Card-linked offers
The Card-Linked Offer That Targeted The People Already Ready To Spend
A realistic measurement scenario showing how a useful advertising channel can still produce a misleading read when the comparison group is wrong.
Archetype: Card-linked offer network using bank transaction histories to target merchant promotions
Bias mechanism: Targeting uses recent category behavior and purchase histories. The measured treatment group therefore contains shoppers with a higher baseline chance of buying even without the offer.
Business model pressure
Merchants or advertisers pay to reach consumers through bank or card-linked promotional inventory, often with value framed around measurable purchases.
Advertiser proof claim
The platform can show that activated users spent more at the merchant than a broad comparison group. The question is whether transaction-history targeting selected likely buyers before the promotion.
Statistical result
| Metric | Naive read | Stratified read | Modeled benchmark |
|---|---|---|---|
| Conversion lift | 14.8 pts | 1.6 pts | 0.9 pts |
| Incremental conversions | 8,962 | 996 | 544 |
| Incremental revenue | $483,950 | $53,789 | $29,353 |
| ROAS after media cost | 1.86x | 0.21x | 0.11x |
The naive analysis is 9.0x larger than the stratified estimate in this worked example.
The advertiser-facing story
The reported result shows a large spending gap between activated offer users and non-activated consumers. The gap looks like merchant lift when it may partly be a targeting artifact.
What broke
Prior purchase behavior is both a targeting input and a predictor of future purchase. If the control group is not drawn from the same eligible population and propensity strata, the measured effect inherits the targeting model's selection.
Better design
Randomize offers inside the eligible audience, pre-register the outcome window, separate new and existing customers, and report incremental spend after removing baseline spend differences.
Propensity-strata audit
The adjusted estimate compares treated and control users within similar treatment-propensity strata. That does not replace a randomized holdout, but it shows how much of the naive result was carried by who entered treatment.
| Propensity stratum | Users | Treated conversion | Control conversion | Within-stratum lift |
|---|---|---|---|---|
| 1 | 447 | 2.4% | 5.2% | -2.8 pts |
| 2 | 9,655 | 9.0% | 7.6% | 1.4 pts |
| 3 | 14,035 | 13.9% | 11.0% | 2.9 pts |
| 4 | 10,828 | 16.4% | 16.1% | 0.3 pts |
| 5 | 8,555 | 21.6% | 19.4% | 2.2 pts |
| 6 | 10,497 | 25.7% | 24.4% | 1.3 pts |
| 7 | 16,621 | 31.9% | 31.4% | 0.5 pts |
| 8 | 22,146 | 40.8% | 38.8% | 2.0 pts |
| 9 | 16,109 | 52.4% | 50.0% | 2.3 pts |
| 10 | 1,107 | 63.8% | 61.8% | 2.0 pts |
Takeaway
A strong advertiser report should not stop at attributed conversions. It should show how the comparison group was built, whether treatment users had stronger prior intent, and how much of the result survives a better counterfactual.
Use this case in a readout review
After reading the failure mode, move from diagnosis to a better question set with these practical tools.
- Claim confidence rubricScore whether the readout supports strong, qualified, or tentative claim language.
- Measurement method selectorChoose the method that best fits the question: holdout, matched market, MMM, survey readout, or QA.
- Retail media incrementality checklistSeparate offer targeting, prior category demand, and buyer-would-have-bought-anyway risk.