The VALORANT analyst's job is moving from reading other people's dashboards to asking your own AI about your own data.
We wrote up how that works today, with a real prompt and screenshot for every step. https://x.com/i/article/2118452281532190720
The VALORANT Analyst in 2027
How VALORANT data is moving from Excel dashboards to AI automation that you only have to feed with your own tools.
A scrim day ends at 17:45. The analyst has the VOD on one monitor and a stats page on the other and starts copying, round by round: who got the opening kill, where the spike went down, what each side bought. When the sheet is done the players have logged off, and the review waits until tomorrow.
We think most of the effort spent on boring, menial tasks like this can be made much more efficient soon. The numbers will look much the same as they do now. The person choosing the question will change: today it is whoever designed the screen, and soon it will be the analyst, typing into a chat they already use, with the team's own data plugged in.
This article walks through that change one step at a time, with a real prompt for each step. The examples use Claude connected to kogni, the VALORANT tool we build, so read our view with that in mind. The team in the screenshots is Aurora Five, a demo team with fictional players and twelve scrims on file. The pro data is real, read from VCT 2026 broadcasts.
1. Today's struggle: hours lost copying between screens
Almost all VALORANT data reaches a team as a screen someone else designed. A stats site shows the columns its creators chose, based on the markers they think are useful. A team dashboard answers the questions its builder expected people to ask. A VOD tool lets you watch and tag rounds.
Those screens work well until your question isn't on one of them. Then you open a spreadsheet and carry numbers from tab to tab, and the carrying eats into the hours you're meant to spend elsewhere.
Here is what the same job looks like with kogni: one prompt about your own team's data, answered in seconds to minutes instead of hours to days. This example is Aurora Five's last scrim.
One request instead of an evening of copying: the last scrim, round by round. Demo team, fictional names.
2. Tomorrow's shift: your data, accessible through your own AI
From here the direction flips. You stop visiting each tool's screen. The data comes to the assistant you already use, and you ask in your own words.
What makes this possible is a connector, a link between an AI chat (Claude, in these examples) and a service where your data lives. You add it once, sign in to the service and approve what the AI may read. After that, when a question needs data, Claude asks the service for it and works with what comes back. You get the answer in whatever shape you asked for: a table, a chart, a paragraph for the team chat.
Most connectors are built on MCP, the Model Context Protocol, an open standard Anthropic published in 2024 that many AI apps now support. You can skip the technical side. Before you trust any connector, though, find out three things.
What it can see: in kogni, what your account sees in the app. A player sees what a player sees, a coach what a coach sees.
Who decides: a team manager switches AI access on or off for each kind of team data: scrims and stats, player performance, ranked games, the schedule, the team workspace and drafts.
What it can change: nothing without a person. Claude can prepare a task or a schedule event as a draft, and nothing happens until a manager approves it in the app. The one thing Claude creates directly is a watch room for pro VODs, shown in section 6.
Public data and private data
kogni gives Claude two kinds of data and keeps them apart.
Public data is pro play: VCT 2026 matches read from the broadcasts. Any kogni team on the PRO plan can ask about it, and asking shows nothing about your team to anyone else.
Private data is your own team's: scrims, player stats, ranked games, the schedule and the workspace. Claude reads it only when a manager has switched that kind of data on, and only within your permissions in the app. Other teams can't see it, and it never goes into the pro numbers.
That leaves the choice with the team. It can switch off every kind of team data and still use the pro data, or open its own data when it wants answers about itself. When an answer uses both, it labels which numbers are yours and which are the pros'.
The first thing to ask a new connector is what it can see. Here is that question for Aurora Five, with Claude's answer.
Ask a new connector what it can see, and what your team has switched off. Demo team, fictional names.
3. Public data: an even playing field
Pro VODs are public, and every team watches the same ones. kogni reads VCT 2026 broadcasts round by round: kills, plants, economy, and the position of every player the minimap shows, sampled every 2 seconds. That adds up to more than a thousand maps from this year's leagues, both Masters events, Champions and the Game Changers Championship.
You can open those positions inside the chat. Ask for kogni's pro positions viewer and it appears under the reply. For Lotus defense alone it holds about 395,000 position samples from 3,187 rounds on 153 maps, and you can change the side, the time in the round, the team or the player without leaving Claude.
kogni's pro positions viewer inside the chat: where pro defenders stand on Lotus. Pro data read from VCT broadcasts.
Your rivals can pull the same numbers, so public data alone gives nobody an advantage. Teams start to differ at the follow-up question.
Start with the obvious one: where do defenders die first on Lotus? At Champions 2026 the busiest spot is C Main, with 13% of the 139 first deaths on defense, followed by B Main and A Rubble.
Where defenders die first on Lotus at Champions 2026, from 139 rounds on 15 maps. Pro data read from VCT broadcasts.
Anyone can get that picture, so ask a narrower question. Losing the first duel on defense usually loses the round, but not always. Where was the first death in the rounds the defenders won anyway?
The same first deaths split by round result, on one colour scale: 38 rounds won, 101 lost.
