Team Liquid tackles esports data with AI

In the 2022 Spring Playoffs for multiplayer video game League of Legends, Team Liquid suffered a shocking 0-3 loss to rival team Evil Geniuses, which prevented it from moving on to the 2022 World Championships.

“Out of nowhere, they pulled some counter picks [during the draft] that we just had no visibility on,” says Jesse Hart, senior director of sports science and analytics for Team Liquid. “It was a dark day for us, but it’s a dark day that’s become a lesson. We have to be aware and prepared for these things.”

While there’s no shortage of data when it comes to League of Legends — game data volume for this arena battle game runs to more than 2.1 million hours — Team Liquid was sideswiped by Evil Geniuses’ picks during the draft, in part because it was unable to bring insights from that data to bear. Hart and his team already worked closely with partner SAP to analyze their data. Now it was time to leverage the power of AI to make sure they never suffered another surprise like the one inflicted by Evil Geniuses.

The initial phase of each League of Legends game is a draft to select champions, as they’re called, or the characters players use during a match. The game has a pool of 166 champions, each with unique abilities and synergies with other champions. During the draft, each team also gets to ban several champions from the pool, meaning that neither team can pick those champions during the draft. They then take turns picking the champions that their teams will play, with several more bans available during the process. The goal is to pick a team of champions with strong synergies, while countering the opposing team’s abilities with those of your own picks. And to add to the pressure, players have less than 30 seconds to make each pick during the draft.

AI changes the game

In the past, Team Liquid’s coaches prepared for the draft manually. They watched footage to analyze opponents’ draft behaviors and champion preferences. They attempted to anticipate the opposing team’s picks and bans to develop their draft strategies — no easy task in a game with 445 quadrillion possible options for team selection.

While most teams still do things manually, Hart and his team worked with SAP to use SAP AI Core to train an AI model to automate the process.

“We have the AI trained on each team’s drafting profile, which is our approximation of their value judgment systems when they approach their pick,” Hart says. “What things do they prioritize or don’t? What things have they got in their back pockets that they might pull out? Being able to get ahead of the conversation when they pull out a surprise is invaluable.”

They built the solution on SAP Business Technology Platform (SAP BTP) and store 1.6 TB of game data from past games in SAP HANA cloud. The AI delivers suggestions of the best draft picks and bans to optimize win chances, and during the draft, it visualizes the predictions and provides the current winning probability after each pick and ban.

The tool also allows the team to train for drafts against any professional team.

“Draft is probably one of the most impactful things we do as a coaching staff,” Hart says. “It’s setting up our team for success and making sure the other team is limiting their success. Any advantage we can get in that realm is hugely beneficial to us, so it’s a very natural candidate to start looking at for AI applications.”

Still on the ascent

Since its founding in 2000, Team Liquid has won more than $43 million in prize money and 120 major titles and counting across the 15 esports games in which it competes, including Dota 2 and League of Legends. In the past several decades, esports, or competitive video gaming, has quietly turned into a massive industry.

“The viewership of esports is massive; it will boggle most people,” says Steve Arhancet, co-CEO of Team Liquid. “The League of Legends World Championship in 2023 had as many people watching it as the Super Bowl. There are a lot of gamers and there’s a lot of people that watch other people play.”

Because esports are natively digital, everything generates data, from the way players move and use their champions during games to the players’ communication with each other over their headsets during play.

“The average esports game, we’re taking 50,000 data points from each match and we’re talking 20,000 games a day,” Hart says. “That’s a ridiculous amount of data.”

Team Liquid’s use of AI isn’t limited to draft, though. Hart notes that it’s using voice transcription software with the players’ comms — players talk to each other during every moment of the game — and leveraging AI to analyze that chat for insights to help players improve. It’s also using gen AI to support coaching staff.

“The strength of generative AI is its ability to make non-experts look like experts,” he says. “Let’s say we’ve got a really talented coach who’s great at using the data and making cool things happen with it, but they don’t know how to get that data. Gen AI bridges that gap. We can round out the skill sets of all of our coaching staff using generative AI to level up the team.”

Honing skills on demand

Gen AI also helps players and coaches pull clips of footage to study. For example, Arhancet says a player might ask the bot, “Please show me 10 games where Cassiopeia is being played by Faker, in the first five minutes, where Cassiopeia gets a solo kill.” From there, it digs and pulls up videos on demand with links to the exact times.

And recently, the team won the League of Legends Championship Series (LCS) 2024 Spring tournament, beating FlyQuest 3-1.

“I know it’s working because I’ll walk by one of our players while he’s on the computer playing solo queue, and on the second monitor he has the SAP solution up, and he’s using it between games,” Arhancet says. “To me, that’s the moment where you know you’ve built something incredibly powerful because it’s being endorsed by the players and coaches themselves.”