Jack Williams on iTero, Giant X, and the future of AI coaching in esports

iTero, the AI-powered coaching app from Giant X (formerly XL Esports) recently hit the 1,000,000 download milestone. The milestone comes after an impressive period of growth following the 2024 acquisition by the LEC organization.

The story of the company starts a little less glamorously, with Jack Williams, a former Deloitte employee finding himself banned from every UK betting site for winning too consistently when betting on esports. iTero was born from a visit to Twickenham, and a realisation that the model he’d built could help professional (and, with a shove from Overwolf, your average League player, too).

We sat down with founder Jack Williams to talk through iTero‘s journey from a betting model capitalizing on poor pricing to being an LEC scouting machine. We covered everything from Riot’s rulesto where AI coaching is headed next, and when it becomes AI-assisted cheating.

ESI: Congratulations on hitting the 1,000,000 download milestone with iTero. Could you give some background on your success and how iTero came about? 

Jack Williams: Let’s start from the top — I’ll tell the full story. Honestly, what interests me most is the AI-and-esports angle. A lot of that work has fed into the consumer product, which has done really well, and it’s a genuinely useful internal tool — but on its own, “an app that helps people get better at the game” only makes for so much interesting reading. What people actually want to know is how to beat a top team on the weekend and make it to Worlds.

I was previously in professional services as an AI engineer — before “AI engineer” was really a job title, so at the time, it was more of a data scientist role. First at HSBC, then at Deloitte. I have a Master’s Degree in AI, so I’m fairly technical on that side. During COVID, while at Deloitte, I was betting on esports games — not because I was a degenerate gambler, but because I felt I had an edge. 

I noticed the betting odds weren’t moving once the draft was locked in, which struck me as odd, because everyone in the scene knows the draft matters — it’s not a perfect predictor of who wins, but it at least tells you whether a team got a strong pick in the first rotation, or got counter-picked.

So I started building AI models that analyzed the game state at the point of the draft and predicted the winner from that. I made a bit of money on the side doing it — until my betting accounts got banned. Apparently, in the UK, bookmakers can just cut you off if you start winning consistently, which I thought was a bit unfair. 

But I now had this tool and no real use for it, so I reached out to Kieran Holmes-Derby (Co-Founder of Excel Esports) and basically said, “I’ve built this thing, it works, and I’ve been banned from every betting site. Could I come in and show your esports team, just to see if it’s useful?”

I went to see him at what was then their base in Twickenham Stadium — a bit surreal, walking up to Twickenham with the England rugby team training on the pitch and the esports team training inside the stadium. I came away from that thinking there had to be something I could build to help players draft better and win more games. So I resigned from Deloitte and went all-in on the idea: build tools for the team, help them draft, they win more, everyone’s happy.

ESI: How challenging was it to leave a full-time job and launch a standalone product back then? 

Jack: I spent about six months building it and three months trying to sell it, before realizing that in 2020, nobody could afford — or particularly wanted — anything AI-related. It wasn’t even a fashionable concept yet. More importantly, most teams didn’t have competent analysts at all; it was largely people on Google Sheets making things up, with no real grasp of sample size. 

Walking into that environment and saying, “Here’s an advanced, black-box AI model telling you to pick this champion, trust me” — that didn’t land well with coaches or players. I sold zero contracts and was fairly stuck, so I planned to go back to Deloitte.

Then, at an ESI conference, I was approached by Overwolf — I’d won a pitch competition that year, so I’d made a bit of a name for myself. Someone there asked if I’d considered building the tool for consumers instead of professional teams. I hadn’t, but figured it was worth a try. Three months later we had a live product. It looked genuinely terrible, but we went from zero to 10,000 downloads very quickly, and retention was much higher than typical apps in that category — most apps in this space see 10–20% retention, with people uninstalling almost immediately. Ours held on: people thought it looked rough but liked what it did for their drafting, so they kept it.

Overwolf then gave me a non-dilutive grant — no equity taken — which let me hire a developer, and we spent the next two years building it into what it is today: an AI companion app that helps players progress toward their next rank. After two years we were at around 100,000–200,000 downloads. Then we were acquired by Giant X — originally XL Esports — which brought things full circle back to Kieran. That happened in 2024.

