A voice of the customer program is a steady routine for asking customers what they think, then acting on it. With AI phone calls, you can ask on a schedule. Insight7's voice campaigns, in beta, carry a type, calling hours and retry rules.
Start with one question and one owner
Pick one business question before you pick any technology. A good one is "Why do people sign up and never fund?" or "What do customers think of the install visit?" A program that tries to learn everything learns nothing in particular.
Then build a small skeleton:
- One question. Write it in a sentence a director could repeat.
- One call type for that question. In Insight7's voice campaigns, which are in beta, a campaign has a type: customer survey, lead follow-up, cold outreach or win-back. Add your own labels, such as onboarding drop-off or post-service check-in.
- One named owner per theme. Product owns feature gaps. Operations owns delivery. A person owns each theme, not a team inbox.
- One success metric. Pick a fix you can count, such as KYC completions or fewer repeat complaints.
Why use the phone? People say more out loud than they type into a form. When you write the script, design in a follow-up for vague answers, such as "You said it was confusing. What part?" A tap on a 1 to 5 scale cannot do that.
Where should you start? We advise starting with outbound. Much inbound volume exists because companies never send proactive updates. Lenders, fintechs and investment firms leak customers at the same few points: people sign up but never fund, or start identity checks and never finish. Operators outside banking want proactive check-ins and surveys before a complaint arrives.
Some customers prefer a message to a call. Ask early, and keep a text option for those who decline.
Set call cadence for each call type
Give each call type its own trigger and rhythm. A win-back call and a post-install check-in need different timing.
| Call type | Trigger | Rhythm |
|---|---|---|
| Post-service check-in | Install or visit completed | Once, 2 to 5 days after |
| Onboarding drop-off | Sign-up with no funding or KYC after a set wait | Once per drop-off event |
| Relationship check-in | Calendar | Quarterly |
| Win-back | Customer lapsed | One-off round, then stop |
Set the attempts and hours next. A consumer lender we spoke with tested its calling times and found evenings answered better than midday. It also settled on 3 to 4 attempts per contact, because many customers only picked up on the second or third try. Test your own audience, since habits differ by sector and region.
Put the rules into the campaign settings:
- Calling hours and retries. Insight7's voice campaigns, which are in beta, include calling hours, retry rules and a do-not-call list.
- Carrier refusals. A carrier's temporary refusal does not use up a retry.
- Real reach. A call where only the AI spoke does not count as reached. Your reach rate then reflects real conversations.
Add a fatigue cap for each purpose. One customer should not hear from you more than once per quarter for the same purpose, so trigger calls should skip anyone called recently for that reason. A post-service check-in and a quarterly relationship check-in are different purposes. Adjust the rule after you read opt-out rates.
Start small. Buyers want a short pilot on a small sample and payment only after a measured lift. Run one batch of a few hundred contacts, read the transcripts, fix the script, then scale.
Score and tag every call the same way
Use one short theme list and one scoring rule for every call, so the results can be compared. Start with four themes: price, product gap, service and trust. Add a sentiment label and a one-line verbatim for each call. Your product owner can then read ten quotes instead of ten transcripts.
Check the tags against people before you trust them. Have two reviewers tag about 20 calls by hand. Compare their tags with your tagging tool's, and fix the theme definitions wherever they disagree.
A checklist score can hide what matters. A call can follow every step and still leave the customer angry. One critical error should fail the whole call. Flag critical issues, such as a threat to leave or a complaint about fees, as their own tag. Do not average them into a score.
Insight7 supports the review side of this step. In its voice campaigns, which are in beta, a dashboard shows calls in progress and lets a reviewer read each transcript when the call ends. Yes/no answers and call outcomes are pulled from the transcript. Every call can also be scored against a weighted scorecard your team owns, not a sample. Alerts can watch keywords and low scores. To see recurring pain points in calls you already have, try the VoC Feedback Analyzer, which surfaces them from uploaded transcripts.
