Slippage is the gap between the price you intended and the price you got. Every AI execution model that reports slippage is also, quietly, reporting an assumption about how size moves price. Reading the assumption is more useful than reading the slippage number.
The three impact shapes you meet most often:
- Linear impact — cost grows in proportion to order size. Simple, and fine for small orders, but it overstates impact for large ones and ignores the diminishing-return shape that real markets tend to show.
- Square-root impact — cost grows with the square root of size. Closer to what is observed in many liquid venues, but it is a rule of thumb, not a law, and it breaks down at the extremes of size and volatility.
- Venue-fitted impact — coefficients fit on recorded fills for a specific venue and size band. The most honest in principle, and the most fragile in practice: the fit is only as good as the regime it was calibrated on.
The slippage number a model reports depends on which of these it uses. A linear-impact model will tend to report higher cost on large orders than a square-root model on the same order — not because one model is better, but because they are answering a slightly different question.
So the reading order is: identify the impact shape, find the calibration range, and only then look at the slippage figure. A slippage number without its impact model attached is a headline without a story.
None of this is a verdict on whether AI “works” for execution. It is a reading habit. The point is to know which question the model is actually answering before deciding whether the answer is useful.