What Matters When You Pick a Champion?

A guide to reading the draft, from first pick to last.

Two rows of carved stone draft pieces on wet slate, some still under cloth, with one, two and three gold stars set beside them.

You have two comfort picks. One usually wins lane against the enemy top laner. The other can protect your carry when the enemy jungler dives. Which advantage is more likely to matter after the first turret falls?

We looked at the drafts of more than 5.3 million ranked Solo/Duo games to see which factors showed up in wins. Then we checked those patterns against 174,000 newer games that the first study never saw. One finding matters even if you never look up a win rate: who else is in the game changes what makes a good pick.

What the stars mean

The stars are a guide to what deserves your attention when comparing champions you play. They describe the importance of each factor, not your chance of winning.

  • ⭐⭐⭐ High priority: Start here when choosing between picks.
  • ⭐⭐ Worth considering: Can separate two otherwise good options.
  • ⭐ A small edge: Usually a tiebreaker.
  • – Doesn’t apply: Only the bot laner and support share a lane partner.
  • Strength in your role: How often the champion wins at your rank in that role.
  • The rest of the enemy team: How the pick fares against opponents outside your own role.
  • Your teammates: How often the pick wins alongside the allies already chosen.
  • Your direct opponent: How often the champion wins the game against the enemy in the same role.
  • Your bot-lane partner: How often this bot laner and support win when they play together.
  • Early gold: Who tends to be ahead in gold at 15 minutes in the matchup.

What changes from first pick to last?

Picking first, you know your role and little about your opponents. Look for a champion you can still play comfortably into a difficult matchup. A strong average win rate can hide a counter you would hate to face.

Later, you can answer a different question. Suppose your support has already picked an engage champion. Which of your usual bot laners can follow when they go in? Now check the other side. Can that same champion keep attacking if the enemy dives?

What to look at as picks come in

Read across a row to see when that information becomes useful.

What to consider Opening picks Mid-draft Late picks
Strength in your role⭐⭐⭐⭐⭐⭐
How safe it is to pick blind⭐⭐⭐⭐⭐⭐
Other enemy champions⭐⭐⭐⭐⭐⭐
Fit with your team⭐⭐⭐⭐⭐⭐
Direct matchup or bot duo⭐⭐⭐⭐⭐
Gold at 15 minutes⭐⭐⭐

How much does winning lane matter?

A gold lead at 15 minutes tells you who tends to come out of lane ahead. Matchup win rate answers a different question. It includes the games where a champion falls behind early and still gets to decide the last fight.

What to compare in your role

The stars summarize the research in broad terms. Read across your role to compare factors.

RoleOther enemiesTeam fitDirect matchupBot duoEarly gold
Top⭐⭐⭐⭐⭐⭐⭐⭐–⭐⭐
Jungle⭐⭐⭐⭐⭐⭐⭐⭐–⭐
Mid⭐⭐⭐⭐⭐⭐⭐⭐–⭐
Bot⭐⭐⭐⭐⭐⭐⭐⭐⭐⭐⭐
Support⭐⭐⭐⭐⭐⭐⭐⭐⭐⭐⭐

In top lane, early gold deserves a closer look. In bot lane, check your partner as well as your opponent. If you play an immobile carry, the enemy jungler who can reach you at dragon may matter more than the marksman you can beat in lane.

How we checked the numbers

Big numbers can still mislead. A lucky streak, a quirk of one rank or a test that already knows the answer can all make a pick look better than it is. Here is how we guarded against each one.

Short streaks don't count for much

Would you trust a counter that won eight of its first ten games? A streak like that can happen by luck. Until a matchup has enough games behind it, we pull its result toward the average instead of taking it at face value.

Empirical Bayes shrinkage

Ranks share most patterns, not all

Most of what makes a pick good holds from Iron to Challenger, but some matchups play out differently at each rank. This approach learns both at once, so a pairing with plenty of games overall but only a few in your rank still gets a sensible rating.

Mixed-effects models (GPBoost)

No peeking at the result

To test our predictions, we always held some games back and checked against those. All ten players from a match stayed on the same side of that split. If one player's game helped build a rating and a teammate's game was used to test it, the test would already know who won.

Grouped cross-validation

Checked at every stage of the draft

On the newer games, we also tested the advice with only the opening picks visible. A change that looked better once both teams were revealed gave worse first-pick advice. That's why we check both situations before trusting a change.

Validation by draft stage on newer matches

Further reading

  1. Sigrist, F. (2022). Gaussian Process Boosting. Journal of Machine Learning Research (JMLR), 23(232), 1–46.
  2. Efron, B., & Morris, C. (1975). Data Analysis Using Stein's Estimator and Its Generalizations. Journal of the American Statistical Association, 70(350), 311–319.
  3. Scikit-learn Developers. Cross-validation: Evaluating Estimator Performance (Grouped & Time-Series Splits).

More on choosing a champion