ESI: And post-acquisition, what changed for the company, and the product? 

Jack: Once I was integrated into the team, the focus became pure growth: how do we reach more players and build a better product? We quickly found real synergy between an esports organization trying to win the LEC and a product helping the average solo-queue player win their next game — synergy in decision-making, since I now had access to coaches and players who are genuinely excellent at the game, which improved the model, and synergy in marketing, since it’s much easier to promote a tool like this when you’re backed by an esports team.

Fast-forward two-plus years, and we’ve crossed a million downloads. We were at roughly 500,000 last year, and around 200,000 the year before — so very rapid growth since joining Giant X. Users have grown roughly 300% over the same period — downloads have doubled, active users have tripled. We’re in a great place with the product right now; we think it’s the best on the market by some distance.

We’re also excited about what the team can do more broadly. Giant X is known for finding rookies — we don’t have the biggest budget in the league, but we consistently finish around fourth. Part of that comes from the statistical models we use to identify new players — I wouldn’t necessarily call them AI, but they’re advanced. That’s helped us overperform, and while most of the credit goes to the coaches and players, some goes to the systems we’ve built by applying the concepts from our solo-queue AI tools to professional scouting.

ESI: Great stuff. You mentioned it’s been a mad year of growth — what do you think really drove that? Was it simply more investment in the product?

Jack: There are a few things, and I don’t want to miss any of the important ones, but fundamentally the most important factor was that product was always the priority — at every stage. We never decided to just fill the app with ads for quick revenue, or run cheap, spammy ad campaigns to inflate download numbers with no real substance. The only thing I cared about was retention. Because of that, most of our investment went into improving the product and the retention rate, to the point where most people who download the app are still using it a week later — I don’t think any other app in the space can claim that over half their users stick around after a week, though there’s no public data to compare against.

So: product and retention first. Second, Giant X gave us a real advantage as a Spanish organization with strong ties to Spain. Almost all of our growth has come from Spanish-speaking Latin America and mainland Spain. When we joined the organization, roughly 20% of our users were Spanish-speaking — that’s now around 70%. If I listed our top five countries by users, every single one would be primarily Spanish-speaking — Spain, and I believe Argentina, Chile, Mexico, and one more in Latin America.

ESI: I think you’re right…

Jack: What we were able to do was lean on our own community: a company full of Spanish speakers with genuine Spanish-language networks, to grow organically. We reached out to local content creators, usually fairly small ones, and worked with them on a personal level: “Here’s the tool, try it yourself, tell us honestly what you think. We’re not going to ask you to oversell it. If you genuinely don’t like it, let’s not work together. But if there are parts you do like, make a video about it.” That kind of organic product placement performs far better than a scripted ad that just hits marketing talking points. If creators genuinely like the product, they talk about it well.

Third, Giant X has its own internal educational content channels — rebranded a couple of times, originally GX Plus, now called Iter. I don’t know the exact figures, but we’re talking hundreds of millions of impressions annually across those videos — tips, tricks, “how should you play this,” that sort of content. That content naturally lends itself to promoting the product: “Here’s the optimal build this patch — if you want to know the optimal build at any time, download the app.” Very natural placement through channels we already own.

ESI: I recall — and I might be paraphrasing the exact wording — that Riot’s policy bars third-party tools from creating an unfair advantage players wouldn’t otherwise have. Have you ever had that question raised, or had to pull back a feature because Riot said no?

Jack: With Riot, we’re very careful with them. We’re probably the most conservative of the third-party apps: if Riot tells us not to do something, or the wording is ambiguous, we simply don’t do it, because staying on the right side of Riot matters to us. Given our esports connection, we’re deeply embedded in the Riot ecosystem, and we don’t want to jeopardize that. We also want players to trust the app — you don’t want someone downloading it wondering if it’s functionally similar to an aimbot. So we avoid any gray-area features entirely; we only build what’s clearly permitted.

