The draft is the only part of a ranked match you can get right before it starts. It is also the part most players run on autopilot: six bans, three picks, twenty seconds a slot, and whatever brawler feels comfortable today.
This guide is about doing it deliberately. It covers what makes a ban worth spending, why the draft format changes completely between Diamond and Mythic, what your seat in the pick order is actually worth, and how to build a comp that has a plan instead of three good brawlers. Everything here stands on its own — you do not need a tool to use any of it.
Less than draft-obsessed players think, and more than casual players think.
Ranked matchmaking works hard to make teams even. Across the collected ranked match data, the first team wins about 51% of the time — the format starts you at a coin flip on purpose. When a win-probability model built from 1,059,778 ranked matches is asked to predict the winner from the final six brawlers and the map alone, it gets 58.8% accuracy and 0.625 AUC on a 158,966-match held-out split. AUC 0.5 is a coin flip; 1.0 is perfect. So the composition of both teams, by itself, carries a real but minority share of the result.
The practical reading: a good draft does not win you a game you would otherwise lose. It shifts the odds a few points, every single game, for free. Over a climb, a few points compounds. Over one match, it does not save bad positioning.
So draft like you are collecting a small tax, not like you are solving the match. And when you lose after a good draft, that is normal — do not throw out a correct process because of one result.
From Diamond upward, the draft opens with six bans, three from each team, before anyone picks. Below Diamond there is no ban phase at all — you go straight to picking. Six bans out of a roster of a hundred-plus brawlers is not enough to make the map safe. It is only enough to remove the specific problems.
The mistake is banning what is scary in the abstract. A ban is only worth something if the brawler would actually have been picked. If you ban a brawler with a monstrous win rate that nobody ever brings to this map, you have spent a third of your team's ban budget on nothing, and the enemy still gets their real first pick.
So a good ban target needs two properties at once: it has to be strong on this specific map, and it has to be likely to appear. That is the intuition the assistant's ban scoring encodes directly — a threat score of roughly 85% map win rate plus 15% how contested the brawler is on that map, where contest saturates once a brawler is showing up in about a third of games there. Strength leads; how often it is actually picked breaks the ties between similarly strong options.
Two more things worth internalizing about win rates, because they change what "strong" means. First, small samples lie. A brawler at 80% over five games on a map is not an 80% brawler; every rate in the tool is pulled toward 50% with 20 pseudo-games of prior, which turns that 80% into something near 55% until real evidence accumulates. Do the same in your head — treat a spectacular number with a tiny sample as ordinary. Second, the meta moves. Match data is weighted so it fades to half influence every three weeks, which means a brawler nerfed last patch stops dominating the numbers on its own. If you are working from a tier list, check when it was written.
One honest caveat: the official Brawl Stars API never exposes what was banned, only the final picked teams. So ban value here is inferred from win rate and pick rate, not measured from real ban data. Nobody has real ban data.
This is the single biggest structural fact about ranked drafting, and most players never think about it explicitly: the draft format is not the same at every rank.
At Bronze, Silver, Gold and Diamond, you draft blind. There is a ban phase, and then you pick your own three brawlers without ever seeing the enemy's picks. No snake, no reveals, no counter-picking. The enemy is drafting at the same time, in the dark, exactly like you.
From Mythic upward — Mythic, Legendary, Masters, Pro — you get the full draft: a 1-2-2-1 snake where picks are revealed as they happen, and countering becomes possible.
These demand different thinking. In a blind draft, counter-picking is not a weak strategy, it is an unavailable one. Any reasoning that starts "I'll wait and see what they take" is wasted, because you will never see it. Your last pick has exactly as much information as your first. What you can control is entirely internal: how well your three brawlers fit the map, and how well they work together.
So in blind pick, optimize your own comp and assume the worst about theirs. The enemy is looking at the same map data you are, so assume they took the map's strongest brawlers — specifically, the strong ones you did not ban. Draft something that beats the map's default answer, not something that beats a specific brawler you are hoping to see. And avoid brawlers that are excellent when unopposed and useless against one specific counter, because in the dark you cannot dodge that matchup.
This is also why the assistant's counter signal simply switches off in blind mode: with no revealed enemy, there is nothing to counter, and the score renormalizes over map fit, the model, internal synergy, and role fit. It is not that countering stopped mattering — it is that the information does not exist.
The 1-2-2-1 snake means the first-pick team picks at slots 1, 4 and 5, and the second-pick team picks at slots 2, 3 and 6. That produces four genuinely different jobs.
First pick (slot 1). You get map priority — the single most contested brawler on the map is yours if you want it. You pay for that with zero information and full exposure: the enemy's next two picks can be aimed straight at you, and you have no response, because your next turn is not until slots 4 and 5. So a first pick should be strong and hard to punish, not a narrow specialist. Take the brawler the map is built around, the one that stays useful regardless of what shows up opposite it. Never spend first pick on something whose whole value is beating a brawler that has not been picked yet.
Slots 2 and 3 (second-pick team's double). You have seen exactly one enemy brawler and you get to fill two slots at once. This is the best shaping seat in the draft. Use one pick to answer their first pick and one to establish what your team actually wants to do. Spending both on countering a single brawler is the classic overreaction — you have removed one threat and told the enemy everything about your comp, and they still have three picks left.
Slots 4 and 5 (first-pick team's double). Two enemy brawlers are on the board and this is your last chance to shape the comp. Fill your holes now — frontline, range, whatever the objective needs — because you do not pick again. And plan for the fact that the enemy picks last: assume their final pick will be the best remaining answer to your comp, and try to finish in a state where no single brawler wrecks you.
