A CX lead opens the monthly report and sees an NPS of 30, steady again. The support inbox tells a different story. A short AI survey call can ask the feedback questions a single score cannot, and here are 15.
Why one score hides the real story
NPS asks one question and returns one number. It does not say why a customer scored you a 6. Quiet customers skip the survey, and a polite 8 can sit on top of a real complaint. A score can look clean while single conversations still hold objections.
A short voice call fixes part of this. People often find it easier to explain things out loud than to tick boxes. The AI can also ask one follow-up on the spot.
Use the list like this:
- Ask 5 to 6 questions per call.
- Keep the call short.
- Open with an open question, and save yes/no questions for last.
Related reading: 20 Churn Interview Questions for AI Exit Calls With Cancelled and Inactive Customers, Voice of the Customer Program Guide for Teams Using AI Phone Calls and How to Turn Customer Call Transcripts Into Product and Marketing Decisions.
Questions 1 to 5: What really happened
1. The last thing that went wrong
Ask: "What was the last thing that went wrong for you with us?"
Why it matters: NPS gives a number with no event behind it. This question gets a specific moment.
What the answer reveals: A concrete story points to a fixable process. A vague "nothing" often means the customer has stopped expecting much.
2. What almost made them leave
Ask: "What almost made you stop using us?"
Why it matters: Customers who nearly leave can still give a polite 8.
What the answer reveals: The near-exit trigger. In fintech, lending and investment, it is often a stall such as KYC left unfinished or an account never funded.
3. The expectation gap
Ask: "What did you expect to happen that did not?"
Why it matters: A gap between promise and result drives disappointment, even when service was technically correct.
What the answer reveals: Where your marketing, sales script or onboarding set the wrong expectation.
4. The real path taken
Ask: "Who else did you talk to before you got an answer?"
Why it matters: Each extra contact is a handoff that the score never counts.
What the answer reveals: The path the customer really took, such as repeat calls, social posts, a friend, or a rival's help page.
5. The unexpected moment
Ask: "What did we do that you did not expect?"
Why it matters: Surprises shape memory more than routine service does.
What the answer reveals: Good surprises show what to repeat. Bad ones are early warnings. This suits proactive check-ins after an outage, a delay or an incident.
Keep these five open. Let the customer talk for a full minute, then ask one follow-up such as "What happened next?"
Questions 6 to 10: Effort, trust and switching
Customers rarely leave because of one bad moment. They leave because staying felt like work, or because someone else looked easier.
6. How hard it was
Ask: "How hard was it to get this sorted?"
Why it matters: A study of more than 75,000 people found that the Customer Effort Score predicts loyalty better than satisfaction measures or NPS.
What the answer reveals: Repeat contacts, channel switches and long waits. Ask for a number from 1 to 5, then "What made it hard?"
7. The switching point
Ask: "What would make you switch to someone else?"
Why it matters: Loyal-looking customers still have a price or pain point where they would leave.
What the answer reveals: The competitor pull, such as a named rival, a feature or a fee.
8. Trust and human handoff
Ask: "What do you trust us with, and what would you rather speak to a person about?"
Why it matters: 87% of 3,566 customers said companies using GenAI for service must offer a route to a human.
What the answer reveals: Which topics need a human path. Disputes, money and sensitive issues usually top the list.
9. Advice to a friend
Ask: "What would you tell a friend not to do with us?"
Why it matters: Customers often find it easier to be blunt about a friend than about themselves.
What the answer reveals: The traps in your process, such as a form that fails or a step that confuses.
10. What to stop doing
Ask: "What do you wish we would stop doing?"
Why it matters: Teams keep habits that irritate customers because nobody has dared to remove them.
What the answer reveals: Cheap fixes, like extra reminders, scripted greetings or repeated identity checks.
Score question 6 on every call. Read the "what made it hard" answers together, because patterns show up fast.
Questions 11 to 15: Fix and follow up
The last five questions turn complaints into a work list. They also tell you how to reach each customer next time.
11. The first fix
Ask: "If you could change one thing, what should we fix first?"
Why it matters: A forced single choice gives you a ranking. A long wish list gives you noise.
What the answer reveals: The top pain in the customer's own words. When many customers name the same fix, that is your first project.
12. What would bring them back
Ask: "What would bring you back, or keep you with us?"
Why it matters: Lapsed and wavering customers rarely get asked. A score cannot tell you what a win-back needs.
What the answer reveals: The price, feature or promise that would change a decision. A flat "nothing" is a strong churn signal.
13. The unasked question
Ask: "What did you almost ask us, but did not?"
Why it matters: Customers often hold back questions because they feel silly or rushed.
What the answer reveals: Gaps in your help pages, onboarding and agent scripts. Each unasked question is a future call or a quietly lost customer.
14. Topic and channel choice
Ask: "What would you like us to call you about, and would you rather we call or send a message?"
Why it matters: Proactive contact about outages, delays and check-ins works best on the right channel.
What the answer reveals: Topic and channel preference. Store the answer and respect it in every later campaign.
15. Anything we missed
Ask: "Is there anything else I should have asked?"
Why it matters: Your script reflects what you already believe. This question lets the customer step outside it.
What the answer reveals: Unmet needs and blind spots. Tag these answers separately, because new themes often start here.
Save question 15 for the very end. Pause after asking it, and the best answer often arrives in the silence.
How to run these on AI survey calls
Pick one goal per campaign, then choose five or six questions that serve it. A lender checking the first month after disbursement might ask questions 3, 6, 10, 13 and 14. A mobility firm calling riders after a trip might ask 1, 5, 6 and 14. Add one follow-up to each open question, such as "What happened next?"
Post-trip surveys by phone are repetitive and high in volume, and a wrong word costs little. That makes them a safe first thing to move onto an AI caller.
Be honest about the AI. Say it is an AI in the first sentence. In the EU, Article 50 has applied since 2 August 2026, and automated phone systems must say they are AI from the first interaction. In the US, AI-voice calls already need consent under the TCPA, following the FCC's 2024 ruling. Offer a route to a human, keep to agreed calling hours and honour do-not-call requests. Your legal team should confirm the rules for each region you call.
Read transcripts, not only tags. A yes/no tag tells you a customer was unhappy. The transcript tells you why.
Insight7 has a beta Voice AI feature for this. Managers build and launch outbound AI calling campaigns by setting a goal, an audience, the AI caller and the rules. A campaign can be a customer survey, with consent marking, calling hours, retry rules and a do-not-call list. You can test the agent in the browser, with no phone number needed. A dashboard shows calls in progress and lets a reviewer read each transcript. Calls where only the AI spoke are not counted as reached.
Run a small test first. Call a few dozen customers, read every transcript and drop any question that gets vague answers.
Start with five questions
A score tells you something is wrong. A question tells you what. This week, pick five questions for your worst-scoring segment. Write them into one short survey script for a small test batch of calls.

