E-commerce case interview questions on returns, COD and baskets

Returns and cash on delivery

1. "Return-to-origin rate rose from 12% to 18% last month. What do you look at?"

Metric first: RTO rate split by payment mode (COD versus prepaid), then by pincode or city tier, then by category.

Model answer: RTO — orders that go out but are refused or undeliverable — is usually driven by COD orders, where the buyer has no money at stake. If the rise is concentrated in COD, look at what changed: a new acquisition channel bringing impulse buyers, a longer delivery promise giving buyers time to change their mind, or a courier partner with poor first-attempt delivery in some pincodes. If prepaid RTO also rose, suspect delivery operations rather than buyer intent.

Trap: treating RTO and customer returns as one number. A return after delivery is a different process with a different cost.

This query computes RTO by payment mode and city tier on a small synthetic orders table:

CREATE TABLE orders (order_id INTEGER, payment TEXT, city_tier TEXT, outcome TEXT);
INSERT INTO orders VALUES
 (1,'COD','tier1','delivered'),(2,'COD','tier1','rto'),(3,'COD','tier3','rto'),
 (4,'COD','tier3','rto'),(5,'COD','tier3','delivered'),(6,'prepaid','tier1','delivered'),
 (7,'prepaid','tier3','delivered'),(8,'prepaid','tier3','returned'),
 (9,'COD','tier1','delivered'),(10,'prepaid','tier1','delivered');

SELECT payment, city_tier,
       COUNT(*) AS shipped,
       SUM(outcome = 'rto') AS rto,
       ROUND(100.0 * SUM(outcome = 'rto') / COUNT(*), 1) AS rto_pct
FROM orders
GROUP BY payment, city_tier
ORDER BY rto_pct DESC;

In SQLite a comparison returns 1 or 0, so SUM(outcome = 'rto') counts matching rows. Note that the returned prepaid order is not counted as RTO — it is a delivered order returned later.

2. "Should we switch off COD in the pincodes with the worst RTO?"

Model answer: Compare the cost of RTO with the margin lost from buyers who would not order without COD. For each pincode group, estimate RTO shipping cost per COD order against contribution margin per delivered COD order, and the share of COD buyers likely to convert to prepaid. Softer options come first: partial advance payment, order confirmation calls, or COD limits by basket value. Suggest testing one change in a few pincodes before a blanket switch-off.

Trap: optimising RTO alone. A zero-RTO policy can be achieved by removing COD entirely, which may remove most of the sales in those areas.

3. "Returns in the apparel category are 3x electronics. Is that a problem?"

Model answer: Not by itself — size and fit returns are structural in apparel. Compare apparel return rates with their own history and across sub-categories and brands. A brand with returns far above its category peers, with "size mismatch" as the reason, points to a size-chart problem the seller can fix.

Basket and pricing

4. "Average order value fell 8% during the festival sale. Did the sale fail?"

Metric first: revenue, orders and AOV together, plus contribution margin after discount.

Model answer: AOV often falls during sales because more first-time and price-sensitive buyers place smaller orders. If orders rose enough that revenue and margin went up, the sale did its job. Look at AOV for returning buyers separately; if that also fell, discounts may have pulled basket size down rather than bringing new buyers in.

Trap: judging a sale on one ratio. AOV is a ratio of two things that both moved.

5. "Would a free-delivery threshold of ₹499 raise basket size?"

Model answer: Look at the current distribution of basket values. If many orders sit just below ₹499, a threshold there can nudge them up; if most orders are far below, it will mostly annoy buyers. Estimate the shipping cost absorbed on orders above the threshold, then test the threshold on a random share of users and measure orders per user, AOV and margin, not only AOV.

6. "A 20% discount coupon had 40% redemption. Was it good?"

Model answer: Redemption rate says how many used it, not how many would have bought anyway. The question is incremental orders: compare a random holdout group that did not receive the coupon. If the holdout bought at nearly the same rate, most of the discount was a giveaway to existing buyers.

Customers and sellers

7. "How would you measure whether customers are coming back?"

Model answer: Cohort retention: group customers by month of first order and track the share who order again in each following month. Report both "ordered again within 90 days" and orders per customer over six months, and separate buyers acquired in sales from those acquired at full price, because their behaviour often differs.

8. "Cart abandonment is 75%. How do you reduce it?"

Model answer: First, see where people drop: cart to address, address to payment, payment to success. A spike at payment points to failure rates on a particular method — check UPI and card success rates by bank or app. A drop at the address step may point to serviceability: buyers discovering their pincode is not deliverable only at checkout. Fix what the funnel shows, rather than sending reminders to everyone.

Trap: assuming abandonment is a single behaviour. Many carts are wish lists and will never convert.

9. "Which sellers should the category team talk to this week?"

Model answer: Rank sellers by a small set of quality metrics that matter to buyers — cancellation rate, late dispatch, return rate with seller-fault reasons, and ratings — weighted by order volume so a tiny seller with two bad orders does not top the list. Show the list with the reason for each seller so the category manager knows what to raise.

The pattern behind all nine

Each strong answer names the metric and its split before any action, separates two outcomes that look similar (RTO versus return, redemption versus incremental), and ends with a test or a decision rule. Practise by writing down the order lifecycle of any app you use and the metric at each stage.

For a similar case in lending, see the banking and credit case bank. E-commerce order, return and cohort data are among the case datasets in the Data Analyst programme.