Why we're still waiting for Counter-Strike's data revolution

You can collect advanced data from a Counter-Strike demo at every single tick. But that potential is still waiting to explode.

The advent of advanced statistics and empowered data departments has transformed traditional sports since the turn of the Millennium. Moneyball, the book (and movie) about how data and a bold GM took the middling Oakland Athletics into back-to-back playoffs, thrust this concept into the public sphere but change has only accelerated since.

In Basketball, the mid-2010s brought the '3 point era,' as teams eliminated most shots between the 3-point line and right next to the hoop.

Over the pond, long shots went on the decline in European Football as the advent of Expected Goals (xG) led teams to prioritize creating higher-quality chances over low-percentage snapshots.

These two trends are easy to visualize but only scratch the surface. Backroom sports departments are unrecognizable now compared to twenty years ago. The age of scouts, of the prioritization of 'eye tests' and years of experience, is all but over. Their skills remain important, but where scouts go, who they watch, and what they watch for are all data-driven decisions.

In Counter-Strike, though, data departments are barely a dream — and might never even be needed. It is easy to imagine CS2 is on the same path as these traditional sports, just a few years away from its 'Moneyball' moment as teams race to hire data-savvy general managers to make every Rostermania count.

You could argue something similar to the 3-point-era and xG has already happened, when Astralis developed precision grenade stacks and kickstarted the saving era as more and more teams shied away from retakes. But that was not the beginning of a grand data revolution.

Scouting now is still done in much the same way it was in 2016. The coach, in-game leader, and a representative from the organization sit down and sieve through a longlist via trust in their eye test by watching demos and picking up references from a player's former teammates.

"Stats aren’t that important when choosing new players," kassad tells us. "There are dozens of more important factors when making a team. First of all, I like to get feedback from people that worked in the past with players I’m interested in signing. After that comes demo reviews mainly focused on seeing how the player handles high-pressure situations, how he reacts to certain situations happening mid-rounds, does he have the amount of initiative needed for his position, decision making, and so on."

In some organizations, it's the players in kassad's role. In a world where eye test is king, the players are the overmighty barons. It is an art, not a science.

It is easy to draw comparisons between this status quo and the antagonistic old-school scouts pictured in Moneyball. When it is a personal decision, personal relationships factor in the decision. You don't need to trust the word of a former coach if you've worked with the player yourself before. It's easy then, for kassad to sign Joakim "⁠jkaem⁠" Myrbostad and Nemanja "⁠nexa⁠" Isaković because they are known quantities.

It's the same for agencies. It is not a coincidence that all five Vitality players were signed to Jérôme Coupez's Prodigy Agency when they arrived.

Jesper "⁠JW⁠" Wecksell has talked about how talking to former players has meant EYEBALLERS have missed out on a "whole generation" of Swedish talent. "We've missed some incredibly good signings because we may have listened to rumors more than giving the people in question an honest chance ourselves," he told Dust2.se.

But this is a system that, broadly, works. The best talents still make it to the best teams. Spirit did not find Danil "⁠donk⁠" Kryshkovets with some super-advanced algorithm; they did so with an empowered old-school scouting system.

Trying to follow traditional sports can be a trap itself, an anachronistic application of principles that seem good but simply don't fit an esport like Counter-Strike.

For some, positions and roles fit this example, as the community increase their focus and scrutiny when transfers turn a player's game upside down. Previously, eyelids may not have been batted but now Mario "⁠malbsMd⁠" Samayoa's signing to G2 — a clear prodigy with a 1.30 rating for 2024 in M80 — was questioned on this basis.

"I hate this trend of the last few years, where everyone is so fixated on roles," said Richard "⁠Richard Lewis⁠" Lewis on stream talking about the new G2 roster. "People talk about roles as if they are immutable qualities that are totally inflexible. I'm from a time where roles were almost a secondary concern: We get players who vibe well, we've got a good tactical system, and we will adapt."

Yet, the reality is that this community trend is following the scene. Roles and positions discourse is simply catching up to what was already a problem.

Peter "⁠dupreeh⁠" Rasmussen spoke frequently about his unhappiness with being moved to lurking positions in 2016 to free up Markus "⁠Kjaerbye⁠" Kjærbye. When he was back in the map control pack after Emil "⁠Magisk⁠" Reif's arrival — a happy accident, given the ultra-aggressive Kristian "⁠k0nfig⁠" Wienecke was their primary target — Astralis kickstarted their era.

