VCT 2025 Player Analytics

TL;DR: An interactive Power BI dashboard that analyzes pro VALORANT play across the 2025 VALORANT Champions Tour: which agents ran the meta, which players performed best, and how teams build their agent pools. I built it to learn Power BI end to end. Most of the work turned out to be making messy data trustworthy before charting any of it.

Overview

VCT 2025 Player Analytics is a three-page dashboard built on 9,800+ pro match records from 15 tournaments, from the four regional Kickoffs through Masters Bangkok, Masters Toronto and Champions 2025. Anyone can pick a tournament or a team and get answers for themselves, without needing an analyst to pull the numbers.

• Role: Solo analyst: data cleaning, modeling, measures and report design
• Stack: Power BI (web), DAX, Power Query, Python (pandas)
• Data: Valorant Champion Tour 2021–2026 Data on Kaggle, 2025 player stats
• Code and README: GitHub: VALORANT-VCT2025-Player-Analytics
• Status: Complete

Why I built it

More and more of the internships I want, especially in energy and enterprise tech, ask for Power BI and dashboard experience, and I had none. I wanted to close that gap with a real project rather than a tutorial, so I picked a subject I already know well. I play and follow VALORANT, which meant I could tell when a number looked wrong. That turned out to matter a lot.

The questions

I framed the dashboard around three questions a coach, analyst or fan would actually ask:

1. Meta: Which agents dominated pro play in 2025, and did that change by tournament or region?
2. Players: Who performed best once you account for how much they played?
3. Teams: How does a given team play, and how does it compare to the league?

Cleaning the data

The raw file looked ready to use. It wasn't. Three problems would have produced confident, wrong numbers:

Problem	What it would have broken	Fix
Rollup rows mixed in with detail rows: "All Stages" totals and multi-agent rows like "astra, omen"	Every stat double- or triple-counted	Removed both, going from 17,996 rows to 9,805 at one consistent grain: one player, on one agent, in one match
About 6,200 clutch records like 1/3 auto-converted to dates (03-Jan) by Excel	Clutch stats unusable for a third of the data	Decoded them back into wins and attempts, then validated against the dataset's own clutch %. All 2,654 checkable rows matched.
Percentages stored as text ("44%")	Couldn't be aggregated or formatted	Converted to decimals

I did the cleaning in Python with pandas so every step is repeatable and documented in the repo, then loaded the clean file into Power BI.

Modeling and measures

The most important decision in the model was to never average averages. The source gives per-match stats like ACS (average combat score). Averaging those across matches would treat a 13-round map the same as a 26-round one. Instead, every DAX measure rebuilds its rate from raw counts:

• ACS: weighted by rounds played
• K/D, KPR: from total kills, deaths and rounds
• Entry success: first kills ÷ (first kills + first deaths)
• Agent pick %: an agent's rounds ÷ all rounds in the current filter, so it adjusts to whatever tournament or team is selected
• League benchmarks: the same measures with the team filter removed, so a team can be compared to the average

I also added an agent role column (Duelist, Initiator, Controller, Sentinel) in Power Query so the meta could be viewed by role.

The dashboard

Page	What it answers
Agent Meta	Headline cards (players, agents played, top agent and its pick rate) and pick rate for every agent, colored by role and filterable by tournament
Player Leaderboard	Top performers by round-weighted ACS (500+ rounds), plus a scatter chart with a trend line comparing ACS to entry success
Team Scouting	Pick a team to see its stats next to the league average, its agent pool under a title that updates with the team, and each player's main agents

What the data showed

• Omen was the backbone of the 2025 meta, at about 13% of all player-rounds, ahead of Viper and Sova at about 8% each. He was the most-played agent in every region and at international events.
• The meta was remarkably uniform across regions. All four leagues shared nearly the same top agents.
• Winning opening duels goes with impact. Across 211 regular players, entry success and ACS are strongly correlated (r ≈ 0.61).
• But there's more than one way to be elite. Among the top 25 players, that link almost disappears. ZmjjKK posted the highest ACS with below-average entry success, while aspas and Kai were the best opening-duel players.

Judgment calls

A few decisions changed what the dashboard says more than any chart did:

• Minimum sample size. Without a floor, players with one great map topped the leaderboard. I set 500 rounds as the cutoff. I tried 1,800 first, but that only left players from teams that went deep at international events, which ranked playtime more than skill.
• Choosing a chart that says something. My first scatter chart used bubble sizes and was an unreadable blob. Narrowing it to the top 25 players removed the trend entirely. Showing all regular players with a trend line told the real story.
• Admitting gaps. The source has no clutch data for China's leagues, so I left clutch % out of player comparisons rather than show misleading numbers. The README lists every limitation like this.

What I'd do next

1. Agent icons. Replace agent names with their portraits by pulling icons from a community VALORANT API in Power Query. Just to make it more visually appealing :)
2. Map-level data. Add which map each match was on, so teams can be scouted by map, not just overall.
3. Automated refresh. Connect to a live source so the dashboard updates during a season, the way a real business dashboard would.

Reflection

I went in expecting to simply learn a tool and came out understanding data analytics workflows a lot better. The Power BI skills came quickly. The harder lessons were the ones that apply to any data work: check the data before trusting it, define every number so it means what people assume it means, and design each page around a question someone actually has. The validation step is the part I'm proudest of! Catching thousands of corrupted values before they reached a chart is the difference between a dashboard that looks right and one that is right.