Return-rate analysis: use comparable cohorts and reasons

A D2C ethnic-wear brand in Jaipur reviews a weekly merchandising dashboard on 15 October to decide which of two festive kurta launches (Launch A: SKU-KRT-101, launched 15 September, selling price ₹2,000.00; Launch B: SKU-KRT-202, launched 8 October, selling price ₹2,000.00) should receive a 1,500-unit factory re-order under a 15-day post-delivery return policy. Both styles have 1,000 gross dispatched units (₹20,00,000 gross value each). The naive dashboard divides total returns received to date (including courier RTOs) by gross dispatched units and displays a 26.00% return rate for Launch A (260 / 1,000) versus an apparent 10.50% return rate for Launch B (105 / 1,000)—prompting the buyer to double down on Launch B. In reality, Launch A's 1,000 dispatches are 100% matured past D+15 (120 courier RTOs = 12.00% RTO rate, leaving 880 delivered units, of which 140 were returned by customers = 15.91% matured CIR rate). Launch B's 1,000 dispatches have 110 courier RTOs, 490 units still in transit or within Day 1–4 of delivery, and only 400 delivered units that have reached Day 7—with 88 customer returns already logged (88 ÷ 400 = 22.00% partial CIR rate, tracking toward 26.50% at D+15) due to a miscut shoulder seam!

The operational decision for a customer experience analyst and merchandise owner across D2C, apparel, and marketplace sellers is how to define comparable order cohorts, separate logistics RTO from product-driven customer returns, and validate return reason codes before changing a garment pattern, size chart, or supplier allocation (Shopify Help Center — Returns and exchanges management, Amazon Selling Partner API (SP-API) — Seller use cases and return analytics, StateSet — Customer support and return operations automation). By linking this analysis to promotion contribution margin after returns and failed delivery (NDR) and RTO decision workflows, customer experience analyst teams prevent premature buying bets on immature cohorts and fix the exact defect driving returns.

Why calendar-month return ratios and unverified dropdown reasons mislead buyers

Three methodological errors routinely corrupt retail return-rate reporting:

1. Conflating courier RTO with post-delivery customer returns (CIR): A parcel that never reached the customer because of an incomplete address or cash-on-delivery refusal (NDR and RTO workflow) is a checkout/logistics failure, not feedback on garment fit or fabric quality. Mixing RTO and CIR into a single percentage hides whether you need an address-verification rule or a pattern-grading correction.
2. Right-censoring bias on immature cohorts: Under a 15-day post-delivery return policy (D+15), an order dispatched 5 days ago (3–4 days transit + 1–2 days with customer) has had almost zero time to generate a return request (Shopify return lifecycle tracking). Comparing a 7-day-old launch against a 30-day-old launch guarantees that the newer launch looks artificially superior.
3. First-dropdown click bias versus physical inspection truth: Customers routinely click the top dropdown option in a return portal ('No longer needed' or 'Ordered wrong size') to speed through RMA creation, while writing 'Colour bled on first cold wash' or 'Stitching open at sleeve' in the free-text note—or returning a garment whose chest measurement is 4 cm smaller than the size chart (Amazon SP-API return reason and customer comment structures, StateSet return triage workflows).

Use these four cohort-normalised equations for every return-rate review:

Courier RTO Rate (%) = (Pre-Delivery RTO Units ÷ Total Dispatched Cohort Units) × 100

Matured Customer-Initiated Return (CIR) Rate at D+15 (%) = (Customer Returns within 15 Days of Delivery ÷ Delivered Units with >= 15 Days Elapsed Since Delivery) × 100

Net Retained Delivered Units (Matured Cohort) = Matured Delivered Units − Matured Customer-Initiated Return Units

Verified Defect Reason Share (%) = (Returns Confirmed with Specific Defect at Warehouse QC ÷ Total Inspected Customer Returns) × 100

Enforcing cohort maturity and reason-conflict retention stops merchandising teams from scaling a defective new launch just because its return window has not closed yet.

