Return rate on Flipkart has gone up. How would you diagnose and improve it?

A return rate moving is only meaningful once I know exactly what's being counted, so let me pin the definition and the shape before I hypothesise.

Before I chase the obvious suspect, I'll lay out the whole space so I don't tunnel.

0. Is the rise even real? Cheapest to rule out first:

• Definition / instrumentation: the return window was extended so more orders become return-eligible, reason codes or the return flow changed, or we started counting differently.
• Denominator: if orders fell, the rate rises mechanically even with flat absolute returns.
• Mix shift: order composition tilted toward structurally high-return categories. Fashion returns at 25 to 40%, electronics and BGM far lower, so a shift toward fashion lifts the blended rate on its own.
• Seasonality / calendar: a post-sale return wave (a Big Billion Days hangover), festive returns.

Then, if it's real, a return is a gap between what the buyer expected or needed and what they got. Why any return happens is MECE by origin:

1. Purchase-time expectation gap (listing): misleading images, wrong size chart, inaccurate specs, inflated reviews, counterfeit → reason codes "not as described", "size/fit", "quality".
2. Fulfilment / condition (in transit): damaged, defective on arrival, wrong item, missing parts, late → "damaged", "wrong item".
3. Buyer behaviour: bracketing (order several sizes, keep one), buyer's remorse or "changed mind", COD refusal at the door, serial or abusive returners, or a new low-intent buyer cohort.
4. Policy / external: a relaxed returns policy (free or extended returns) that changed behaviour, a competitor, or a promo that pulled in deal-seekers.

That is collectively exhaustive: a return is triggered at purchase, in transit, by the buyer, or by a policy or mix change. Now I prune, locate it in time, then segment by reason code, category, buyer cohort, seller and logistics lane.

I want to pin the rise to a date and see what moved alongside it.

Before I blame the policy, I want to see exactly which returns rose.

Even with a prime suspect, I'd keep a short hypothesis set:

Most likely (policy-induced, fits the data): free returns on fashion both shifted mix toward fashion and, within fashion, lowered the cost of returning, enabling bracketing on size and casual "changed mind". The underlying enabler is weak size/fit guidance, buyers can't get size right first time, so free returns turn that into bracketing.

Worth ruling out (listing / quality): if "not as described" or "quality" were also rising, that's a seller-quality or counterfeit problem, not fit. The data says size/fit dominates, so this is secondary, but check.

Worth ruling out (external / abuse): serial returners exploiting free returns, or a competitor. The clean within-fashion size/fit signal makes broad abuse unlikely, but a five-point jump earns one look at the repeat-returner tail.

To confirm the leading hypothesis rather than assume it:

• Reason-code and bracketing funnel: the size/fit share of fashion returns over time, and a bracketing signal, baskets with two or more sizes of the same style, and the keep-one-return-the-rest rate.
• Control cohort: sub-categories or regions not yet on free returns, compare the rate trend. This is the falsifying test from the pushback.
• Buyer conversations and returned-item condition: are size/fit returns genuine fit misses (bad size charts) or habitual bracketing? Are returned units even resalable?
• Boring failure modes: did a release break or hide size-chart data? Did a reason-code taxonomy change inflate "size/fit"?

Two horizons, and I'd lead with the durable fix, not a policy clamp.

Reduce returns at the source (cut the expectation gap).

• Size/fit, the dominant reason: better size charts, on-model measurements, fit predictors that learn from a buyer's past kept sizes, verified fit reviews, and standardised brand sizing. This attacks the biggest reason code directly.
• Listing quality: accurate images and specs, and penalise sellers with high "not as described" return rates.

Shape behaviour without nuking a buyer-friendly promise.

• Free returns very likely lifts fashion conversion and GMV, that's the trade we made on purpose. So I would not blanket-reverse it. Instead target the abusive tail: cap free returns per buyer per period, add gentle friction or a fee only for serial returners, and flag bracketing patterns.
• Make first-time-right good enough that bracketing isn't needed.

Prevent recurrence. Never ship a returns-policy change without a return-rate guardrail and a staged rollout, and model the mix and behaviour effect before going wide.

What I would not do. I wouldn't just reverse free returns to make the metric look good. Return rate is a means; the goal is profitable, return-adjusted GMV and retention. Killing free returns could cut returns and cut fashion conversion and repeat purchase, losing more than it saves.

I'd track the fix in three layers:

• Leading indicators (days to weeks): fashion size/fit reason-code share, bracketing rate, first-time-right (kept-on-first-order) rate, and adoption of the new size tools.
• Lagging headline (the outcome): blended return rate easing back toward baseline, decomposed into mix versus within-segment so I know which lever worked, plus return-adjusted contribution per fashion order.
• Guardrails (must stay intact): fashion conversion and GMV (didn't kill the goose), returns-experience CSAT, fashion-buyer repeat purchase and retention, reverse-logistics cost, and refund-fraud rate.

Two things could make me wrong, and I'd guard against both:

The free-returns read could be confounded by the sale. That's exactly why the control cohort and the mix-versus-within-segment decomposition come before any policy change, not after.

The fix could overcorrect. Friction or fees on returns can depress fashion conversion and retention, trading a metric win for a GMV loss. So I'd target only the abusive tail, lead with the sizing fixes, treat any return friction as a time-boxed scalpel, and watch the conversion and retention guardrails the whole way.

So: the rise is a real five-point step concentrated in fashion, driven by a free-returns rollout that both shifted mix toward fashion and lifted within-fashion size/fit returns through bracketing. I'd confirm it with a non-migrated control cohort and the reason-code funnel, then fix at the source with sizing and listing quality plus targeted friction on the serial-returner tail, rather than killing a policy that lifts GMV, and I'd guard conversion, retention and return-adjusted contribution so I don't win the return-rate metric and lose the business.