Riot’s own wording isn’t especially precise — their third-party tools policy says tools can’t provide a “competitive advantage,” but that’s a fuzzy standard. Technically, sorting a stats site by win rate and picking the top champion is also using available data to make the best decision — that’s a competitive advantage too, by that logic. In my view, what would actually be unfair is surfacing information that isn’t accessible through the game’s own UI — pulling from the game’s underlying code, capturing events that aren’t meant to be visible, and displaying them in ways Riot doesn’t sanction.

Sometimes Riot’s decisions are a bit arbitrary, but we just follow them. For example, summoner-spell timers — a simple UI element you’d click when someone used Flash, showing a countdown until it was back up — were fairly standard across tools. Riot decided they no longer wanted them in the ecosystem, so we removed ours. Generally we stay conservative, and if Riot ever asks us to change something, we do it immediately.

ESI: How do you see this changing as tech continues to grow and develop? 

What will get more interesting is when it becomes harder for Riot to enforce this kind of control. Right now, they control me largely through API access — if I upset Riot, they pull my API access, and I no longer have an app. That’s a significant deterrent.

But tools that run advanced AI models directly on a player’s monitor, generating their own data by reading in-game visuals rather than using the API, are much harder to control. That’s a genuinely concerning direction. Maybe less so in League, more so in FPS titles. 

You could end up with something running invisibly, not on the computer but effectively reading the monitor, providing an unfair advantage that’s very hard to detect. For example, a high-end monitor might be able to identify exactly where someone’s head is on screen, or feed instructions a player wouldn’t otherwise have. That’s a worrying space for competitive integrity. But broadly, we stay more conservative than we strictly need to be, specifically to stay in Riot’s good graces.

ESI: Do you ever worry that someone might just copy your feature, given the lack of IP protection? 

UI is trivial to replicate now with AI coding tools. Someone can screenshot a UI, feed it to an AI model, and ask it to recreate the interface. What’s genuinely difficult , or even close to impossible today, is recreating the predictive models behind it. Those results come from a lot of training, a lot of data, and a lot of time. That may become easier for AI to replicate eventually, and probably sooner than people expect, but as of today, you can copy our UI — you can’t copy the underlying intelligence.

ESI: How does it work being embedded within an esports organization? Do you still offer services to other professional teams, or does trust become an issue given your Giant X affiliation?

Jack: No — we work entirely within Giant X, including our LEC team and our Spanish league (LVP) team, and would extend to any other titles we compete in going forward. We’re internal-only. There’s real demand out there. Teams with modest budgets would love a tool like this to find undervalued players, but for now we haven’t opened it up to other teams, purely for competitive integrity reasons. Any team with the budget to pay for something like this is a team we’re likely to face at some point, so for now the answer is no, though that could change in future.

ESI: Does your model try to account for less quantifiable factors — player tilt, greed, tendencies that don’t show up cleanly in the final match data? Do you break things down by early-game phases, and how deep does the modelling go at the competitive level?

Jack: Because of exactly the issues you’re describing, we tend to stick to solo queue — it’s much better suited to AI predictive modelling. There’s far more data, and you avoid the noise of, say, one Fnatic-vs-G2 result on one patch with two specific drafts. There’s very little to learn from a single data point like that. In solo queue, you can observe the same or similar draft repeated at high elo many times over and draw real conclusions.

Most of what Itero does day-to-day is solo queue analysis rather than pro-play prediction. You’re right that the number of variables involved in pro-play prediction is enormous — even with perfect data, you’re probably capped around 80–85% accuracy. 

And even a modest edge, say a 15% chance of an upset, will still play out at least once across a split if a team plays ten games. 15% sounds low, but across enough matches it still shows up. There are all sorts of stories, not just in esports but in football too, of people finding spurious correlations. 

This can take the shape of tracking a player’s Instagram activity, relationship status, whatever. Chasing an “edge” often is really just noise. Anyone can make a data point say almost anything if they ignore causation versus correlation.

ESI: Has your model ever contradicted established draft theory — going against what’s considered “meta” — and ended up shifting the meta itself? Or does it generally reinforce what’s already accepted?

