that publishes no standard error cannot yet be assessed.

I used this approach to read correctly a match the whole world read wrong. In 2026, at the World Cup in Qatar, Japan beat Germany 2-1. Korean media poured over the German coach's tactics. I opened the data table right after the match and found Japan had recorded 247 sprints against Germany's 201, and all five of their substitutions came before minute 74. Running intensity after minute 60 was the decisive variable, not reputation. My analysis that night reached 120,000 views.

I retell this to make one point: I am not opposed to new tools. I am opposed to evaluating new tools on faith rather than on indicators. The same principle applies to iTero. Give me the sample size, give me the control group, give me the confidence interval. Then we can continue the conversation.

What I want to see in the next cycle of this story is concrete.

I want to see whether GIANTX discloses deal terms. If they publish the contract length and scope of use, the market can price this more accurately. If they keep it sealed, I will read that as a signal the marginal edge is smaller than the marketing claim, because large edges always have an incentive to be displayed.

I want to see whether other LEC organisations sign equivalent agreements within the next two transfer windows. If they do, the market is flattening itself and exclusivity was merely the first phase of an ordinary commercial cycle. If they do not, one member holds a variable nine others cannot reach.

And I want to see whether Riot Games or Valve updates its rules on analytics tooling used during between-game breaks. I have watched this industry long enough to know publishers write rules only after a contentious precedent exists. The precedent has just appeared.

For my part, I will add one line to my notebook for every GIANTX LEC match next season: elapsed time from the previous game's end to the first ban. I will count it across eighteen regular-season matches. If that number falls steadily, I will know the tool is working, and I will know where it is working. If it stays flat, I will know that four minutes and forty seconds is still four minutes and forty seconds, and that the name iTero is only a name.

The question I leave for the next cycle is not who wins the LEC. The question is this: when publishers finally finish writing the definition of a coaching tool, will they define it as a purchasable privilege, or as a utility that must be shared equally among all members? Answer that, and you will know where next season is decided — on the rift, or inside a contract clause nobody reads.