A customer activation call playbook is a plan for phoning people who signed up but stalled before their first key action. Give each stalled segment one call goal, a short script and a capped retry cadence. Measure results against a holdout group.

Segment customers by where they stalled

Start with five segments, one for each place customers leak out of the funnel.

  • Signed up, never funded. The account exists but holds no money.
  • Started KYC and stopped. The identity check is half done.
  • Abandoned an onboarding step. The customer hit a form, an app download or a verification and left.
  • Funded but inactive. The customer paid in once and never used the product.
  • Dormant or win-back. The customer used the product before and went quiet.

Build each segment from an event. Your CRM or app logs "KYC started" with no "KYC completed" after a set wait, and that event adds the person to a call list. Pick the wait for each segment. A day may suit a KYC stall. A week may suit a funded but inactive account.

Then split customer-caused drop-off from product-caused drop-off. A lending team we spoke with tracked app download, KYC tier one and loan process started as separate wins. That let them see whether a failed final step meant lost interest or a product limit. Calling someone about a step your product breaks wastes the call.

Our advice is to start with the KYC-stalled segment. It has one clear next step, and its cause is the easiest to fix. Move to reactivation calls once that segment works, because those customers are the coldest.

Give every call one goal

Give each call one goal, and make it a yes/no result. A short call with one ask gets a clear answer. The system can read that answer from the transcript and tag the contact. A call that tries to fix KYC, pitch a loan and run a survey ends with no clear tag.

Segment Call goal Single next action CRM outcome tag
Signed up, never funded Learn what stopped the first deposit Agree a funding date funding_committed
Started KYC and stopped Get KYC finished Open the app to the ID step kyc_resumed
Abandoned an onboarding step Remove one blocker Retry the step today step_retry_agreed
Funded but inactive Prompt a first use Make one transaction first_use_agreed
Dormant or win-back Find out why they left Accept an offer or decline winback_yes / winback_no

Start with outbound. Much inbound volume exists because companies never send proactive updates.

Decide the handover before launch. Buyers ask what happens with sensitive issues or a disputed amount, and whether anyone chases a handover nobody picked up. Send disputes, fees and complaints to a person with a queue and an owner.

Write a short script for each segment

Every script has five beats: say it is an AI, give the reason for the call, make the single ask, answer one objection, and offer a human. Keep each call to about two minutes. Say it is an AI up front, because EU AI Act Article 50 now requires it for AI that talks to people.

Use this base and swap the middle:

  • Opening, all segments: "Hi, this is an AI assistant calling for [Company]. Is this [Name]? I'm calling about your new account. Do you have two minutes?"
  • Never funded: "Your account is ready but has no deposit yet. Can I help you pick a day to add your first funds?" Objection "I'm not sure yet": "That's fine. What's the main thing holding you back?"
  • KYC stalled: "Your identity check is almost done. Can you open the app now and finish the ID step?" Objection "No time": "It takes a few minutes. Can I call back this evening?"
  • Onboarding step: "You stopped at [step]. Was something unclear or not working?" Objection "It kept failing": "Thanks. I'll pass this to our team today."
  • Funded but inactive: "You've added funds. Would you like to try one small transaction this week?" Objection "I don't see the point": "What would you want it to do for you?"
  • Win-back: "We haven't seen you for a while. Can you tell me why you stopped?" Objection "Not interested": "No problem. Should we stop calling?"
  • Handover line, all segments: "I can connect you to a person now, or have someone call you back. Which do you prefer?"

Test the script by talking to it before launch. Listen for robotic phrasing and cut any line longer than two sentences.

Set a cadence with retries and windows

Treat three to four attempts per contact, spread across different times of day, as a starting point. Test retry timing and morning, evening and midday windows against your own answer data before you lock them in.

