Fix the comparison boundary before calculating
Records and assumptions to align
- Use one enrolment cohort, purchase period, policy version and observation horizon. (not complete)
- Separate requested, approved, paid, partial and reversed refunds. (not complete)
- Record payment/platform fee credits and delivery/support costs already incurred. (not complete)
- Map purchase, payout, refund and fee-credit dates. (not complete)
- Keep any conversion, completion or trust effect outside the base case unless supported by a controlled comparison. (not complete)
| Field | Required unit | Preferred evidence |
|---|---|---|
| refund paid | CU/refund event | payment/refund record |
| unrecovered delivered variable cost | CU/refund event | delivery and provider records |
| realised fee/cost recovery | CU/refund event | provider statement or labelled scenario |
Build a reproducible scenario
refund economic loss/event = refund paid + non-recoverable fees + incremental administration + unrecovered delivered variable cost - realised fee/cost recovery; cohort impact = refund events × loss/event
- refund paid
- Cash returned under the observed policy event (CU/refund event) — payment/refund record
- unrecovered delivered variable cost
- Materials, licences, contractor/support cost already incurred and not recovered (CU/refund event) — delivery and provider records
- realised fee/cost recovery
- Only credits or recoveries actually received or explicitly scenarised (CU/refund event) — provider statement or labelled scenario
Display dated gross refund cash separately from economic loss. Define whether the cohort baseline already deducts normal delivery cost: if it does, do not add that cost again in refund impact. Do not assume a provider reverses fees or that a guarantee causes incremental enrolment.
Fictional 100-enrolment cohort
Eight refund events and all monetary values are invented scenario inputs. The 140 CU baseline is contribution after normal delivery costs, including the 25 CU delivered variable cost, but before refunds and refund-specific fees, administration and credits.
| Case | Declared inputs | Substitution | Result |
|---|---|---|---|
| Loss per refund | 300 refund; 12 non-recoverable fee; 8 administration; 5 realised credit | 300 + 12 + 8 - 5; the 25 CU delivered cost is already in the baseline | 315 CU/event |
| Cohort refund impact | 8 refund events; 315 CU/event | 8 × 315 | 2,520 CU |
| Retained cohort contribution | 100 enrolments × 140 CU contribution after normal delivery cost and before refunds; 2,520 CU refund impact | 14,000 - 2,520 | 11,480 CU |
Stress-test the uncertain inputs
| Variable | Lower case | Higher case | What it tests |
|---|---|---|---|
| Refund events | Completed-cohort lower case | Explicit downside case | Tests cohort contribution |
| Fee/cost recovery | No unsupported credits | Documented realised credits | Tests economic loss separately from gross refund |
| Refund timing | Before payout/expense | After payout and delivery | Tests peak cash exposure |
Stop and review when
- Refund records cannot be tied to a policy version and cohort. (not complete)
- Fee credits or conversion uplift are assumed without evidence. (not complete)
- Cash paid and economic loss are reported as one number. (not complete)
Turn the scenario into a controlled decision
- Reconcile the baseline to current records and name the evidence owner.
- Run the base case, then change one uncertain input at a time.
- Record the chosen response, approval limit, review date and stop trigger.
- Compare actual results with the original boundary before reusing the assumption.
Avoid these mistakes
- Multiplying revenue by a generic refund benchmark. (not complete)
- Calling gross refunded cash the final loss. (not complete)
- Using the model to write or validate a legal guarantee. (not complete)
Questions to resolve before approval
- Does a guarantee increase sales enough to pay for refunds?
- This arithmetic cannot establish causation. Measure comparable cohorts or treat uplift as a separately labelled sensitivity.
- Are payment fees always returned after a refund?
- No global assumption is safe. Use the current provider contract and realised statement credit.
Methodology and source boundary
- Refund Cash & Profit Impact methodology — Margin101: Product-owned refund cash, fee credit and economic-loss separation.
- Returns Profit Impact methodology — Margin101: Product-owned event/cohort frequency and recovery boundary.
- 13-Week Cash-flow Forecast methodology — Margin101: Product-owned dated cash exposure boundary.