  <path d="M219 86 L248 86" stroke="#475569" stroke-width="1.8" marker-end="url(#arrow-r050)"/>

  <rect x="248" y="40" width="195" height="92" rx="8" fill="#fef9c3" stroke="#ca8a04" stroke-width="1.5"/>
  <text x="262" y="64" font-weight="700" fill="#854d0e">2. Enforce Cohort Maturity</text>
  <text x="262" y="84">A (D+15): 140/880 = 15.91% CIR</text>
  <text x="262" y="102">B (D+7): 88/400 = 22.00% CIR</text>
  <text x="262" y="118" font-size="11" fill="#475569">Block naive 10.50% B ranking</text>

  <path d="M443 86 L472 86" stroke="#475569" stroke-width="1.8" marker-end="url(#arrow-r050)"/>

  <rect x="472" y="40" width="195" height="92" rx="8" fill="#ffedd5" stroke="#ea580c" stroke-width="1.5"/>
  <text x="486" y="64" font-weight="700" fill="#9a3412">3. Reconcile Reason Codes</text>
  <text x="486" y="84">A: 140 customer returns</text>
  <text x="486" y="102">52 'Mind Changed' -&gt; Bust -3cm</text>
  <text x="486" y="118" font-size="11" fill="#475569">True fit share: 98/140 (70.0%)</text>

  <path d="M667 86 L696 86" stroke="#475569" stroke-width="1.8" marker-end="url(#arrow-r050)"/>

  <rect x="696" y="40" width="200" height="92" rx="8" fill="#dcfce7" stroke="#16a34a" stroke-width="1.5"/>
  <text x="710" y="64" font-weight="700" fill="#166534">4. Targeted Interventions</text>
  <text x="710" y="84">A: Update PDP size guide</text>
  <text x="710" y="102">B: Hold ₹30L factory re-order</text>
  <text x="710" y="118" font-size="11" fill="#475569">Fix B shoulder seam defect</text>

  <rect x="24" y="166" width="872" height="164" rx="8" fill="#ffffff" stroke="#cbd5e1" stroke-width="1.2"/>
  <text x="44" y="194" font-weight="700" font-size="13" fill="#0f172a">Governance rule: Only matured delivered cohorts may be ranked; conflicting customer &amp; QC reasons are retained</text>
  <text x="44" y="220" fill="#334155">• Gate A: Denominator rule separates 120 Launch A RTOs (12.00%) from 880 delivered units (140 CIR = 15.91% at full D+15 maturity).</text>
  <text x="44" y="246" fill="#334155">• Gate B: Maturity guard flags Launch B as Immature (only 400 delivered units at D+7, already at 22.00% CIR vs naive 10.50% blended ratio).</text>
  <text x="44" y="272" fill="#334155">• Gate C: Reason reconciliation retains both dropdown ('Changed mind': 64) and QC tape measurement (52 of 64 had -3 cm bust variance).</text>
  <text x="44" y="302" font-weight="600" fill="#0369a1">Outcome: Prevents a flawed 1,500-unit (₹30,00,000 retail) re-order on Launch B and fixes Launch A's size chart.</text>
</g>
<defs>
  <marker id="arrow-r050" viewBox="0 0 10 10" refX="8" refY="5" markerWidth="6" markerHeight="6" orient="auto-start-reverse">
    <path d="M 0 0 L 10 5 L 0 10 z" fill="#475569"/>
  </marker>
</defs>

Worked illustrative example: comparing a matured September launch against an immature October launch

Illustrative example (hypothetical scenario): Consider the Jaipur D2C apparel brand evaluating Launch A (SKU-KRT-101) and Launch B (SKU-KRT-202) on 15 October, where both styles retail at ₹2,000.00/unit and have 1,000 dispatched units (₹20,00,000 gross dispatched value each) under a 15-day post-delivery (D+15) return policy.

Step 1: Decompose dispatched units into RTO, delivered maturity, and customer returns

1. Launch A (SKU-KRT-101, dispatched 15–22 September — 100% matured at D+15):
   • Gross Dispatched Units = 1,000 units (₹20,00,000)
   • Pre-Delivery Courier RTO Units = 120 units (12.00% RTO rate; ₹2,40,000)
   • Matured Delivered Units (Reached D+15) = 1,000 − 120 = 880 units (₹17,60,000)
   • Post-Delivery Customer Returns (CIR at D+15) = 140 units (₹2,80,000)
   • Matured CIR Rate (Launch A) = 140 ÷ 880 = 15.91%
   • Net Retained Delivered Units (Launch A) = 880 − 140 = 740 units (₹14,80,000 net retained sales)
2. Launch B (SKU-KRT-202, dispatched 8–12 October — Immature cohort on 15 October):
   • Gross Dispatched Units = 1,000 units (₹20,00,000)
   • Pre-Delivery Courier RTO Units Logged So Far = 17 units (with 93 additional COD parcels failing first delivery attempt in the NDR queue; projected RTO = 110 units)
   • Units Still In Transit or <D+5 Post-Delivery = 490 units (return window barely opened)
   • Early Delivered Sub-Cohort at Day 7 (D+7) = 400 units (₹8,00,000)
   • Customer Returns Logged on the 400 D+7 Delivered Units = 88 units (₹1,76,000)
   • Naive Blended Return Ratio (88 CIR + 17 RTO = 105 returns ÷ 1,000 dispatched) = 10.50% (misleading!)
   • Actual D+7 Partial CIR Rate on Delivered Sub-Cohort = 88 ÷ 400 = 22.00% (already +6.09 percentage points worse than Launch A's full D+15 rate of 15.91%!).

