Definition
The percentage of promotional conversions that represent genuinely incremental purchases, as distinct from pulled-forward demand or purchases redirected from full-price channels, measuring whether promotional AI is creating new demand or redistributing existing demand at lower margin.
The Measurement Gap
The Gap
A promotion runs for a week. Revenue spikes 40%. The promotional AI reports a 40% lift. The marketing team celebrates. But what happened in the two weeks after the promotion ended? Did revenue drop below the pre-promotion baseline? If it did, some of that promotional "lift" was not new demand -- it was demand pulled forward from future periods. Customers who would have bought next week bought this week instead, at a lower price.
Promotional lift, the standard metric for evaluating promotions, measures the revenue increase during the promotional window compared to a baseline period. It cannot distinguish between genuinely new purchases and demand that was merely accelerated or redirected. Conversion rate improves during promotions because price dropped, but a higher conversion rate on discounted revenue is not the same as incremental value creation. Revenue attribution assigns credit for conversions during the promotional window but does not track whether those conversions represent demand that was already forming.
The structural problem is that promotional measurement systems are built to look at promotional windows in isolation. They measure the spike but not the valley that follows. A promotion with 40% lift and a 30% valley in the following two weeks produced only 10% genuine incrementality -- but legacy metrics reported 40%. When promotional AI optimizes for lift, it is optimizing for a metric that systematically overstates its own contribution.
The Decision
We propose PIR to answer: Of the "lift" our promotional AI claims, how much is real new demand versus demand that would have occurred anyway? This metric would require holdout-based measurement with extended observation windows that capture both the promotional spike and the post-promotional valley, providing the true incrementality rate rather than the inflated lift number.
Formula
Plain Language
Run the promotion to a test group while holding a control group at regular price. Measure revenue for both groups across an extended window: the promotional period plus 2-4 weeks after the promotion ends. The difference between the test group's extended-window revenue and the control group's extended-window revenue is the genuine incremental revenue. Divide that by the revenue the promotional system attributed to the promotion. A result of 100% means all attributed lift was genuine. A result of 50% means half the "lift" was redistribution. A negative result means the promotion destroyed net revenue -- the valley exceeded the spike.
Notation
PIR = Revenue_incremental / Revenue_attributed_to_promo x 100
Where:
Revenue_incremental = Revenue_promo_group_extended_window - Revenue_control_group_extended_window
Variables
| Variable | Definition | Units | Source System |
|---|---|---|---|
| Revenue_attributed_to_promo | Revenue credited to the promotion by the standard promotional lift calculation (lift percentage x promoted revenue) | Dollars | Promotional Platform |
| Revenue_incremental | True incremental revenue measured via holdout group across the extended observation window | Dollars | Experimentation Platform (Derived) |
| Revenue_promo_group_extended_window | Total revenue from the promoted group measured across the promo period plus 2-4 weeks post-promo | Dollars | E-commerce / CRM / Billing |
| Revenue_control_group_extended_window | Total revenue from the holdout control group measured across the same extended window | Dollars | Experimentation Platform |
Proposed formula, pending validation pilots.