Route each insight to a named owner
An insight that reaches no one changes nothing. Before the first call goes out, give every theme an owner, a speed and an expected action.
| Theme | Owner | Speed | Expected action |
|---|---|---|---|
| Threat to leave, fee complaint, request for a human | CX lead | Same day | Call the customer back |
| Risk or compliance concern | Compliance | Same day | Review the call, decide on a pause |
| Delivery or service issue | Operations | Weekly digest | Fix it and log the fix |
| Pricing or trust | Collections or sales lead | Weekly digest | Change the script or offer |
| Product gap | Product | Monthly review | Rank it, then accept or decline |
Urgent routes need a person, not a shared inbox. Buyers ask on their own who chases a handover that nobody picked up. Name a backup owner for every same-day route. A Gartner survey of 3,566 customers found that 87% say a company using GenAI must offer a way to reach a human.
For the slower routes, a weekly digest of the weakest calls keeps owners current without a flood of alerts.
Close the loop on every item. The owner replies with a decision (fix, defer or decline) and a reason, and the customer hears what changed.
Then prove it worked. Log each fix with a date, and ask the same question in the next call round. Training buyers want proof that behaviour changed, not that a session happened. Gartner's analysis of 432 AI customer service use cases found that only a quarter show positive ROI, and 42% cannot be measured. A before-and-after answer from your own customers closes that gap.
Govern the program from day one
Write the rules down before the first call, and give one person the right to pause a campaign.
- Consent and disclosure. Insight7's beta voice campaigns include consent marking. Whatever tool you use, mark consent on every contact. Have the agent say it is AI at the start. Rules differ by region, so see the questions below.
- Data storage and access. Buyers ask where recordings live, which AI models touch them, and who owns them. Record the answers. Limit transcript access to the program owner, theme owners and reviewers.
- Tag calibration. Repeat the 20-call human check each quarter and log any change to theme definitions.
- Monthly review. The program owner meets theme owners to review volumes, opt-outs, open items and fixes logged.
- Retention. Set how long audio and transcripts are kept, and delete on schedule. Honour do-not-call requests at once. Insight7's voice campaigns, in beta, include a do-not-call list.
- Pause rights. The program owner or compliance can pause a campaign after a spike in complaints, opt-outs or silent calls.
Related reading: 15 Customer Feedback Questions to Ask on AI Survey Calls That Your NPS Score Misses, 20 Churn Interview Questions for AI Exit Calls With Cancelled and Inactive Customers and How to Turn Customer Call Transcripts Into Product and Marketing Decisions.
Questions people also ask
What is a voice of the customer program?
It is a repeating loop: ask customers what they think, tag the answers the same way, send each theme to an owner, fix the issue, then ask again. The loop separates a program from a one-off survey.
How do you build a voice of the customer program?
Choose one business question, one call type, and one owner per theme. Set the cadence and a shared tag list. Pilot with a small batch, fix the script, then widen the audience.
What is voice of customer data, and where does it come from?
It is anything customers say about your company: calls, surveys, reviews, tickets and chats. With AI phone calls, the data is a transcript plus pulled yes/no answers and call outcomes.
How often should you call customers for feedback?
Match the rhythm to the trigger. Call once after a service event, once per drop-off, and quarterly for relationship check-ins. Never call one person twice for the same purpose in a quarter.
Do you have to tell customers an AI is calling?
Plan to disclose everywhere. In the EU, AI that talks to people must say it is AI from the first interaction. In the US, the FCC's 2024 ruling brings AI voices under the TCPA's consent rules. Colorado adds a disclosure duty from 1 January 2027.
Can AI phone calls replace customer surveys?
For most reasons-why questions, yes. Calls let you ask open questions of many customers. Keep a one-question survey for a quick score trend, then use calls to explain the movement.
Who should own a voice of the customer program?
Give one program owner, usually in CX or operations, the cadence, the tag list and the monthly review. Theme owners in product, operations and compliance act on what is routed to them.
Ask often, ask the same way
Hidden customer truth appears when you ask often, ask the same way, and send each answer to someone who must act. This week, pick one call type, such as a post-service check-in, and name its owner, themes and pilot batch.