Jack: On the pro side, I don’t want to overstate our influence. Those decisions really sit with the coaches. Any notable calls or innovations are down to the coaching staff and players; I wouldn’t claim we’ve reshaped the LEC meta. Where you probably would notice a pattern is in solo queue. As more people use the tool, more players are taking objectively favorable matchups. 

We’re now at a level of market penetration where it’s visible in the data: even at high elo, where usage is comparable across the elo spectrum, more drafts are being optimally picked than before the tool existed.

That gets interesting at a deeper level, though. If drafting optimally becomes so widespread, through us or competing tools, that everyone can predict “if I pick X, they’ll pick Y,” players start looking for a third move: flex picks, or off-meta picks. 

Say you’re a Vayne top player, and every time you pick Vayne top, you know the opponent will pick a hard counter like Teemo, you might start swapping roles mid-lobby more often: “I’ve been counter-picked again, let’s switch it up — I’ll go Vayne ADC, someone else take top.” So there’s potential for that kind of disruption.

On the pro side specifically, I think our scouting approach has been more interesting. What we’ve done is identify players who are fundamentally very good at the game mechanically, and then build coaching support around them. 

This is the opposite of the old approach of picking players purely because they’ve been in the scene for years, know how to handle scrims, live the “pro player” lifestyle, or can cope with stage pressure. That approach limits you to a small, recycled pool of “known” players. 

We weren’t interested in that. We look for someone fundamentally strong at the game, which doesn’t necessarily mean the highest-ranked solo queue player. Ranked skill doesn’t automatically translate to performing well on stage. We don’t just rank by LP and pick the top name. But we’ve built a system that identifies players likely to be fundamentally strong performers on stage, bring them in regardless of background, and build the supporting systems around them.

Jackie is our best example. He wasn’t coming from a highly competitive LFL or EU Masters team. We found him through the data, brought him in, built the systems around him, and he’s now genuinely our MVP in that role, as a result of a fundamentals-first approach rather than traditional esports scouting.

ESI: From a player experience angle — pre-game, in-game, post-game — which stage do you think matters most for someone trying to improve? Do people mostly use it as a pick overlay and move on, or are people meaningfully engaging with post-game reports? Do you track usage patterns to see which behaviors correlate with the most improvement?

Jack: It’s an odd one, because ideally you’d measure it by win rate, but the tool is essentially designed to push everyone back toward a 50% win rate over time. Play better with the tool, and you climb elo until you stabilize again — that’s just a statistical reality. 

So gains from the tool tend to be short-term by nature, but real. What we do know: once someone starts using the tool, their win rate rises as a baseline. And if you follow the champion recommendations closely, your win rate rises quite significantly in the short term. 

We’re talking roughly a 6% increase. That sounds modest, but sustained over time, that’s enough to climb from, say, Gold to Platinum just by following the top suggestion consistently.

The main friction isn’t really an issue, more a choice players make: people don’t love always playing the exact same recommended champion. If you’re first or second pick, the top recommendation is often similar to what was suggested the game before, since not much has changed. Maybe you know one teammate and a couple of enemy picks, but unless it directly affects your lane matchup, the recommendation doesn’t shift dramatically. So people look for variety. This means turning down the “prioritize meta pick rate” weighting, for instance, to surface more off-meta options.

The clearest example is our champion-mastery weighting. By default it factors in how much experience you have on a champion, which is optimal for winning. Left on the default setting, you’ll win more. Almost everyone turns it down to zero, like, “I don’t care about my experience, just tell me the best pick,” then first-times a random champion in ranked and loses. 

That’s fine, honestly; the game isn’t only for people trying to optimize their climb. Some people just want to have fun and use it to discover new champions. Fundamentally, it’s hard to claim the tool “does” a certain win rate, because most people voluntarily set it to be less optimal, because that’s more enjoyable.

ESI: Where do you see AI coaching heading across gaming and esports more broadly? Do you think it’ll become standard, something publishers build themselves, and could it help level the global competitive landscape?

