Analyse RTO by cohort before changing your COD policy

Last month's RTO rate was 20%. This week's dashboard shows 5%, and the team concludes that its new COD policy is working. More than half of this week's shipments are still in transit.

Analyse return to origin using a stable cohort and explicit outcome states. A lower reported percentage can reflect younger shipments, a changed denominator or a different order mix. The business needs to establish which part of the delivery process changed before tightening payment policy.

Define the population and the event

Choose a cohort entry event, such as confirmed carrier handover, and keep its shipment identities fixed. Order creation can include cancelled orders that never entered delivery. Label creation can include parcels that were never picked up.

Define the RTO event used in the metric. Initiation indicates a return process has begun. Physical return receipt and warehouse acknowledgement occur later. A case may still be disputed or corrected between those stages.

Shiprocket describes distinct shipment and remittance reporting uses, and its NDR guidance presents actions that can precede return. Match exported statuses to the metric rather than assuming one report's label is the final outcome. Shiprocket — Choosing scheduled reports Shiprocket — What is NDR?

Record the observation cutoff and cohort age. Keep unresolved, lost, cancelled-after-handover and disputed cases visible according to the declared taxonomy.

A young cohort can look deceptively good

The following illustrative cohorts each contain 1,000 handed-over shipments. The older cohort has resolved outcomes; the younger one is observed much earlier. The example uses RTO initiated as the return outcome for this comparison.

The younger cohort's 5% is an observed-to-date rate, not a forecast of its eventual rate. Dividing fifty by the 450 resolved shipments gives about 11.1%, but that also has selection bias if quickly resolved shipments differ from those still travelling.

The useful display is the full state distribution: 40% delivered, 5% RTO initiated and 55% unresolved. Compare cohorts at the same elapsed age and continue showing open cases beyond the chosen review horizon.

A fixed fourteen-day or thirty-day horizon is not universally mature. Select it using the actual service and observed resolution pattern. Long-running exceptions should remain visible rather than being declared delivered to close the report.

Separate RTO from physical return performance

Suppose the older cohort's 200 initiated returns include 170 received at the origin, twenty still returning and ten under investigation. The return-receipt completion measure is 170/200, or 85%, for that initiated population.

It answers a different question from the 20% RTO rate. It should not replace that numerator or be added to it. The 170 receipts are a subset of the 200 initiated returns.

Warehouse disposition is another stage. Received goods may be held, damaged or awaiting identity checks. Treating every completed RTO as available stock mixes a transport outcome with an inventory decision.

Look for a changed mix before blaming payment mode

Compare COD and prepaid shipments within relevant groups: origin and destination lane, carrier service, product type, order age, first-time versus repeat order where lawfully and appropriately available, and confirmed address quality.

Use only segmentation that helps the operating decision and respects access and privacy requirements. A postcode or customer category should not become an unsupported proxy for dishonesty.

Consider a separate illustrative mix example. One lane group has a 10% RTO rate and another 30%, unchanged between periods. A period with 80% of shipments in the first group has an overall rate of 14%. A period split equally has an overall rate of 20%.

The six-percentage-point increase comes entirely from mix: 0.8 × 10% + 0.2 × 30% versus 0.5 × 10% + 0.5 × 30%. It does not demonstrate that either lane or the COD process deteriorated.

Investigate the reason chain

Reported return reasons are useful leads, but verify them against attempt and NDR records. A final “customer refused” label may follow an earlier address problem, delayed attempt or unresolved payment-mode error.

Assign the action to the process that can change it. A blanket COD restriction can reduce orders without resolving the underlying dispatch or delivery issue.

Test the proposed policy on a defined basis

Select a limited, approved change, such as improved address confirmation for a particular exception group. Preserve comparable cohorts and relevant service conditions. Where a controlled comparison is feasible, define it before seeing results.

Measure delivered contribution, delivery outcomes, order conversion and customer impact, not only the RTO percentage. A policy that excludes many legitimate orders can lower RTO while reducing useful business.

Use a worksheet or dashboard with frozen cohort identities and reproducible status rules. A predictive score needs separate validation, including false positives and changes in the population. Do not treat the score as an explanation of why a particular customer returned an order.

Change the process the evidence identifies

A useful RTO analysis distinguishes time, mix and execution. That lets the business address a specific failure without mistaking an immature dashboard or different order population for a policy result.

Next actions

1. Define the cohort entry event, RTO state and observation cutoff.
2. Show delivered, returned and unresolved populations together.
3. Compare like-for-like segments at comparable ages.
4. Trial the smallest process change supported by verified reasons and measure its wider effect.