Janusz "⁠Snax⁠" Pogorzelski joined MOUZ in 2018 and, despite winning ESL One New York, wilted away in Martin "⁠STYKO⁠" Styk's challenging supportive role and was removed. SK signed the super-aggressive João "⁠felps⁠" Vasconcellos (a Fernando "⁠fer⁠" Alvarenga clone, really) as Lincoln "⁠fnx⁠" Lau's replacement, but looked far better after adding Ricardo "⁠boltz⁠" Prass.

More recently, Astralis did not even qualify for the Major after trying to fit Benjamin "⁠blameF⁠" Bremer, Martin "⁠stavn⁠" Lund, and Jakob "⁠jabbi⁠" Nygaard into the same team.

Players often believe they are more flexible than they are. 'We can play any role,' they say, with faith that a modern system can distribute fragging opportunities more evenly. Even as malbsMd's old teammates question his G2 role change, the man himself has full faith that he can adapt.

Read more

Swisher: "malbsMd doesn't deserve those roles [in G2], he should be in ones he shines in"

Mostly, the move has paid off, even if he has gone from a 1.30-rating matchwinner to an X-factor, with a clear reduction in the number of duels his new roles allow him to take. He is still winning fights, fair and unfair, but he is seeing fewer of them.

But other teams are taking a different road, with GamerLegion being a shining example with Ashley "⁠ash⁠" Battye at the helm of prioritizing data and positions in his scouting.

"I'm a big Liverpool fan," ash tells HLTV, "and this is the approach that they've had for the last few years: Finding people via statistics, maybe undervalued, and going for those guys."

The basic approach is still rudimentary, and what you would expect: ash and assistant Adrian "⁠imd⁠" Pieper put together a longlist, go through as many statistics as they can, and then use those stats to narrow the list down to a manageable number of players to look at demos for.

The difference is in what stats they are using.

"Skybox has been a godsend for us," ash told us. "What we found was that the statistics are broken down into roles. For example, you can pick five teams and it separates people's ratings on buy rounds, eco rounds, against ecos, and it finds the rating they 'should' be [based on their positions]."

It's similar to HLTV's own 'target rating', which gives players a ballpark average rating you would expect based on their positions. For example, a player with donk's CT positions would average a 1.10 rating, and someone in William "⁠mezii⁠" Merriman's would have 1.02. A player in tough CT roles with a 1.10 rating, therefore, deserves attention in a way a star rifler might not.

Read more

Target Rating: Identifying over and under performers on CT side

Skybox then lets ash and imd go even further, by looking at rounds between teams on full buys.

"Stats on HLTV, they might say 1.15 and then you look into it on Skybox and you can check ecos and he has a 2.16 rating against ecos for the year or something, and it's inflating their rounds like crazy. But some players with that same 1.15 rating might only have a 0.97 vs ecos over the year.

"We ended up finding FL4MUS, his ratings and [opening] attempts were insanely low versus ecos, but on buy rounds his attempts were very high and his rating was 0.2 above the average in every position he played, compared to all these top teams."

Compare that to fnatic's blameF, who has 36% opening attempts in anti-eco rounds and just 18% in full-vs-full since July 23.

This was early this year, back when GamerLegion thought they had missed out on the Copenhagen Major (and before Nemiga and Timur "⁠FL4MUS⁠" Marev were on the radar) as they searched for a replacement for Nicolas "⁠Keoz⁠" Dgus.

"I literally had no idea who FL4MUS was," ash continued. "He just stood out. From there, he went to the shortlist, and we watched him afterward, and that's why I saw him pass the eye test unbelievably as well. So the statistics make the shortlist, and then from there I go through and break down if the numbers are that good, is it sustainable the way he's playing, can he fit in a system, and go from there."

This use of Skybox, ash admits, isn't even what the advanced Anders "⁠Anders⁠" Blume-backed service is designed for. Scouting is a natural extension of their main goal: Automating laborious analyst work.

"Our overall ambition is to make the research you do on opponents way faster," Skybox CMO Jacob Dietz says.