Slot 6 (last pick). Maximum information: you see all three enemy brawlers and two of your own. This is the only seat where a hard specialist is safe, because nothing gets picked after you and nobody can counter your counter. The cost is that the map's best options are usually gone, so you are choosing from leftovers. Pick the answer to their biggest threat, or the piece your comp is missing, whichever is more urgent.
The assistant's model signal folds some of this in automatically — it scores the board exactly as it stands, so a pick's number already reflects how real drafts that reached that position tended to end. What it cannot do is anticipate a specific opponent's response, so the reveal-order reasoning above stays your job: before you lock a pick, ask what you would take against your own comp if you had the next turn.
Brawl Stars sorts brawlers into seven classes: Tank, Assassin, Controller, Marksman, Support, Damage Dealer and Artillery. Memorizing the labels is not the point. Two axes cut across them and explain most bad comps.
The first is frontline: Tanks and Assassins. Frontline is how you contest space head-on — win the ball, hold the mine, take the zone, force the fight. A team with no frontline cannot take ground it does not already have. It can only poke, kite, and hope the enemy walks into it. Against a mode that requires you to physically occupy something, that is often just losing slowly.
The second is range: Marksmen, Controllers and Artillery. Range is how you punish, zone and answer poke. A team with no long range gets walled out by throwers, loses every trade it cannot close, and has to run a gauntlet to do anything.
Support and Damage Dealer flex between the two, which is why they show up in almost every comp.
The other failure mode is redundancy. Three of any one class is one-dimensional — you have one plan and one weakness, tripled. Two of the same class is often fine but worth a second look, especially two Tanks, two Marksmen or two Artillery, where the duplicate usually adds the same strength you already had and the same hole you already had. The engine's role-fit score reflects this directly: each additional copy of a class already on your team docks that class's fit.
Fit is also mode-dependent, and the engine's class preferences are a decent summary of conventional wisdom. Heist rewards Damage Dealers most, then Marksmen. Bounty and Knockout reward Marksmen and Controllers. Brawl Ball rewards Tanks and Assassins. Gem Grab rewards Controllers and Support with a Tank to contest the middle. Hot Zone rewards Controllers above everything.
The test for a finished comp is not "do I have one of each." It is: can you say the win condition in one sentence? Two frontline and no range is an aggressive comp — force fights, close distance, snowball early picks. Two or more range and no frontline is a control comp — poke, zone, punish over-extension, never brawl. A mix is the standard plan: frontline engages, backline follows up. If you cannot name which of those you are, you drafted three brawlers, not a team.
One concrete cross-comp rule worth keeping: if the enemy has two or more Tanks and you have no Marksman, you are in trouble, and a Marksman is the answer. That specific pattern is common enough that the tool flags it as a warning.
Ranked runs six modes: Gem Grab, Brawl Ball, Knockout, Hot Zone, Heist and Bounty. The draft does not carry equal weight across them.
In a held-out ablation over 40,208 ranked matches, the same draft signals were measurably more predictive in some modes than others. Heist was the highest at 0.628 AUC, then Hot Zone at 0.593, Brawl Ball 0.580, Gem Grab 0.578, Bounty 0.567, and Knockout lowest at 0.563.
The practical translation: in Heist, the draft decides more of the game than anywhere else. A Heist comp with no burst damage genuinely cannot complete the objective, and no amount of good play fixes that. Spend your attention and your bans there. At the other end, Knockout and Bounty drafts are more interchangeable — rounds get decided by who takes the first bad fight and who gets caught out of position. If you are losing Knockout, the draft is probably not why.
None of these numbers are large in absolute terms. Even the most draft-decided mode is well short of predictable. The ordering is the useful part, not the magnitude.
Almost all draft losses come from a short list.
Comfort-picking into a bad map. Your best brawler is your best brawler on average, not here. A map has geometry — walls, lanes, bush, open ground — and a brawler that is excellent in the abstract can be nearly unplayable on a specific layout. The honest version of this instinct is worth respecting, though: a brawler you have played a hundred times is genuinely better in your hands than one you have not, which is why the tool weights your own history with a brawler by how much of it there is. A thin personal record nudges. A deep one speaks up. But neither overrides a map that does not want the brawler.
Ignoring the enemy comp when you can actually see it. In snake ranks, information is handed to you and players routinely throw it away. In the tuned signal weights, the counter term sits at 0.20 — close behind map win rate at 0.25, and four times synergy's 0.05. That ordering came out of the held-out ablation, which found counter was the strongest head-to-head signal and had been under-weighted. If you can see two enemy brawlers and your pick reasoning does not mention either of them, you are drafting blind by choice.
Over-valuing raw win rate over fit. A brawler at a big global win rate on a map it does not suit is worse than a modest brawler that fits the layout and your comp. And raw rates on thin samples are mostly noise — that is why every rate is smoothed toward 50% and every recommendation carries a sample-size confidence. Check the confidence before you trust the number.
Banning what is scary rather than what is likely. Covered above, but it is the most common wasted resource in the whole draft.
Three brawlers who all want to do the same thing. Usually the result of three players each picking their own comfort brawler with no reference to each other. It is the one mistake that is entirely free to avoid.
Forgetting the enemy picks last. If you are on the first-pick team, your comp is finished before theirs is. Draft as if their final pick will be the worst possible answer for you, because a good opponent will make it exactly that.
Brawl Draft is unofficial fan content. It is not affiliated with, endorsed by, or sponsored by Supercell.