AI Calling Campaigns
Churn prevention calls: phone the customers about to leave
Churn prevention with phone calls: which warning signs deserve a call, what to say, what it costs per saved customer, and how to trigger calls from your CRM.
Churn prevention fails most often in the gap between a warning signal and the cancellation, when an email goes unread and nobody talks to the customer. A phone call closes that gap, but only for the signals where a conversation can change the outcome: a usage cliff, a complaint that went quiet, a repeat payment failure, a downgrade. Ask what changed before offering anything. At roughly 14–20¢ a talk minute with an AI caller, the break-even is one saved customer in a few hundred calls for most subscription businesses.
Read the top-ranking churn prevention guides and you'll find predictive analytics, onboarding, surveys, in-app nudges and loyalty programmes. Almost none of them mention picking up the phone. That's an odd gap, because the phone is the one channel where you find out, in the customer's own words, why they're about to go.
This article is about that channel: which churn signals deserve a call, what the call should sound like, what it costs, and how to fire it automatically from the tool that spots the signal.
The number everyone quotes, and what it actually says
Churn articles love the claim that a 5% increase in retention lifts profits by 25% to 95%. The source is Fred Reichheld at Bain & Company, and his original Bain brief is more modest than the quote: "In financial services, for example, a 5% increase in customer retention produces more than a 25% increase in profit." One industry, one illustration. The mechanism he describes, that returning customers buy more over time and cost less to serve, is the part worth taking away. Your own number comes from your own margins.
A second figure is more useful because it changes what you do. In Recurly's churn benchmarks (July 2026 network data), involuntary churn, meaning failed payments rather than decisions to leave, is 1.25 points of a 3.60% overall rate, about a third of all churn. Those customers didn't choose to go. Many of them will fix a card in a two-minute call.
Which churn signals deserve a phone call
Not every signal is worth a conversation. The test is whether a person, talking to the customer, could plausibly change the outcome. Here's how common signals sort out:
| Signal | Call? | Purpose of the call | Good outcome |
|---|---|---|---|
| Payment failed twice | Yes | Tell them before access lapses; point them to update the card | Card updated, no lapse |
| Usage dropped sharply over 2–4 weeks | Yes | Ask what changed; offer help or a pause | Re-engaged, or a pause instead of a cancel |
| Complaint closed without a reply from the customer | Yes | Check it was actually solved | Issue confirmed fixed, or flagged to your team |
| Downgraded plan | Yes, for high-value accounts | Understand why; check the new plan fits | Stays on the lower plan rather than leaving |
| Renewal in 30 days on an annual plan | Yes, for high-value accounts | Confirm they're getting value before the bill lands | Renews, or a known risk you can act on |
| Visited the cancellation page | Sometimes | Only if you can call within hours | Save offer accepted or reason captured |
| Opened fewer emails | No | Email engagement is weak evidence on its own | Leave it to your email programme |
| Hasn't logged in for a week | Depends | Normal for some products, alarming for others | Use your own baseline |
The last row is the important one. "Inactive" means different things for a daily app and a quarterly service. Define each signal against your own usage pattern before you attach a call to it.
What a churn prevention call should sound like
Most save calls go wrong in the first thirty seconds by leading with a discount. A discount offered to someone with a support problem fixes nothing and teaches them that complaining pays. The order that works:
- Who you are, and that you're an AI if you are. Disclosure is on by default in Dial Milo, and in the EU Article 50 of the AI Act requires people to be told they're interacting with an AI system from 2 August 2026.
- Confirm the right person, before naming anything account-specific.
- A reason they'll accept. "I noticed you haven't used the scheduling tool since mid-September and wanted to check everything's working."
- Ask, then listen. "Has something changed?" The answer is your churn research, captured verbatim in the transcript.
- Respond to the reason given. A support problem gets flagged for your team. A price objection gets the written save offer, presented once, with its exact terms. A "we just don't need it now" gets the pause option if you have one.
- Record the outcome. Accepted, declined, wants a callback at a specific time, opted out, or no decision.
Milo: Hi, this is Milo, an AI assistant calling from Brightdesk. Am I speaking with Priya?
Priya: Yes.
Milo: Thanks, Priya. I noticed your team hasn't used the booking calendar since mid-September, and I wanted to check it's still working for you. Has anything changed?
Priya: Honestly, we moved two people to another office and the calendar sync kept breaking.
Milo: That sounds frustrating. I'll note that for the support team so someone who can fix the sync gets in touch. Is this the best number for them to reach you?
Priya: It is, yes.
Milo: Done. Is there a good time for them to call?
Priya: Tomorrow morning, before eleven.
Notice what didn't happen: no discount. Priya wasn't leaving over price, and a discount would have cost money without fixing the sync. The call ends as "callback requested" with a time, the AI campaign report should list Priya under "needs attention", and the transcript says exactly what to fix. (Illustrative call; Brightdesk is a fictional business.)
Write your save offer, but tell the caller to use it only when the customer names price or value as the reason. If you're using Dial Milo, the offer and its exact terms go in the campaign, and Milo presents them once rather than repeating the pitch.
Triggering the call from the signal
A churn call made a week after the signal is a much weaker call. The practical answer is to let the system that spots the signal start the call.
