Card-linked offers
The Card-Linked Offer That Targeted The People Already Ready To Spend
Organization byline: Measurement Press editorial desk. This page does not list an individual author or reviewer. Read the editorial method and corrections policy.
About 9 min read
A realistic card-linked offer scenario showing how closed-loop transaction proof can overstate lift when prior purchase behavior is both a targeting input and an outcome predictor.
Simulated Worked Example. This is a constructed, synthetic, and anonymized editorial scenario using simulated data; no named advertiser, campaign, client, or observed result is asserted. The figures are for method demonstration and are not reported client results.
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, new customer reach, and closed-loop transaction reporting. The advertiser needs to know whether the offer changed behavior, not only whether the platform found shoppers whose bank histories already showed category demand.
Advertiser proof claim
The platform can show that activated users spent more at the merchant than a broad comparison group. The counterfactual question is whether transaction-history targeting selected likely buyers before the promotion, and whether the comparison group had the same eligibility, purchase recency, category affinity, loyalty status, and offer-availability path.
Primary-source boundary
These official references define the measurement concepts used in the review. They are not evidence for this fictional scenario, and they do not establish that a real card-linked offer produced the figures below.
- Conversion Lift measurement guide — distinguishes attributed conversions from treatment-versus-control incremental conversions and defines incremental value and iROAS.
- Store sales measurement guide — describes online-to-offline measurement, first-party transaction data, and the distinction between reported store-sales counts and values.
- Store-sales upload format — documents transaction-field, timing, and data-handling requirements that a real offline measurement review would need to verify.
Statistical result
| Metric | Naive read | Stratified read | Modeled benchmark |
|---|---|---|---|
| Merchant purchase lift | 14.8 pts | 1.6 pts | 0.9 pts |
| Attributed incremental purchases | 8,962 | 996 | 544 |
| Attributed merchant spend | $483,950 | $53,789 | $29,353 |
| Spend ROAS after media cost | 1.86x | 0.21x | 0.11x |
The offer-activation dashboard is 9.0x larger than the eligible-audience estimate in this worked example.
Reproduce the targeting-bias downgrade
The generator uses a fixed seed and 110,000 fictional shopper records. Re-run scripts/generate-case-studies.mjs with this scenario JSON to reproduce the displayed rates; the rounded table values follow the formulas below.
| Item | Calculation or assumption |
|---|---|
| Eligible-group split | 60,398 activated-offer records + 49,602 eligible-comparison records = 110,000 total; activated share = 60,398 / 110,000 = 54.9063%. |
| Naive purchase lift | Activated purchase rate 35.6419% - eligible-comparison purchase rate 20.8036% = 14.8383 points, displayed as 14.8 pts. |
| Propensity-adjusted lift | Weighted mean of the within-decile activated-minus-comparison rate differences = 1.6492 points, displayed as 1.6 pts; this is an adjustment diagnostic, not a randomized estimate. |
| Attributed-value sensitivity | Naive lift x 60,398 activated records = 8,962.0 purchases; x 54 USD fictional value = 483,950.36 USD and 483,950.36 USD / 260,000 USD fictional media cost = 1.8613x. |
| Adjusted-value sensitivity | Adjusted lift x 60,398 activated records = 996.1 purchases; x 54 USD = 53,789.47 USD and 53,789.47 USD / 260,000 USD = 0.2069x. |
| Modeled benchmark | 0.9% modeled benchmark x 60,398 activated records = 543.6 purchases; x 54 USD = 29,353.43 USD and 29,353.43 USD / 260,000 USD = 0.1129x. |
| Headline inflation check | 483,950.36 USD naive value / 53,789.47 USD adjusted value = 8.9971x, displayed as 9.0x. |
Assumptions: The treatment assignment, outcomes, propensity deciles, 54 USD value per conversion, 260,000 USD media cost, and 0.9% benchmark are simulated inputs. The within-decile adjustment assumes the modeled propensity variables capture the main selection differences and does not remove unmeasured confounding.
Limitations: No real offer-eligibility file, randomized availability flag, baseline merchant or category spend, confidence intervals, transaction deduplication audit, discount/refund handling, privacy review, or observed advertiser result is supplied. A real readout would need those materials before supporting incremental demand, ROAS, or renewal claims.
Readout audit questions
Use this pass before the case becomes a budget argument, buyer proof point, or channel lesson. The goal is to separate what the dashboard observed from what the campaign plausibly changed.
| Check | Question to ask | Evidence in this case | Safer claim boundary |
|---|---|---|---|
| Credited outcome | What is the report counting as campaign impact? | Separate attributed outcomes from incremental outcomes. In this case, the naive lift is 14.8 pts while the stratified read is 1.6 pts. | Treat path, click, form, survey, or matched-outcome reporting as descriptive until the counterfactual is visible. |
| Comparison group | Were treatment and control groups comparable before the campaign? | 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. | Ask for a protected holdout, balanced market, suppression test, or matched comparison before using causal verbs. |
| Budget threshold | Would the decision survive the adjusted read? | The naive read is 9.0x larger than the stratified estimate, and adjusted ROAS is 0.21x after media cost. | Use the result for diagnosis, repair, or test planning unless the adjusted evidence clears the decision threshold. |
| Next evidence | What would change the conclusion? | Separate offer targeting, prior category demand, and buyer-would-have-bought-anyway risk. | Name the missing evidence request before turning one worked example into a general rule. |
For the next review, pair this case with Retail media incrementality checklist and the claim confidence rubric.