Jack: That’s a loaded question, so let me try to unpack it. First: I think it’s unambiguously good for game engagement generally, separate from esports entirely. If League of Legends existed with no esports scene at all, would tools like this still make the game better? I think yes. And whether that comes from third parties like us or from publishers building it themselves, I think this category will keep growing and improving, hopefully in a healthy direction.

My favorite reference point is chess.com. It’s very good at leaving the actual game entirely up to you — no move suggestions, no live coaching, no win-probability readout while you play. But the pre-game training and especially the post-game analysis are excellent. That’s really the direction I’d like to see games go: making it clearer why you lost, so it feels less arbitrarily frustrating. 

The hardest part of League, if you’re playing a couple of games a day at a 51% win rate — which is genuinely good, and will get you climbing eventually, is that each individual game still feels like close to a coin flip, which is demoralizing in the moment.

What these tools let you do is give players something to work on that isn’t simply “win or lose.” We have a macro coach that reviews your recent games and tells you, for example, “over your last fifty games, here’s the mistake you make most often” — so you can direct your practice time deliberately. 

Say the pattern is: you win lane, but consistently lose experience between the 14- and 21-minute mark. You can go into your next game focused specifically on improving that number, and that gives you a sense of control over your own improvement, independent of whether the other nine players in that game cause you to win or lose. You can lose the game and still feel good about it, because the specific thing you were working on — not throwing away your early lead — actually improved. That’s fundamentally better for the game and the player.

For esports specifically, will it close the gap with Asia? Probably not, largely because it comes down to economics and funding. If your regional market doesn’t generate enough revenue, you can’t reinvest in improvement, and it becomes very difficult to catch up. 

I won’t pretend to have the answer for fixing European esports economics. Everyone knows it’s a tough environment. Fundamentally, these tools cost money, the same as good coaches, good players, good facilities. Even things like a gym and a performance trainer attached to a team show measurable benefit. We’ve seen that in the data, and the same with sleep tracking and player wellbeing work. Can European teams broadly afford all of that? No. Same logic applies to AI tooling. If you can’t afford to build it, you can’t benefit from it as much.

Will it improve everyone over time? Yes. We’ve seen this in football, the NFL, baseball, basketball, it doesn’t matter the sport. AI helps teams improve. But it helps everyone improve roughly proportionally, so the relative gap doesn’t necessarily close. We’ll probably still be losing to Korea in ten years. We might have a good AI model by then, but they’ll likely have a better one.

ESI: So, it sounds like in-game coaching is something you want to swerve? 

Jack: I touched on it with the chess example. Not everyone in the market shares that view, so it’ll be interesting to see how Riot handles it as tools get more capable.

That’s when hen in-game tools start veering toward AI-assisted cheating rather than legitimate coaching. I’ve seen one example: a “phantom marker” that shows an icon on the map at the last location a champion was seen, even after they’re no longer visible. This is effectively remembering positioning information for you. That’s well over the line, in my view. Almost everything beyond the basic stat-comparison features we already have — like “how does your farm compare to the average player at your elo” — starts to cross into that territory.

There was another one that used voice prompts — literally telling you, “go do Dragon now,” “go do Baron now.” That felt like a clear step too far, especially when they used recognizable esports personalities’ voices for it. Fun as a novelty, but well past the line.

ESI: With improvement focuses such as the 14-to-21-minute XP tracking example, do players ever tunnel on that specific metric? Like, improving their own XP number while their team falls apart in every other lane. Is there a risk that focusing so heavily on individual performance metrics undermines overall team play and win rate?

Jack: I worried about that too, honestly. What we’ve actually seen is close to the opposite. Players who’ve played thousands of games develop very fixed habits, and the hardest part is breaking that routine at all. 

If we can temporarily turn a dedicated farmer into more of a ganker, even for a handful of games, I think that’s net positive. Even if they eventually revert to their old play style, because they’ve picked up a new skill along the way. 

Same logic as a support player in CS suddenly playing entry for a stretch: even if they go back to support afterward, they’ve gained something from the experience, and they’ll likely understand that play style better when facing it too. It’s fine if a player over-indexes on one improvement area for a couple of games and dips elsewhere, that’s part of learning to balance it. When they revert to their usual style, they carry that extra skill with them.