Two dates, two different questions
A purchase date tells you when exposure was created. A dispute notification date tells you when an outcome reached your team. Both matter. Counting disputes received this week is useful for staffing, but dividing them by this week’s purchases mixes populations.
Compare cohorts at the same age
Group purchases by a shared starting point, such as purchase week. Compare their outcomes after the same observation period. A recent cohort has had less time to accumulate disputes than an older cohort. Apparent improvement may be incomplete observation.
Keep the numerator attached to its denominator
Suppose 20 disputes relate to a cohort of 10,000 settled payments observed for 60 days. The observed dispute rate is 20/10,000 = 0.2% for that cohort and age. It is not a final lifetime rate or automatically a fraud rate. Keep count-based and value-based measures separate.
Do not erase the reason
Disputes may concern authorisation, delivery, cancellation or other issues. Read the reason and relevant evidence before classifying the outcome. Operational dashboards and analytical evaluations can use different views, provided each view states its population, definition and observation period.