A promotion has two effects, and standard promo math measures one of them. The visible effect is incremental volume: units sold during the window above the baseline you would have sold anyway. The invisible effect is what the discounted price does to the number a customer carries in their head as what this thing costs. Pricing researchers call that number an internal reference price, and it updates from what people have recently paid rather than from what the list price says. Once it moves down, full price stops reading as the price and starts reading as a markup.

Picture a subscription business that runs a thirty-percent-off promotion every quarter to hit its number. In the first year the promo weeks look excellent, and the analysis credits them with every signup above baseline. By the third year the baseline itself has moved: signups in non-promo weeks have thinned, because the segment that would have paid full price has learned that waiting eleven weeks is worth thirty percent. The promotion still posts strong incremental volume, and it is now largely harvesting customers whose waiting habit it created.

The measurement error here is structural rather than careless. Incremental volume is computed against a baseline that the promotions themselves have been depressing, so the worse the damage, the better the promotion looks. Any metric comparing a period to its own recent history inherits this problem whenever the intervention has been reshaping that history. Escaping it takes a baseline the promotion cannot touch: a holdout region, a holdout cohort, or a pre-period long enough to predate the habit. The alternative is a number that improves precisely as the underlying business gets worse.