Targeting-balance decision logic
Before this case becomes proof that a card-linked offer created merchant lift, read the transaction result through eligibility and baseline demand. The useful question is whether comparable shoppers without offer availability would have spent differently.
| Review point | Evidence to request | Decision consequence |
|---|---|---|
| Eligibility rules | The exact eligible audience, offer-ranking rules, bank or card purchase-history inputs, recency windows, category-affinity thresholds, merchant exclusions, and whether shoppers had a fair chance to receive the offer. | If activated users are compared with broad non-activated consumers, describe the result as targeted-shopper response until the eligible population is visible. |
| Pre-period spend balance | Baseline merchant spend, category spend, purchase frequency, days since last purchase, loyalty or existing-customer status, and pre-campaign basket value for activated and comparison shoppers. | If activated shoppers had stronger merchant or category demand before the offer, remove incremental spend and ROAS language from the readout. |
| Offer availability | A randomized offer-availability flag, suppression group, holdout construction, exposure logs, activation rate, and whether activation is separated from purchase response. | If treatment requires claiming or activating while the comparison group never faced the same opportunity, treat the gap as selection-contaminated. |
| Outcome accounting | Net spend after discount or reward cost, new-versus-existing customer splits, return/refund handling, purchase window, and whether matched transactions include normal repeat buying. | If the outcome is gross spend among redeemers, call it tracked redeemed spend rather than merchant lift. |
| Repair path | Locked eligibility rules, randomization inside the eligible audience, baseline balance reporting, a protected holdout, and a prewritten claim boundary before launch. | If renewal budget depends on incrementality, require the repair path before using the transaction gap as causal proof. |
Worked downgrade
The headline version of this case says activated card-linked offer users produced a 14.8 point merchant purchase lift, 8,962 incremental purchases, and $483,950 in attributed merchant spend. The report looks persuasive because the transaction match is real and the purchase outcome is close to revenue.
The weaker read appears when the eligibility file shows that the offer model favored shoppers with recent category transactions, prior merchant-like behavior, and stronger purchase recency before the promotion. After balancing shoppers by offer propensity and baseline demand, the estimated movement falls to 1.6 points and $53,789 in attributed spend against $260,000 of media cost. The modeled benchmark is 0.9 points.
The safer readout sentence is: activated card-linked offer users spent more at the merchant, but the evidence does not prove incremental merchant demand because offer eligibility and prior purchase behavior were not comparable. The next action is to rerun with randomized offer availability inside the eligible audience, baseline-spend balance, and separate reporting for activation, redemption, new-customer share, and net incremental spend.
The advertiser-facing story
The reported result shows a large spending gap between activated offer users and non-activated consumers. Because the platform can match card transactions back to the merchant, the readout feels more concrete than click or view reporting. The risk is that closed-loop visibility makes a selected-audience gap look like created demand.
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. The issue is not that transaction data is weak; it is that the same data can identify likely buyers before the offer changes anything.
Better design
Randomize offer availability inside the eligible audience, pre-register the outcome window, separate activation from purchase response, report new and existing customers separately, and calculate incremental spend after removing baseline spend differences and reward costs.
Propensity-strata audit
The adjusted estimate compares activated and comparison shoppers within similar offer-propensity strata. That does not replace randomized offer availability, but it shows whether the headline transaction gap was carried by shoppers whose prior card behavior already predicted a merchant visit.
| Propensity stratum | Eligible shopper records | Activated-offer purchase rate | Eligible-comparison purchase rate | Within-stratum difference |
|---|---|---|---|---|
| 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 card-linked offer readout should not stop at activated-user spend or closed-loop transaction matching. It should show the eligible audience, offer randomization or suppression, pre-period merchant and category spend, new-versus-existing customer splits, discount handling, and how much of the result survives a fair counterfactual.
Move from this case to the next review.
Use the case to tighten claim language, choose a better measurement method, and review the next campaign readout without treating a worked example as a rule.
Claim review
Set the claim language
Score whether this evidence supports strong, qualified, tentative, or diagnostic wording.
Method choice
Choose stronger evidence
Pick the holdout, matched-market, MMM, survey, or QA method that fits the next decision.
Readout route
Inspect the report handoff
Move from the case lesson into delivery, response quality, outcomes, comparison strength, and renewal language.