Defenders still won 38 of those 139 rounds. In the wins, C Main holds a bigger share of the first deaths (16%, against 12% in the losses) and B Main a smaller one (5%, against 11%). Treat that as a lead to check on the VOD and nothing more, because 38 rounds is a small set.
4. Private data: your scrims
Scrims are where a team tests ideas, and nobody outside the team has them. They are the one data set your rivals can't download, which makes them the place where your analysis starts to look different from theirs.
This data is private, and it stays private. kogni keeps a team's scrims and official matches from the match data the team uploads, and only that team's members can see them. Through the connector, Claude reads them with the permissions the member already has in the app, and only while the manager keeps scrims switched on. No other team can reach them, in kogni or through its own AI, and they never go into the pro numbers.
The manager holds the keys. AI access can be switched off for the whole team at once, limited to some team roles, or ended for a single connection, and an ended connection stops working on its next request. kogni records every request an AI makes: which tool, for which member, allowed or refused. What Claude reads lands only in your own chat, under your Claude account's privacy settings.
Here Aurora Five asks how it executes and retakes on Lotus, across six scrims and 117 rounds. On attack the team hit C 20 times and won 18 of those rounds. Defense is weaker: Aurora Five won 4 of 30 retake attempts, and none of the 11 it tried at even numbers.
Executes and retakes on Lotus from six scrims, with the sample stated in the answer. Demo team, fictional names.
With the first scrim alone, the same question came back with a warning that 18 rounds are an anecdote. Six Lotus scrims are still a small set, and the answer says so: most situations rest on 1 to 4 rounds each. Even so, 0 of 11 retakes at even numbers is enough to send someone to the VOD.
5. Joining both: your scrims next to the pros
Each kind of data helps on its own. Put together, they let you ask: is this normal, or is it us?
Joining them doesn't merge them. Your scrims and the pro numbers meet only inside your answer, each one labelled, and nothing of yours goes into the pro data.
That only works when both sides are measured the same way. "We die a lot at A Stairs" says little. "We die at A Stairs far more often than pro defenders do, counted on the same grid of the map" gives you something to check on video.
Here Aurora Five's 229 deaths on Lotus defense, from six scrims, sit next to more than 10,000 pro defender deaths on the same grid. A Stairs and B Upper stand out: each holds 3 to 4% of Aurora Five's deaths, against about 1% of the pros'. Claude adds two limits of its own. Aurora Five's busiest spot holds only 9 deaths, so gaps of about one point are close to noise, and the chart shows where deaths happen, not why.
Lotus defense deaths: Aurora Five's share per spot against pro defenders', on the same grid. Demo team, fictional names.
kogni's coach review works the same way across a team's scrims. It runs 13 checks, such as trades within 3 seconds, winning after the first kill and retakes by players alive, with the same rules it uses on pro broadcasts, so the team number and the pro number can sit in one row. That review comes in the next section, because it is also the clearest example of checking an answer.
6. What stays human
The AI now fetches, counts and charts. The analyst keeps the parts that were always the job.
Asking comes first. The first-death heatmap in section 3 is the one any team would pull. The split by round result exists because someone thought to ask for it. Knowing which follow-up question to ask, the one that separates two explanations of an outcome, takes VALORANT knowledge, and the AI doesn't bring that knowledge to your team.
AI will not replace analysts, and it will not replace coaches. This is a deeply human role, and the context around each number decides whether an improvement found in the data works once it is applied. You have to understand the game, and you get out of a dataset, public or private, what you put into it. kogni is a tool that helps and guides with data. It can't replace the applied knowledge and nuance that coaching and analyst staff bring, or the human side of their job. It is here to help, not to replace, and it makes a mighty fine assistant: it gives coaches and analysts more time for the work that really matters.
Checking the answer is next. Every number has a sample behind it, so ask for it every time: "how many rounds is that?" kogni calls a gap "worse" than the pros only when it is at least 5 points and unlikely to be chance at that sample size. Below 15 cases it says "low sample".
Aurora Five's coach review shows how that works. When the team had only its first scrim on file, it had traded 3 of 45 deaths on defense within 3 seconds, against 17% for pro teams. That looked terrible, and kogni still called it in line, because 45 deaths can't tell a habit from a bad night. Twelve scrims later the rate is 38 of 382, about 10%. The gap is smaller now, and kogni calls it worse, because the sample can carry it. A few rows of the same review still say low sample: the 1v1 clutches rest on 7 rounds.
A coach review where every row carries a status: worse, better, in line or low sample. Demo team, fictional names.
The second half of the answer matters more than the table. Asked which rows not to trust yet, Claude points to the low-sample rows and the ones with no pro reference, and adds that "in line" means no conclusion, not that you're fine. It even ranks its own red flags: the even-number retakes are the weakest of the three.
The rest of the same answer: which rows not to trust yet, and why. Demo team, fictional names.
Broadcast data has gaps of its own. kogni reads about 99.7% of the kills listed on VLR (vlr.gg, the main VALORANT esports stats site), so a few are missing, and players the minimap doesn't show aren't counted. A good answer mentions both.
