How to balance a game with data: win rates and pick rates
By Mike Dalton ·
A game is balanced when no option is so strong that players feel they must pick it, and none is so weak that nobody should. Characters, decks, maps, and starting sides are all options. This guide covers how to measure balance with match data: which numbers to track, how to read them together, how many matches you need before a number means something, and how to check that a patch worked.
Which numbers measure game balance?
Three numbers cover most balance questions:
- Win rate: the share of games an option wins. For a character, it is wins divided by games played with that character.
- Pick rate: the share of games where an option appears. Some tools call it play rate or play share.
- Games played: the number of games behind each win rate and pick rate. It tells you how much to trust the other two.
To calculate a win rate, record each game's result as 1 for a win and 0 for a loss. The average of those values is the win rate. Most analytics tools can calculate these numbers from events you send; the comparison of game analytics tools covers which ones build the reports for you.
What win rate counts as balanced?
The goal is not exactly 50%. Small differences are normal, and some options are meant to suit certain players or situations. A common approach is to set a target band around 50%, such as 45% to 55%, and look closely at anything outside it.
The right width depends on your game. A game with many options, or options that take a long time to learn, can accept a wider band. A game with a few options that everyone plays may need a narrower one.
Read win rate and pick rate together
A win rate on its own can mislead. Pick rate shows how much an option affects the games people actually play.
| Win rate | Pick rate | What it often means |
|---|---|---|
| High | High | Probably too strong, and players have noticed. Change it first. |
| High | Low | Strong for the few players who choose it, or a counter to popular picks. Watch it before you change it. |
| Low | High | Popular but weak, or hard to learn. Check whether experienced players do better with it. |
| Low | Low | Probably too weak. |
In our map balance example, attackers won 68.1% of matches on a map named Foundry, and Foundry was 20.4% of all matches. A high win rate on a map that appears in one match out of five affects far more games than the same win rate on a map players rarely see.
How many matches do you need?
Every win rate has a margin of error that shrinks as more games are played. For a win rate near 50%, the 95% margin of error is about 1.96 × √(0.25 ÷ games played):
| Games played | Margin of error |
|---|---|
| 100 | ± 9.8 points |
| 500 | ± 4.4 points |
| 1,000 | ± 3.1 points |
| 2,000 | ± 2.2 points |
With 100 games, a measured win rate of 55% is consistent with a true win rate anywhere from about 45% to 65%. Treat small gaps from 50% as noise until you have more games, and act first on large gaps.
When you compare two versions, the margin of error on the change is larger, about 1.4 times the margin of each version's win rate. In our fighting game example, each character has about 100 games per version, so a change has to be larger than about 14 points to stand out from noise. A character named Volt dropped 14.8 points after a patch, just past that line, while changes of a few points mean little at that sample size.
Exclude results that are always 50%
Some games always produce one win and one loss for the same option, which pulls its win rate toward 50% and hides real differences:
- Mirror matches: when a character plays against itself, that character wins and loses the same game. Leave mirror matches out of character win rates.
- Team results recorded per player: in a team game with equal team sizes, one team wins every match, so the average win value across all players is 50% on every map. Record results that describe the whole match, such as which side won, on the match instead.
Split the data before you trust it
An option can look balanced overall and still dominate in one part of your game. Split your reports where it matters:
- By game version: always. Mixing versions blends the old balance with the new one.
- By skill tier: if your game has ranks or ratings. An option that is hard to learn can be weak for new players and strong for experienced ones.
- By side or turn order: if your game has them. The first player, or one side of a map, can have an advantage on its own.
Check that a patch worked
After a balance patch, compare the same report for the version before and the version after. In our three example games:
- Characters: a character named Volt dropped 14.8 points to a 49.0% win rate, and one named Frost rose 13.1 points to 48.9%. See how to track character win rates in a fighting game.
- Maps: the attack win rate on Foundry dropped 17.2 points to 50.9%, and every map ended up between 47.6% and 52.5%. See how to compare win rates across maps.
- Deck archetypes: Ember Dragon decks went from a 61.9% win rate to 51.4%, and Stone Golem decks from 40.5% to 50.4%. See how to group card game decks into archetypes.
These are example games we set up to show the method, not released games. In each one, Game Stats AI reports show every value next to its change from the previous version, so you can see the effect of a patch in one view.
What data cannot tell you
Numbers show what is happening, not why. They cannot tell you whether a matchup feels unfair, whether an option is fun, or whether players avoid a character because of how it looks. Use win rates and pick rates to find where to look, then use player feedback and playtesting to decide what to change.
Frequently asked questions
What is game balance?
Game balance is how evenly the options in a game perform. In a balanced game, no character, deck, map, or side wins so often that players feel forced to choose it, and none wins so rarely that choosing it is a mistake.
What is a good win rate for a character?
Close to 50%, within a band that fits your game, such as 45% to 55%. Look at games played as well: with only 100 games, a win rate can be 10 points away from its true value.
How many games do you need to measure win rate?
It depends on how small a difference you want to detect. With 100 games the margin of error is about 10 points, with 1,000 games about 3 points, and with 2,000 games about 2 points.
What is the difference between pick rate and win rate?
Win rate is how often an option wins when it is played. Pick rate is how often it is played at all. Read them together: a high win rate matters most when the pick rate is also high.