The company has a host of tools for just that: A 2D replayer you can layer rounds over, a veto simulator, leaderboards for each position, and the aforementioned economy filters on team and player stats.

But these measures aren't designed for recruitment purposes, not really. The 'Playbook' page, pictured below, is more the type of data work Skybox wants to be their niche: They have analyzed every team's rounds to show trends (how often they rush, use set strategies, or default) and specific versions of their rounds.

Astralis use a B execute with a CT smoke in 24% of their rounds, 10% more than the average team. Something like this is something an analyst would notice in their extensive demo review, but a tool like Skybox points them in the right direction.

"We want to save teams a lot of time," Dietz continues. "We want to do the analysis they do very manually. We want to be able to do that at the click of a button."

You can even see your team's hold success on a bomb-site, retake success, and post-plant success, more hints into where a team is strong, and more importantly, where they are not.

Skybox is a fantastic tool, there is no doubt. But it is still in development; clients are dependent on their dev team for new features. You can give feedback, but it can still be hard to get bespoke statistics.

For that, you need data analysts — who cost a lot of money. The average US data analyst can earn $79,113 a year, according to Indeed. It's next to impossible to justify that cost if you're an ordinary esports team.

Liquid analyst Jay "⁠DeMars DeRover⁠" Li is a rare example of a data-first Counter-Strike analyst.

"I didn't play the game at all," he tells HLTV. "Liquid was very patient with that."

Modest and shy, he's not even sure he'd hire himself. "To be honest I don't think it's that efficient to hire [a data person]. These skills do not overlap. Building good software, good math, understanding statistics and probability theory or whatever, and the skills of knowing what a default is and the current meta — they don't overlap at all."

Traditional analysts, for DeMars DeRover, have an easier time using public tools like Skybox or Leetify to gain data insights than he did learning the game from scratch. This does not mean he thinks data is useless in CS, far from it. It is just that, from the perspective of the budding data analyst, a team is not the home he would recommend.

In sports, teams and consultancy companies hired by teams drive the data revolution. In Counter-Strike, it's external, public-facing companies like Skybox and Leetify and hobbyist research professors like Peter Xenopoulos, whose research led to teams force-buying after losing the pistol round more aggressively, at the wheel.

The key reason for this is the quality, and availability, of Counter-Strike data. "This data is free, very cheap to get, very easy to work with, and a lot of the parsing algorithms are written already," Li explains.

"I don't think any sport, professional or whatever, has as good data as Counter-Strike. I would stand by that for sure. Frequency, accuracy, freeness, how quick it is to download, how readily available it is after each match."

What shines through our conversation is the respect Li has for traditional analysts, a role he has basically taken up. "At this point I'm basically just an analyst who made some tools and made his job easier." There is no Moneyball data department at Liquid. Roland "⁠ultimate⁠" Tomkowiak seemed like one, but since then it has become clear it was a traditional scouting job by Russel "⁠Twistzz⁠" Van Dulken to uncover him.

"I'd say [scouting and transfers] is a small part of my role," Li says. "[Analysts] are not as experienced as the players when it came to the eye test, and don't have as much say as coaches when it comes to roster moves.

"The stats can help us look at a wider range of players. Maybe we don't have time to watch every tier two demo but we can analyze it, and get us a shortlist.

"Roster moves are so precious, you only have five of them, and so much is invested into them, that it's really hard to justify with just numbers. The people with the biggest input are the key decision makers on the team not the math numbers folks in the back."

So we are back where we started. Data is involved, in so far as helping to filter a longlist, but it cannot be relied upon compared to the eye test of a professional player.

But this is not the universal view.

"You have to look at [a data department] as a multi-year investment. The ultimate reason why you want to build a team that can do this stuff in-house is where you're going to set yourself apart as an organization isn't going to be the access to the data. It's going to be what you do with the data."

That's Soham "⁠valens⁠" Chowdhury, major-winning coach with Cloud9 and now Head of Gaming & Data Science at Evil Geniuses.

The organization does not enjoy the best reputation in Counter-Strike, it's fair to say. valens' data-driven insights were not enough to make the 'Blueprint' project a success, as various North American rosters tripped up on themselves before the organization left the game in January 2024.

But there is no doubt valens' work is groundbreaking, far beyond what any other team is doing. Just because it did not translate into trophies, or turn Paytyn "⁠junior⁠" Johnson into Oleksandr "⁠s1mple⁠" Kostyliev, does not make it any less interesting.

