Original practice scenario: A fictional card program awards points for purchases, then handles refunds, charge corrections and partially redeemed rewards. Design its ledger.
Read model answer +
Use immutable reward movements linked to the originating activity, with explicit reversal and redemption semantics. A mutable points balance alone cannot explain how a correction should affect prior awards.
Work through it
Assign stable purchase and reward-event identifiers, version eligibility rules and make posting idempotent across repeated transaction notifications.
Record awards, redemptions, expirations and compensating adjustments separately, defining what happens when a refund follows redemption under fictional program rules.
Reconcile materialized balances to ledger totals and test partial refunds, duplicate events, late corrections and concurrent redemptions against declared invariants.
Tradeoffs to discuss
An append-only ledger improves auditability but requires careful policy for negative balances and corrections to historical rule versions.
Common pitfalls
Do not erase an award after it has affected later activity.
A refunded purchase and its reward reversal need independent, linked event identities.
Original practice scenario: A hypothetical card program sends selected customers a merchant offer. Analysts propose comparing redeemers with customers who did not redeem it. Improve the design.
Read model answer +
Redemption occurs after assignment and reflects customer preferences, so redeemers are not a comparable control group. Estimate the effect of offering the promotion using the original assignment.
Work through it
Randomize eligible customers or appropriate clusters before exposure, retain assignment records and define primary spending and cost outcomes over a fixed window.
Analyze intention-to-treat differences with uncertainty, checking balance, spillover, merchant substitution and the total eligible population rather than only active redeemers.
Report redemption as a secondary behavioral outcome, and use any treatment-on-treated analysis only with explicit identification assumptions and limitations.
Tradeoffs to discuss
Assignment-based effects may look smaller than redeemer comparisons but answer a clearer policy question and avoid selection on the outcome path.
Common pitfalls
Do not exclude customers who received the offer but never used it.
Measure net incremental value rather than gross promoted-merchant spending alone.