Deciding stays with the staff. Pro numbers describe what pro teams did. They don't tell your team what to do, and kogni's answers are written that way on purpose.
The review itself still happens together, on video. Numbers point at rounds, and the rounds still need the players in the room. In the last example Claude looks for the utility pro defenders used after the plant in Lotus rounds they won at Champions 2026. It finds 287 casts in 70 rounds, led by Omen's Dark Cover and Paranoia, and opens a watch room on a playlist of those 70 rounds in VOD LAB, kogni's pro VOD viewer on kogni.pro.
Pro utility after the plant on Lotus at Champions 2026, counted by ability. Pro data read from VCT broadcasts.
The team joins from a link, with or without a kogni account, up to 10 people. Whoever runs the room plays, pauses and skips the video for everyone, and the chat and the drawing on the video are shared. Opening that room is the one thing Claude does directly, and the room belongs to the person who asked for it: the answer says to open it signed in within 30 minutes to keep control.
The end of the same answer: the watch room link and how the room works. Room code hidden.
The watch room on VOD LAB: the playlist runs along the bottom and the round feed on the right.
7. The job in 2027 (our guess)
Everything in sections 2 to 6 already works today, even though most teams don't work this way yet. This section is our opinion about where it leads.
By 2027 we expect analysts to keep a library of questions the way they keep a callout sheet now: a weekly set, plus the ones they open after a bad night. They will rerun those questions after every block of scrims and spend the saved hours in the VOD.
We also expect teams to care more about where their data can go than about how many screens a tool has. When the data reaches the assistant your staff already uses, you can build the report your coach wants instead of the one a product ships with.
Players will ask their own questions too, inside the limits the team sets. Part of the analyst's work then becomes teaching what a fair question looks like and how many rounds it takes before a number means anything.
As we said in section 6, tools like this won't replace analysts or coaches. Porridge (@porridgeorge), a coach and kogni's co-founder, described what the saved time means from the inside:
I firmly believe in the human side of esports, but I also think we spend far too much time on necessary, menial work that can be assisted or automated. Specialists will keep their jobs and do more with the time they save. I can't tell you how many hours I've spent on manual input over the last few years, even with tools, because I wanted to dig into specific variables or was building specialist ratings and systems. This helps on many levels: it teaches less experienced staff what to look for at a foundational level, and it automates the menial parts of the job so coaches and analysts can move on to more advanced approaches, with more data and better-informed decisions than energy or time used to allow.
We also expect the approach to become more human, backed by data, with a virtual assistant you can hold a conversation with about a dataset. Feedback we've received tells us that many coaches find affirmation in the data and in the questions they ask kogni. When they have a feeling or spot a pattern, the data often shows what they already sensed, and when the sample is big enough to carry it, that leads to more assertive, confident map and scrim preparation. Seeing where and how often your team struggles in a scenario or an area of the map, with the data behind it, helps the staff work out why and how to approach it. Refining that, and leaving more time to think about both the theory and the practical side of VALORANT, can only be a good thing.
The bottom line is time. The hours saved can go into other parts of the work, inside and outside the game: sustaining, implementing or advancing your approaches, and the things that matter, instead of recording and copying data every week. More time with the team, in the server, or for yourself.
Time is a finite resource, and we should invest ours as well as we can.
8. How to start
You need a kogni account that is an active member of a team on the PRO plan, and a Claude account that can add custom connectors (this depends on your Claude plan). Your manager decides what the connector may read: a team that wants only the public side can leave every kind of team data switched off, and the pro data still works.
1. In claude.ai, open Settings, then Connectors, then Add custom connector.
1. Paste https://mcp.kogni.pro/mcp and name it kogni. The kogni app shows the same address under Settings, AI connections.
1. Click Connect, sign in to kogni, pick your team and approve. The connection lasts 60 days, or 7 if you switch that option off.
1. In a new chat, click +, then Connectors, and switch kogni on.
kogni also adds ready-made requests to the same menu, under Connectors, then Add from KOGNI: scrim review, weekly team report, opponent prep, coach review, a pro review room and a few more. Claude asks you for the missing details (which map, which opponent), so they make a good first try.
Claude's + menu: switch kogni on under Connectors, and find its ready-made requests under Add from KOGNI.
If you'd rather type, these work on day one:
- "What can you see about my team, and how much data is there?"
- "Review our last scrim round by round."
- "Give me a coach review of the last 30 days, with the rounds to rewatch."
- "Where do we die first on Ascent defense, compared with pro teams?"
- "Open a watch room with Derke's first kills at Champions 2026."
Name the map, the side, the period and the format you want, and ask how many rounds each number comes from.
About the examples: every screenshot comes from a Claude chat connected to kogni in November 2026, except the watch room page, which is VOD LAB on kogni.pro. Aurora Five is a demo team; its players and its opponents are fictional. Pro data comes from VCT 2026 broadcasts. kogni isn't endorsed by Riot Games and isn't affiliated with Anthropic.