Evil Geniuses previously hired data analysts in each game, including occasional HLTV feature writer Juan "Hepa" Borges. Now, it is more stripped down, with a team consisting of valens, an annual intern, and much of the "hands-on keyboard" work being done by technology partner HP enterprises.

As far back as 2019, valens — who has a Master's degree from Stanford — was using in-game data in combination with communication files to analyze which players were over-calling or under-calling based on information demos proved they had available.

In-house, the team has their own rating, named 'EGR,' because, obviously, that factors in communication, utility quantity, utility quality, mechanical skill, and how well you perform in what valens calls "moments that really dictate the outcome of a game."

Communication, if you ask anyone else, is an intangible. But EG, when they trial players, use their comms as a tangible skill to analyse and turn into quantitative data. The theoretical recruitment benefits are obvious.

They also emphasise specific types of situations. Rather than an overall kill count, what is important is a combination of how many 'fair duels' a player wins and how often a player can make a duel 'unfair' in their favor.

They are now using Machine Learning, specifically 'LSTM,' to evaluate player performance in a sequential manner. It is, valens admits, a bit of a "black box," but it is a crucial part of EGR.

But data, even for valens, is not a silver bullet. A cursory glance at EG's results can tell you the same thing. It is an enhancer, something to increase efficiency, to save coaches time or an organization money in the off-season. It is not a miracle worker.

"The majority of CS, you need to have a feel for the game. If you lose touch with what's in front of you, if you started as a player thinking, 'they're going to hit B because they usually hit B in 3v3s,' that shouldn't impact your rotation over your teammates' comms about where the opponents are."

There are objective data insights — 'this team is weaker on B than A against pistols.' But that does not mean it is objectively correct for an in-game leader to always hit B on those rounds.

It's also hard to be universal. Should there be a line where you always save? Take a 2v3 on Inferno's A site, a situation where many teams would save without question. It's a low probability retake, however you cook it. But what if data shows your opponent is particularly bad at post-plants there?

You also want to stay unpredictable. Doing the same thing, again and again, is rarely optimal in a game with as complex a decision tree as CS even if the numbers, on paper, say it is.

Anti-strat work from traditional analysts, hoovering up VODs or using tools like Skybox, has already made it a game of cat and mouse. You might identify a clear trend in data or a demo, but turn up to the best of three and your opponent has completely changed everything.

This is true in sports too, of course. Misuse of data, and blind faith in a flawed model, is a trap in itself.

Liverpool's return to the height of English football came from the marriage of data guru Ian Graham, former video analyst and eye-test specialist Michael Edwards, and a generational coach in the form of Jurgen Klopp. All three parties had to buy in, and cover each other's weaknesses.

Edwards and Graham, the equivalent of a GM and Data Scientist duo, were most fractious. Early on in their partnership, Edwards would rubbish every part of Graham's models; point out every flaw, every gap in its vision. Only once the model satisfied Edwards did it become part of decision-making.

The potential for data in Counter-Strike is limitless. 64 times a second, information is collected that can lead to insights. Everybody has free access to data libraries traditional sport teams fight tooth and nail for.

Rory Smith's book, Expected Goals, opens with the story of Ashley Flores, a Philipino 'data operator.' His job is to watch football matches, and hit a shortcut on his keyboard every time something happens. CS has no need for such infrastructure. It is built into the game.

But nobody has managed to turn that into trophies. No team is our equivalent of Oakland A's, or Liverpool. Who knows what Counter-Strike's data revolution might look like?

What seems clear is that teams would benefit from an increased focus on positions and specialists, data-driven self-scouting for unpredictability, and a more rigorous scouting process that puts analytics — fair duels, economy-adjusted performance, and detailed utility and communication analysis — at the forefront.

It would take a bold organization chief to invest in the midst of esports winter with 2025 just around the corner. Financial outlay on data, at this moment in time, is a risk, in a field that tends towards short-termism at the best of times.

Innovation, however, always carries risk. A plunge into data, long-term, is one that will increase the hit-rate on transfers (and thus save money on buyouts), improve performance, and give early adopters a discernible edge. We know gold is under the mountain. We just don't know who will find it.