STR.12:5.4 - More paid customers need not mean a better conversion result
A service is considering further spending on a revised onboarding offer. Its dashboard shows paid-customer counts rising from twenty to forty-five. The analyst recovers the registration groups and actual first orders instead of treating that increase as the effect of onboarding. Assume that the permitted records establish each customer’s identity, entry and first-order date, with complete follow-up where stated.
| Registration group | Registered customers | Customers with a first paid order | Observation basis |
|---|---|---|---|
| January | 100 | 20 | Each customer’s first 28 days |
| February | 300 | 45 | Each customer’s first 28 days |
| March | 200 | 12 so far | Each customer’s first 7 days only |
The observed twenty-eight-day proportions are 20% and 15%. More customers paid in February, but a smaller proportion of that registration group paid within the defined period. March’s 6% after seven days is not its twenty-eight-day result. Without the older groups’ seven-day readings, it cannot supply that shorter-period comparison either.
Suppose February also used a different acquisition channel. These records do not isolate the onboarding change’s effect, prove that it is harmful, or establish the result for future customers. The analyst withdraws the claimed effect and returns the supported counts, proportions and uncertainty about that stronger inference. If an effect estimate is necessary for the next commitment, an adequate existing causal result or a worthwhile controlled comparison can supply it. The current descriptive correction can finish without another trial; the funding decision still needs its own comparison and authority.