progressWhen ICM misleads you
What the model leaves out, and how far the answer can move because of it.
ICM is the best simple tool for payout-aware decisions, and that is why it is easy to trust too much. It is a model with clear assumptions. Knowing them tells you when to follow the number and when to treat it as a rough guide.
1. You do not know the payout ladder exactly
Many micro and low stakes tournaments publish the ladder only after registration closes, and during late registration it keeps changing. Here is how far the answer moves for one call on the final table from our risk premium guide (you, 1600, call the 3000 shove) when only the ladder changes:
| Ladder | Equity needed (ICM) | Risk premium (points) |
|---|---|---|
| Top-heavy 38/24/15/9/6/4/2/1/1 | 60.8% | 14.6 |
| Typical 28/19/13/10/8/7/6/5/4 | 63.9% | 17.7 |
| Flat 20/16/13/11/10/9/8/7/6 | 69.2% | 23.0 |
The spread is about 8.4 points of required equity between a top-heavy and a flat ladder. Enter the real payouts whenever you can; the calculator and the trainer accept your own.
2. Counting your own stack wrongly
If you misjudge your own stack by 20% (counting chips wrongly, or forgetting what you have already posted):
| Your stack | Equity needed (chips) | Equity needed (ICM) | Risk premium (points) |
|---|---|---|---|
| 1280 | 45.3% | 62.5% | 17.1 |
| 1600 (base) | 46.2% | 63.9% | 17.7 |
| 1920 | 46.8% | 65.0% | 18.2 |
A 20% counting error moves the requirement by only about one to one and a half points, much less than a wrong ladder (about 8.4 points above). Counting is still worth the effort in close spots, because it also changes how many chips you risk.
3. Bounties change everything
Plain ICM knows nothing about bounties. A rough static model adds the bounty to the winnings of a player who covers the opponent. For the player with 1600 calling the 1000 stack at the same final table:
| Bounty on the opponent | Equity needed in chips | Equity needed (ICM + bounty) | Risk premium (points) |
|---|---|---|---|
| No bounty | 44.2% | 51.9% | 7.7 |
| 2% of the prize pool | 44.2% | 41.0% | -3.2 |
| 5% of the prize pool | 44.2% | 31.1% | -13.1 |
| 10% of the prize pool | 44.2% | 22.2% | -22.0 |
These bounty figures are an approximation from a simple static model, not a full progressive bounty solution. The direction is still clear: a bounty worth 5% of the prize pool turns an ICM fold into a clear call, and the premium can even go below zero.
4. The model has no skill and no future
ICM treats every chip as equally likely to win. It does not know that you may be the best player at the table, that the blinds go up soon, or that the next pay jump is far away. In one large study of tournament results the model underestimates the results of players with larger stacks and overestimates those who are short-stacked (see the note below). So treat close decisions as close, and be careful with long pushes of marginal hands.
5. Our preflop book is chip EV
The solution book behind the ranges and the daily spot is chip EV with an ante, not ICM. It is a good baseline for early and middle stages; near the money, use ICM tools. The book limits page lists what is known to differ.
Practise close spots in the ICM trainer, where the risk premium and the model limit are shown for every hand.
A practical rule
- Ladder known and stacks counted: if your hand is more than three points away from the line, trust the number.
- Within two or three points: reads and the blind spots of the model decide it.
- Ladder unknown: allow about half of the spread above either side of your estimate.
- With bounties: use a bounty-aware estimate and call it approximate.
Keep exploring
- PKO calculator: how much a bounty lowers the equity you need
- Check the bubble factor of any all-in
- Browse the push/fold charts for 3 to 15 BB
Free 7-day ICM course by e-mail
One short lesson a day, each with a worked example in exact numbers: chips vs. money, the bubble factor, who feels the pressure, payout structures, satellites. No account needed.
Related guides
Questions
When is ICM most reliable?
When the payout structure is known and fixed, the stacks are counted correctly, there are no bounties, and the decision is not close. Near the money with a known ladder it is a strong guide.
Can ICM be wrong for good players?
Yes. ICM assumes equal skill and ignores future blinds, so a strong player may be able to take slightly more risk than the model allows, and a weak player slightly less. Studies of real results also find a bias with big and small stacks.
How do bounties change ICM?
A bounty adds value to winning an all-in against a player you cover. In our approximate static model a bounty worth 5% of the prize pool lowers the required equity in the example from about 52% to about 31%.
Share this page
progressGTO guides use our own solution book (v17d, 8-max tournaments with a 12.5% ante from each player, chip EV): an approximation of equilibrium with a modelled postflop, not an exact solution. Numbers are computed from the book and from the exact ICM model.