ICM Explained: Why Folding Big Hands Late Can Be Correct

Every tournament player has a version of the same story. Four left, decent stack, and you pick up queens against an opponent who has been shoving relentlessly. You call, you lose, you finish fourth, and everyone at the rail tells you it was a cooler. Sometimes it was. Often it wasn’t — it was a spot where the correct answer was to slide two queens face down into the muck and go back to stealing blinds.

That sentence still gets an eye-roll from people who have spent years being told never to fold a big pair at 20 big blinds. But the roll-off between chip value and money value at a final table is not a small adjustment. It’s the difference between needing 47 percent equity and needing 65 percent, which is roughly the difference between a call being a clear snap and a call being a five-figure mistake.

The tool that measures this is the Independent Chip Model, and the thing worth understanding about it isn’t the formula. It’s where the formula bites hardest, and — just as importantly — where it stops applying.

What ICM actually converts, and why the exchange rate is rigged against you

ICM takes the stacks at the table and the remaining payouts and estimates each player’s probability of finishing in each paid position, assuming finishing order is driven purely by chip counts. It then multiplies those probabilities by the prizes and hands you a dollar figure for your stack. That’s it. No skill, no position, no reads, no blind structure.

The consequence people quote is that a chip you lose is always worth more than a chip you gain, and that’s true, but it’s worth seeing why. If you double your stack from 2 million to 4 million, you don’t double your equity, because the first prize is capped. If you lose those 2 million chips, you don’t fall to zero equity — you fall to whatever the current pay rung is worth. The upside is compressed and the downside is cushioned, but the downside is also permanent. That asymmetry is the entire game.

The gap between chip-EV and ICM-EV is usually called the risk premium, and it varies enormously by spot. On a soft bubble with flat payouts it might be two or three percent of equity. At a final table with a desperate short stack and steep jumps ahead, it can be fifteen or twenty. Treating it as one fixed adjustment — “play a bit tighter late” — is how people talk themselves into calls that cost real money.

A four-handed spot where queens are a fold

Take a concrete final table. Four players left, payouts of $100,000, $62,000, $41,000 and $28,000. Blinds 50k/100k. Stacks: the chip leader has 5.5 million, you have 2 million, another player has 1.6 million, and a fourth player is crippled at 900,000 — under ten big blinds and about to be swallowed by the blinds.

You’re in the big blind. The chip leader open-shoves and covers you. If you fold, you keep 1.9 million and your equity is roughly $55,800. If you call and win, you sit on 4 million and your equity is about $70,700. If you call and lose, you collect fourth-place money: $28,000.

Run the arithmetic on that and your breakeven equity is about 65 percent. Queens against the chip leader’s shoving range from the small blind, with 20 big blinds effective — a wide range, plenty of suited connectors and ace-rag — runs somewhere around 60 to 62 percent. In pure chips the call needs about 47 percent, so it’s a blowout snap-call. In money, it’s a fold, and not a close one. Ace-king, at roughly 55 percent against that same range, is a considerably worse call and yet it’s the hand people call with most often because “I’m not folding ace-king four-handed.”

Now delete the crippled player. Same payouts, same blinds, but the stacks are 4 million, 2 million, 2 million, 2 million. Your breakeven equity drops to about 60 percent. Still well above chip-EV, still a fold with ace-king, but five percentage points softer. The presence of one player who is likely to bust before you is doing a good part of the work, because every hand you survive while he’s still at risk is free money laddering toward you.

Where the standard advice gets you killed instead

ICM tightens your calling ranges. It loosens almost everything else, and this is the half that gets ignored.

If you’re the chip leader in that spot, you should be shoving an absurdly wide range at the 2 million stack, precisely because he can’t call. His risk premium is your license. The same logic applies to re-steals, to three-bet shoves over the player who is clearly trying to fold into a pay jump, and to relentless small-ball aggression against anyone who visibly wants the bubble to burst. The player folding queens is playing correctly, and the player forcing that fold is printing.

If you’re the short stack, ICM barely restrains you at all. You have the least to lose in dollar terms and your fold equity decays every orbit. Sub-15-big-blind play is a different discipline with its own logic, closer to the short-stack framework than to anything resembling final-table caution. Waiting for a hand while the blinds eat you is a way of guaranteeing the pay rung you’re currently standing on rather than climbing.

And there’s the timing dimension. ICM pressure at nine-handed with shallow jumps is mild. It spikes at the money bubble, spikes again at the final table bubble, and spikes hardest when the top-heavy jumps start — the $41,000 to $62,000 to $100,000 ladder in the example above. Three-handed, with one jump left and everyone deep, it starts to relax again. Heads-up it’s gone entirely: two payouts, two players, chip-EV and money-EV converge.

Satellites, bounties, and other places the model breaks

Satellites are the purest version of the model and the reason folding aces is a real, documented, correct play. When several seats of equal value are awarded, survival becomes everything, and once you’re locked to qualify you should fold even premium hands. There’s no first prize to chase. Winning a pot gains you nothing and losing one costs you the seat. Every chip above the qualifying threshold has a value of exactly zero.

Progressive knockouts run the other way. A bounty is cash that lands immediately regardless of where you finish, so calling ranges widen — sometimes dramatically — against covered opponents. A quarter of your stack in bounty value on the short stack’s head can flip a marginal ICM fold back into a call.

Two other caveats matter. ICM assumes everyone plays identically and has no edge, which is false; if you’re a substantially better player than the table, your realizable equity sits above what the model says, which argues for even more survival. And the standard Malmuth-Harville calculation systematically overstates a short stack’s chance of finishing second, which means it slightly understates how much a big stack is worth. Modern tools — HRC, ICMizer, the ICM sims now built into most solver suites — handle this with future-game simulations rather than a single static snapshot, and they’re worth running on your own final tables after the fact. The broad principle they keep confirming is that at a final table you should play tighter than chip-EV suggests, especially with a medium stack and steep jumps ahead. The endgame thinking in Kill Everyone by Lee Nelson predates the current software by well over a decade and still frames the problem correctly.

The habit worth building before you ever need it

You cannot compute ICM at the table. What you can do is walk in already knowing three things: the payout jumps in dollars, not percentages; which opponents are genuinely at risk of busting before you; and roughly where your calling threshold sits — 55 percent, 60, 65 — given those two facts.

When the shove comes and you’re staring at queens with eleven seconds on the clock, you’re not solving anything. You’re checking whether this is the spot you already identified as a fold. Most of the money in tournament poker gets decided before the cards are dealt, which is an unromantic thing to say about a game built on last-card miracles, but the payout sheet doesn’t care how you got there.

Photo by Jonathan Borba on Pexels