Return-rate methods

Return counts and return rates answer different questions. A return rate needs a compatible activity denominator and an explicit time relationship between the numerator and denominator.

The labels below are analytical labels used by this skill, not official Amazon metric names.

share_of_returns

product return units / all return units in the analyzed returns source

This measures composition or concentration. It is useful without shipment data, but it is not a return rate. Say “share of return units” and identify whether the numerator counts units, requests, orders, or rows.

window_biased

returns whose return event falls in W / shipments whose shipment event falls in W

The numerator and denominator usually represent different order cohorts. Returns in the window often originate from earlier shipments, while some shipments in the denominator will return after the window closes. Treat the result as directional.

Expected directional effect when return latency is otherwise stable:

• For a rapidly growing ASIN, older shipment cohorts are smaller than current shipment volume, so the same-window metric tends to understate the eventual cohort rate.
• For a rapidly shrinking ASIN, older shipment cohorts are larger than current shipment volume, so the metric tends to overstate the eventual cohort rate.
• Promotions, stockouts, seasonality, and changes in return latency can alter the direction. Do not claim a deterministic correction factor.

Quote this as, for example, 8.4% (window_biased) and explain the event dates.

cohort_overlap

This is a supplementary same-window view restricted to products present in both the returns and shipment datasets. It can reveal how unmatched products affect the portfolio number, but it does not fix the cohort mismatch. Never present it as a true cohort rate.

true_cohort

Define a shipment cohort, join later returns to the originating shipment using compatible order and product identifiers, and observe each shipment for a stated follow window:

returned units attributed to the shipment cohort / units in the shipment cohort

State:

• cohort start and end;
• follow-window end or duration;
• join keys and match rate;
• treatment of replacements, cancellations, multiple units, and partial returns; and
• unresolved and late-arriving returns.

The follow window should follow the use case, category, and available evidence. Do not claim that one duration captures a universal percentage of returns.

Alignment checklist

Before quoting any rate, verify:

1. Marketplace: numerator and denominator cover the same marketplace set.
2. Channel: do not divide mixed-channel returns by an FBA-only denominator.
3. Item grain: child ASIN, parent ASIN, SKU, and FNSKU are not interchangeable without a mapping.
4. Unit grain: both sides count compatible units rather than mixing rows, requests, orders, and units.
5. Dates: identify return, return-request, shipment, and cohort dates.
6. Coverage: compare actual source coverage, not only requested dates.
7. Zero denominator: return N/A, not infinity, zero, or 100%.
8. Join quality: for cohort rates, report matched and unmatched records.
9. Comparability: use identical definitions for current and prior periods.

Reporting contract

For each rate, show:

• numerator value and definition;
• denominator value and definition;
• method label;
• date or cohort basis;
• exclusions and unresolved records; and
• the practical limitation that matters to the decision.

If no compatible denominator exists, stop at counts, concentration, and share_of_returns.