NPS & surveys: hear the churn before it happens
A customer who scores you a 6 is telling you they're about to leave, weeks before their order history shows it.
A simple NPS or post-purchase survey is your churn early-warning system. AI clusters the open-text answers into themes so you know exactly what to fix first.
Fastest path: one prompt, end to end
🤖 AI prompt — paste into ChatGPT / Claude
You are a customer-feedback analyst for a small e-commerce brand.
My product: [product] Survey responses I have: [paste scores + open-text answers, or say none yet]
Where I could survey: [post-delivery email / post-support / on-site]
Do this:
1. If I have responses: cluster the open-text answers into themes, rank by frequency x severity, and flag which themes correlate with low scores (my likely churn drivers).
2. If I have none yet: design my first survey. One score question (0-10 likelihood to recommend), then ONE follow-up: "what's the main reason for your score?" Nothing else.
3. Recommend timing: post-delivery (a few days after the product arrives, not at order) and post-support-resolution. Explain why timing changes the answer.
4. Give me a close-the-loop plan: how to reply to detractors personally within 48 hours, and how to show all customers that feedback changed something.
If you don't have my real responses, ask; do not invent themes.
Output: themes ranked (or first survey) + timing + close-the-loop plan.
Or do it in 4 steps
- Ask two questions, not twelve. A score (0-10) plus one open follow-up ("what's the main reason for your score?"). Response rates collapse with every extra question, and the follow-up verbatim is where all the value lives. The score is just a sorting key.
- Time the ask to the moment. Survey a few days after delivery (they've used the product) and right after a support ticket resolves. Surveying at order time measures your checkout, not your product. Timing is the difference between signal and noise.
- Treat detractors as churn alarms, not critics. Anyone scoring 0-6 gets a personal reply within 48 hours, a real one, from a human. Many are one good interaction away from staying, and the ones who leave anyway tell you exactly what to fix.
- Close the loop where everyone can see it. "You asked, we changed the packaging" in your email or on the product page turns feedback into trust. Customers who see their input matter answer the next survey too, and the cycle compounds.
Worked example (labeled): a skincare brand adds a 2-question survey 5 days post-delivery. Score average is 8.1, but detractor verbatims cluster on one theme: the pump dispenses too much product. A personal reply plus a usage tip saves several detractors, and the next production run fixes the pump.
Repeat rate among surveyed customers runs visibly higher than unsurveyed. That's not the survey itself. It's that the worst friction got found and fixed a quarter earlier than order data would have shown it.
This note covers your owned surveys; for listening on social channels, see the social-community-listening note under marketing.
Survey continuously, reply to every detractor, and fix one theme per quarter; the score is the thermometer, the verbatims are the diagnosis.