Step 2: Reconcile conflicting customer dropdown reasons with inspection evidence on Launch A

When the analyst inspects the 140 customer-returned units of Launch A (₹2,80,000):

• Raw portal dropdown counts: 46 units (32.86%) selected 'Size too small', 64 units (45.71%) selected top-of-list 'Changed mind / No longer needed', and 30 units (21.43%) selected 'Colour/fabric issue'.
• Free-text comment & warehouse QC reconciliation: Of the 64 customers who clicked 'Changed mind', 52 wrote in free text 'Tight across chest compared to usual Size M' and physical QC tape measurement confirmed the garment chest spec ran 3.0 cm smaller than the published PDP size chart (46 + 52 = 98 total true fit/sizing returns = 70.00% of all customer returns, representing 98 × ₹2,000 = ₹1,96,000 in returned merchandise). Only 64 − 52 = 12 units (8.57%) were genuine no-fault preference returns.

In this illustrative scenario, cohort-maturity gating stops the buyer from placing a 1,500-unit (₹30,00,000 retail value) re-order on defective Launch B, while dual-reason retention uncovers that 70.00% (98 / 140 units = ₹1,96,000) of Launch A's customer returns stem from a correctable -3.0 cm size-chart discrepancy rather than customer whim.

Return-rate cohort workflow: roles, exception handling, and system choice

To make return-rate reporting trustworthy for buying, catalogue, and quality teams, implement a four-stage workflow:

1. Anchor every return metric to the delivery-date cohort and split RTO from CIR (Owner: Customer Experience / BI Analyst). Build your return dataset on order_line_id with explicit timestamps for dispatched_at, delivered_at, rto_marked_at, and return_requested_at (Shopify return lifecycle data, Amazon SP-API return reports). Never include pre-delivery RTOs in the numerator or denominator of product Customer-Initiated Return (CIR) rates.
2. Gate comparative SKU rankings behind the full return policy maturity window (Owner: Merchandise Planner). Set Maturity_Window_Days equal to your return policy window plus buffer (e.g., 15 days post-delivery). Any cohort where <85% of delivered units have passed D+15 must carry an Immature_Cohort flag and may only be compared using matched early-day curves (D+7 vs D+7).
3. Retain dual reason fields and reconcile conflicts (Owner: Returns QC Lead & Support Analyst). Store customer_dropdown_reason, customer_free_text, and warehouse_inspection_reason as separate columns (StateSet support and return data structures). Whenever customer_dropdown_reason conflicts with free text or physical measurement, flag the row for classification review rather than overwriting either record.
4. Assign one owned corrective action per verified return driver (Owner: Category Buyer / Technical Designer). Feed matured CIR rates and return processing costs into promotion contribution after returns. Route high Size_Mismatch rates to PDP size-chart updates and high Seam_Or_Fabric_Defect rates to vendor corrective action.

Choosing the right tool: Build the cohort-maturity table and dual-reason reconciliation view in SQL or a controlled BI semantic layer before introducing any NLP automation. Once the structured denominator and QC fields are reliable, use lightweight text classification to group free-text return comments—subject to weekly human sample review.

Observable acceptance tests for your retail systems

Before using return-rate dashboards for re-order or delisting decisions, run three acceptance tests (ACT-R050-01 to ACT-R050-03):

Putting cohort-based return analysis into practice

To stop chasing phantom return trends and fix the real drivers of post-delivery returns:

1. Split every return report into Courier RTO (% of Dispatched) and Customer-Initiated Returns (% of Delivered) so logistics delivery failures never masquerade as product quality defects.
2. Add a hard D+15 (or policy-length) maturity filter to your merchandising scorecards so newly launched styles cannot be crowned low-return winners before their return window closes.
3. Add a mandatory Warehouse_Verified_Reason field alongside the customer's portal dropdown and audit a 50-return weekly sample where free-text comments contradict the dropdown choice.
4. Run acceptance tests ACT-R050-01 to ACT-R050-03 on your current BI dashboard and connect the verified CIR rates to your promotion contribution after returns model.