Define a Rebate Campaign Funnel Before Comparing Conversion Rates

“Reward conversion is 40%” is not yet a useful measurement. It could mean that 40% of exposed shoppers purchased, 40% of approved awards were claimed, or 40% of issued cards had an observed spending event. Each statement describes a different question and a different denominator.

Before comparing campaign performance, build a metric dictionary that makes those choices explicit. The dictionary should allow another analyst to reproduce the rate without asking what the dashboard author meant by conversion.

Choose the population behind each rate

Start with the business question. To understand whether shoppers purchase after seeing an offer, the population is exposed shoppers with a defined opportunity to purchase. To understand completion of an approved claim journey, the population is the awards or recipients that actually reached that opportunity. Do not use a convenient export total simply because it is available.

Consider a hypothetical campaign with 1,000 distinct exposed shoppers, 120 purchasers, and 150 orders. Of those orders, 100 meet the defined qualification conditions. A customer can therefore appear once in the shopper population while generating more than one order. Dividing 150 orders by 1,000 shoppers produces orders per exposed shopper, not a 15% customer purchase conversion rate. The purchaser rate is 120 divided by 1,000, or 12%.

Record the observation window with the denominator. A shopper exposed yesterday has had less time to purchase than one exposed three weeks ago. The dictionary should identify the exposure anchor and the permitted follow-up period. It should also state how incomplete follow-up is represented.

Keep exclusion rules fixed and visible. Internal test profiles can be excluded under a predefined rule, but removing customers because they did not engage with the reward changes the question. A funnel meant to locate friction should not quietly discard the people experiencing it.

Separate customer counts from award counts

Build a bridge between entities rather than forcing everything into one count. An order may produce no award, one award, or a program-specific relationship that needs confirmation. A recipient may have several awards. The actual mapping belongs in the data dictionary.

This is an illustrative dictionary. The actual program may use different stages or provide different observations. RebateCardX’s reporting overview should be checked against the data available to the specific merchant before any field is promised in a dashboard.

For every row, name the identifier used to deduplicate records. A count of claim events can exceed a count of claimed awards if retries or repeated page activity are present. The metric should count the business outcome it names, not every row carrying a similar label.

Build a funnel with explicit unknown states

A funnel should show unknown states alongside completed outcomes. Suppose 100 qualifying orders produce 100 award candidates. At the reporting cutoff, 70 are approved, 10 are declined, and 20 remain unresolved. The approval share among completed decisions is 70 divided by 80, or 87.5%. The approved share of all candidates is 70%. Both are valid descriptions if labeled correctly; neither means the unresolved 20 have been declined.

Continue the fictional example: 56 of the 70 approved awards have completed the claim step. The observed claim completion share is 80% if all 70 had the same full opportunity window. If 15 approvals occurred only yesterday, that rate mixes maturity and customer behavior. Add the window qualification rather than presenting the number as a final result.

A difficult case occurs when a provider report supplies issued cards but not the award identifier needed to join them. Do not distribute the card count across awards by guesswork. Mark the mapping unavailable and identify the reporting gap. A complete-looking funnel assembled from incompatible populations is less useful than a partial funnel whose limits are explicit.

Version the dictionary when a definition changes. If “purchaser” previously meant any order creator and now means a customer with a retained paid order, the historical series needs a comparable restatement or a visible break. Changing the label alone conceals a measurement change.

Before publishing, ask a reviewer to calculate one row from the underlying fictional or sanitized records using only the dictionary. Any unresolved choice about dates, identities, or exclusions is a definition gap to fix. The final result should make a drop interpretable: fewer people reached the stage, fewer completed it, or the outcome is still unknown. Those distinctions determine what the merchant should investigate next.

Test the dictionary with a customer who places two orders and receives one award. That customer contributes one purchaser, two orders, and one award to the respective counts. If both orders were eligible but the approved program combines their benefit, the award count should not be presented as a count of qualifying orders. Write the relationship explicitly in the dictionary and show the bridge separately. This small test is useful because a funnel can appear numerically orderly while concealing a many-to-one relationship between stages. The analyst should be able to explain the missing or combined unit without inventing customer abandonment.

Explore which Rebate Card X reporting capabilities can support your campaign metric definitions. Explore capabilities.

Source references

• RebateCardX’s reporting overview

Back to contents