Rule Starting point
First call 1 day after a KYC stall; about 7 days after funding with no use
Retry gap A few hours for the same-day retry, then the next day
Max attempts 3 to 4, then stop
Window One morning and one evening slot, tested against midday
Stop rules Goal tag set, opt-out, wrong number, or max attempts reached

Keep every window inside the legal limit. Quiet hours differ by region. In the US, the eCFR text of 47 CFR 64.1200 bars telephone solicitations to residential subscribers before 8 a.m. or after 9 p.m. Other regions set their own hours, so check each market.

Mark consent on every contact and keep a do-not-contact list. Remove anyone who opts out from all later retries.

Sequence SMS with care. If you text and call the same person in one week, you cannot tell which worked. Hold SMS for a voice-only sublist, or send it after the last call attempt.

Insight7's voice campaigns, which are in beta, can carry much of this setup. A campaign has a type (customer survey, lead follow-up, cold outreach or win-back), plus consent marking, calling hours, retry rules and a do-not-call list. You add contacts by CSV upload or by typing them in. You can test the agent by talking to it in the browser, with no phone number needed. Calls where only the AI spoke are not counted as reached, and a carrier's temporary refusal does not use up a retry. A duplicated campaign leaves out anyone added to the do-not-contact list since the original. A dashboard lets a reviewer read each transcript when the call ends, and yes/no answers and call outcomes are pulled from it.

Judge success against a holdout group

Measure a funnel of five numbers, and compare it with people you did not call. Without that comparison, you cannot tell calls from coincidence.

  • Reached rate: the share of contacts where a person actually answered. Calls where only the AI spoke do not count.
  • Real conversation rate: answered calls where the customer spoke back.
  • Goal-completed rate: the share that got your yes/no tag.
  • Activation within X days: the share who did the key action, such as funding. Pick X per segment.
  • Cost per activated customer: total campaign cost divided by activations. Set it against what you pay to acquire a customer.

For attribution against SMS, carve out a voice-only sublist that gets no texts, so any conversion from it is clearly the call. Hold back a second group that gets no contact at all.

Report weekly: segment, attempts, reached, outcomes, top objections and handover count. Then read the flagged transcripts.

Start with a small pilot on one segment. Banks want a short, low-risk pilot on a small sample, with payment only after a measured lift. Gartner found only a quarter of 432 AI service use cases show positive ROI, and 42% cannot be measured. A holdout group gives you that proof.

Related reading: How to Find the Biggest Drop-Off Points in Your Customer Journey Before Launching AI Calls, 12 Customer Lifecycle Moments Where Automated Outbound Calls Get Better Results Than Email and How to Recover Failed Payments and Expired Cards With AI Outbound Calls.

Questions people also ask

What is a customer activation call?

It is a call to a customer who signed up but has not done the key first action, such as funding an account or finishing KYC. The call has one goal and one next step.

Can voice AI make outbound activation calls?

Yes. Voice AI can call lists built from stalled-funnel events, run a short script, tag the outcome and hand off to a person. Start with a small pilot and check consent rules first.

Do you need consent for voice AI outbound calls?

Often yes, and rules vary by region. In the US, AI-voice calls already need consent under the FCC's 2024 ruling. Record consent per contact and check each market's law with counsel.

Should the AI say it is an AI?

Yes. Say it in the first sentence. EU AI Act Article 50 has applied since 2 August 2026 to automated phone systems, and it was not delayed.

How many times should you call a customer?

Start with three to four attempts across different times of day, then stop. Test that cap against your own answer rates and complaint levels before you scale it.

How do you measure whether activation calls worked?

Compare activation rates between called customers and a holdout group you did not contact. Look at activation within a set number of days and cost per activated customer.

When should the AI hand over to a human?

Hand over on disputes, fees, complaints, sensitive issues or any request for a person. Give the customer a choice of live transfer or callback, and make sure someone owns the queue.

Start with one goal per segment

Activation calls win with one goal per segment, an honest script, a capped cadence and a holdout group. Before you start, check that every contact on your first list has consented to AI calls.

Book a demo