Measuring NPS Properly in an Indian D2C Business

Most D2C teams run NPS the way they read a weather report. A number appears on a monthly slide, someone notes it is up two points, and the meeting moves on. The number is the least useful thing the survey produces. The comments underneath it are the actual output, and in most companies nobody reads them past the first week.

Two surveys, two different jobs

There are two versions of this instrument and they answer different questions.

• Transactional NPS fires after a specific event, usually a delivery or a support interaction. It tells you whether that experience worked. It is operational, high frequency, and the right input for fixing a courier, a packaging format or a support queue.
• Relationship NPS goes to the customer base on a schedule, once or twice a year, and asks about the brand rather than an order. It tells you where you stand overall, and it is the number worth trending across quarters.

Teams routinely run one and report it as the other. A score collected after deliveries is dominated by logistics performance, so it will swing when your courier mix changes and will say very little about whether people value the product. Run both, label them clearly, and never average them together.

When to ask

Not at order confirmation. Not at the moment of delivery either. In most Indian D2C categories the right window is around seven days after delivery, once the customer has used the product. For consumables that need a longer trial, push it out to two or three weeks.

Frequency rules matter more than teams admit. If a customer buys monthly, do not survey every order. Survey every third or fourth. Relationship surveys should reach the same person no more than twice a year. Break these and the response rate degrades while your sample skews towards people with a complaint.

Channel matters in India. Email response rates against a consumer base usually sit in low single digits to low teens. In-app and on-site prompts do better. WhatsApp, where you hold opt-in, typically outperforms both, because that is where the customer already receives order updates and it takes one tap to answer.

Response rates and the sample you need

Realistic ranges: transactional surveys land roughly 25 to 40 percent when triggered in-channel and close to the event. Scheduled relationship surveys land roughly 15 to 25 percent. Email-only programmes come in well below that.

Sample size is where NPS reporting quietly breaks. NPS is a difference between two proportions, so its error band runs about twice as wide as that of a simple percentage. At around 200 responses the 95 percent confidence interval on your score is approximately plus or minus 10 points. At 50 responses it is wider than the score itself.

Two consequences follow. You need somewhere between 100 and 200 responses before a monthly movement carries any information, and a move from 42 to 47 on 200 responses is noise being reported as progress. And if your volume cannot produce that, stop publishing a monthly score. Publish a rolling quarter, or publish the themes and leave the number out.

There is an India-specific hole worth naming. You can only survey customers you can contact. Marketplace and quick commerce orders do not give you customer details, so your NPS covers own-site traffic only. If own-site is 30 percent of your volume, your score describes 30 percent of your customers, and probably your most engaged 30 percent. Say so on the slide rather than letting the room assume it is the whole business.

The score is the weakest output

A score tells you a direction. It does not tell you what to do. The verbatim comments do, and they are cheap to process.

Build a reason-code taxonomy of ten to fifteen codes covering what actually goes wrong: late delivery, damaged packaging, wrong item, product not as described, size or quantity mismatch, price versus a marketplace, missing invoice, unresponsive support, subscription billing, and so on. Code every detractor and passive comment against it. Tag the SKU and the pincode cluster. That is the whole system.

Within a month you will have a ranked list of the three failure modes producing most of your detractors, and the ranking is almost never what the team assumed. In practice a large share of Indian D2C detractor comments trace to logistics and packaging rather than the product, which means the fix sits with your 3PL and your carton specification, not with the brand team.

Routing detractors into operations

A survey with no closing loop is theatre. The minimum working system looks like this:

• Every detractor response creates a ticket automatically, with a named owner in CX or operations. Not a dashboard row. A ticket with an SLA.
• Contact detractors inside 48 hours. Recovery contact converts a meaningful share of them into repeat buyers and produces sharper diagnostic detail than any survey field will.
• Review the top reason codes weekly with the operations and CX leads. Put one of them on the fix list each month with an owner and a date.
• Report the delta on that reason code the following month. If late delivery was 34 percent of detractor comments and is now 21 percent, the programme is working. The headline score is a lagging summary of exactly this work.

Run it this way and the score takes care of itself. Run it as a slide and you have bought a monthly number nobody can act on.