📞 AI voice calling

NPS and feedback surveys by AI voice call

Updated 2026-09-19 · 5 min read · By the Convo360 team

In short

An AI voice survey calls a customer shortly after a purchase or service visit, asks one scored zero-to-ten question, then follows up with a short open question depending on the score. Calls stay under a minute, capture the score and comment, and log both to a sheet or CRM. On Convo360 this is built with a gather block for the score and a Condition branch for the follow-up.

Key takeaways
  • Keep the survey to one scored question plus one follow-up; longer calls get abandoned.
  • Branch the follow-up by score: ask promoters what they loved, ask detractors what went wrong.
  • Call within a day or two of the purchase or visit while it's still fresh.
  • Log every score and comment to Google Sheets or CRM automatically, don't rely on manual notes.
  • Respect calling hours and let the customer decline or skip a question without pressure.

Survey links get ignored, especially on mobile where opening a form feels like effort. A short voice call, timed right after a purchase, delivery or service visit, gets a higher response simply because answering 'how would you rate us, zero to ten' out loud takes five seconds; typing it into a form takes longer.

Voice also captures tone a form can't. A detractor's frustration or a promoter's genuine enthusiasm comes through in what they say next, useful raw material even if you're only logging the number.

Voice surveys also work better for customers who are less comfortable with typed forms in English, since the call can run in their preferred Indian language, whereas a web form usually stays in whichever language it was built in.

The one-question-plus-one script

Keep it to two turns.

  1. The scored question: 'On a scale of zero to ten, how likely are you to recommend us to a friend or colleague?'
  2. One follow-up, chosen by the score: for a high score, 'What did you like most?'; for a low score, 'What can we do better?'

Thank them and end, don't ask a third question. A voice survey that runs past a minute starts losing people mid-call, and the extra data rarely justifies the drop-off. If someone wants to say more, let them, but don't prompt for it.

Open the call with a one-line reason for calling before the scored question itself, for example 'Hi, this is Convo360 calling on behalf of [business] about your recent order, this will take less than a minute.' Skipping this and jumping straight to 'on a scale of zero to ten' tends to confuse people who don't yet know why they're being asked.

Capturing the score and branching by response

Capturing a spoken number reliably needs a bit of care. Ask for digits clearly ('zero to ten') rather than an open scale, and accept both spoken numbers and DTMF key presses for people who'd rather press than say a number out loud.

Once you have the score, branch the follow-up.

ScoreCategoryFollow-up
9-10Promoter'What did you like most?', then optionally ask for a public review
7-8Passive'What would make it a 10 for you?'
0-6Detractor'What went wrong?', then offer a transfer to a human if the issue needs resolving

Detractor calls matter most, don't just log the low score and move on, offer to connect them to someone who can actually help.

If the caller gives an answer that isn't a clean zero-to-ten number, for example 'pretty good I guess', ask once for a number rather than guessing a score from the wording, and if they still don't give one, log the comment without a score rather than forcing a value that wasn't actually given.

Scoring capture: NPS math without the spreadsheet gymnastics

Once scores are flowing in, the standard NPS calculation is simple: the percentage of promoters (9-10) minus the percentage of detractors (0-6), ignoring passives (7-8) in the subtraction. You don't need special software for this, a spreadsheet formula over the logged scores does it, and Google Sheets already holds the raw data if that's where you're logging calls.

Track the score over time rather than as a single snapshot, a monthly view shows whether a change you made, a new packaging process or a revised script, actually shifted how customers respond, instead of relying on the feel of a handful of recent comments.

Building it on Convo360

To build this in the Flow Builder:

  1. Trigger the survey as an outbound campaign a day or two after a purchase or service event, or from a payment trigger for order-based businesses.
  2. Use a gather block to ask the scored question and capture either a spoken number or a DTMF menu key press.
  3. Add a Condition block that routes to a promoter, passive or detractor follow-up question based on the score.
  4. For detractors, add a transfer block offering to connect them to support right away.
  5. Log every score and comment with a Google Sheets integration, one row per call, so your team can review trends without listening to every recording.
  6. Use Save CRM lead to flag detractors for a manual follow-up if they didn't want to be transferred on the call itself.

Every call is also transcribed with an AI summary, useful for a quick scan of comments without opening the sheet. If your customer base spans several Indian languages, set the agent's voice and language per campaign rather than running one script and hoping callers follow along in a language they're less comfortable with.

Response etiquette and timing

Call at a reasonable hour and skip anyone who has opted out of calls. If someone declines to answer the scored question, thank them and end the call, don't push for a number. Keep survey calls occasional, once per purchase or visit is enough; calling the same customer repeatedly for feedback wears out the goodwill a survey is meant to build. See the DND and calling hours guide before running survey campaigns at volume.

Avoid stacking a feedback call on top of other outbound calls the same customer may already be getting, a delivery confirmation and a survey call on the same day about the same order reads as excessive. Where possible, route both through the same campaign so a customer gets one well-timed call instead of two.

Step by step

  1. Trigger the survey after the eventLaunch a campaign a day or two after purchase, delivery or a service visit.
  2. Ask the scored questionUse a gather block for a zero-to-ten question, accepting spoken numbers or DTMF.
  3. Branch by scoreRoute promoters, passives and detractors to a different one-line follow-up question.
  4. Offer a transfer for detractorsLet low scorers connect to a human right away instead of just logging the complaint.
  5. Log every responseSend the score and comment to Google Sheets or save it as a CRM lead.

Example conversation

I'd say an eight, delivery was fine but the packaging was a bit crushed.
Thanks for that, noted on the packaging, we'll look into it. Is there anything else that would have made it a ten?

Launch your first feedback call survey

Start free in minutes — no code, no credit card. Bring your own AI keys and pay no per-message markup.

Frequently asked questions

How long should an AI voice survey call be?

Under a minute is ideal, one scored question and one follow-up. Longer calls see people hang up before finishing, so keep it to two turns and end promptly.

Can the AI capture an NPS score accurately over a call?

Yes, using a gather block that accepts both a spoken number and a DTMF key press, which covers callers who find either method easier. If the answer isn't a clean zero-to-ten number, the agent asks once more rather than guessing a score from vague wording.

What should happen when a customer gives a low score?

Ask what went wrong, then offer to transfer the call to a human who can help resolve it, rather than just logging the score and ending the call.

Where do survey responses get stored?

They can log automatically to a Google Sheet or as a CRM lead entry, one row per call, so your team can review scores and comments without replaying recordings.

Is it okay to call every customer for feedback?

Only if they haven't opted out and you're not calling too frequently. Respect DND registration and calling hours, and skip anyone who declines to answer during the call.

Related reading

Available across India — see WhatsApp & AI calling automation by state and city.