Each Dial Milo campaign has a secret "start a call" webhook. Any CRM or journey tool that can send a webhook can use it: Customer.io, Braze, Optimove, HubSpot workflows, or an in-house system. The tool POSTs the person's phone number, name, an ID, the language to speak, and any context you want Milo to know, such as the plan they're on or the signal that fired. Re-fired journeys and duplicates are ignored, people who opted out are rejected automatically, and an optional at time schedules the call for later. The paced dialler still respects calling hours and pause.
With Customer.io the loop closes the other way too: after each call Dial Milo updates the person with attributes like milo_last_outcome and milo_callback_at and fires events such as milo_offer_accepted, milo_callback_requested and milo_opted_out, so your journey can branch on what the customer actually said. The full setup is in Customer.io AI phone calls.
If you don't have a journey tool, the manual route works fine: export the at-risk segment as a CSV each week and upload it to a campaign, or filter your contact book and add the group to a campaign.
What it costs per saved customer
Dial Milo is pay as you go. An outbound call is the phone line, priced for the exact number dialled and your caller ID and billed per started minute (US numbers from 4¢, UK mobiles from a UK number 8.5¢), plus AI at 9.57¢ per talk minute. That's roughly 14–20¢ a minute all in for US and UK numbers. Calls that don't connect have no minute charge. See pricing for the full picture.
Worked example (illustrative figures). A subscription software business has 4,000 US customers paying $49 a month. Each month its signals flag 160 as at-risk.
- Assume 40% answer within two attempts: 64 conversations. Existing customers recognise the business name more often than strangers do.
- Average talk time 3 minutes, billed as 3 line minutes: 64 × 3 × 4¢ = $7.68 line, 64 × 3 × 9.57¢ = $18.37 AI.
- Unanswered attempts and voicemails: allow $10.
- Total: about $36 a month.
If those calls save just 5 of the 64 people reached, that's 5 × $49 = $245 of monthly revenue kept, and more for every month they stay. Even at one save a month the calls pay for themselves. The point isn't the exact figure; it's that the threshold is low enough that the real constraint is your signal quality, not the call budget.
Don't turn retention into harassment
A customer who's been called three times in a fortnight by three different teams is more likely to leave, not less. Three controls matter:
- A contact cooldown across campaigns. Dial Milo's "don't call a list player again within N days" setting covers all your campaigns, not just one, so the churn campaign and the renewal campaign can't both ring the same person the same week. Details are in the calling hours and contact limits guide.
- Honour "stop calling". When someone says it, Milo records it as an opt-out and adds them to your do-not-call list, which no campaign will dial again.
- Calling hours in their time zone. The default is 10:00 to 20:00 local time.
And in the US, remember that the FCC treats AI-generated voices as artificial voices under the TCPA, so a call that's really a sales pitch needs the consent that comes with that. A save call that presents a discount is closer to telemarketing than a check-in that doesn't. This isn't legal advice; the AI phone call laws article covers the detail.
Reading what the calls tell you
The calls that don't save anyone are still worth having, because they're the cheapest churn research you'll ever commission. Dial Milo's AI campaign report labels every person with a next step (needs attention, follow up later, stop contacting, and so on) and writes a review of what customers asked about and pushed back on. If the same reason comes up in a third of calls, that's a product or pricing decision, not a retention one.
- 1
Pick two signals
Start with failed payments and a usage cliff. They're the easiest to define and the most likely to be fixable on a call.
- 2
Write the call around the question
Opening line, the reason for calling, "has anything changed?", and one written save offer for price objections only.
- 3
Wire the trigger
Point your CRM's webhook at the campaign, or upload the at-risk list weekly.
- 4
Set the guardrails
Contact cooldown, calling hours, a daily budget cap, and a test call to hear it yourself.
- 5
Review after 50 conversations
Read the report and the transcripts. Fix the signal definitions first, then the script.
Churn prevention by phone isn't new; account managers have always done it for their biggest clients. What's changed is that the same call can now reach the customer paying $49 a month, on the day their usage drops, for less than the cost of the coffee you'd drink while making it. For the wider outbound picture, the AI outbound calling guide is the place to go next, and the AI sales calls article covers the offer side in more depth.
Frequently asked questions
- What is churn prevention?
- Churn prevention is the work of spotting customers who are likely to leave and acting before they do. It starts with warning signals in your own data, such as falling usage, a failed payment, a downgrade or an unresolved complaint, and pairs each signal with an intervention. A phone call is the most direct intervention, so it's best kept for the signals where a conversation changes the outcome.
- When should you call a customer who might churn?
- Call when the signal suggests a fixable reason and the customer is worth the cost of a conversation: a sharp drop in usage, a complaint that went quiet, a payment that failed twice, a downgrade, or an approaching renewal on a high-value account. Avoid calling on every small dip; a contact cooldown stops the same customer being called again within a set number of days.
- Should a churn prevention call include a discount?
- Not as the opening move. Ask first what's changed, because the answer often isn't price, and a discount offered to someone with a support problem teaches them to complain for money. Keep a written save offer ready for customers who say price is the reason, and have the caller present it once with its exact terms.
- Can an AI make churn prevention calls?
- Yes, for routine check-ins and fixed save offers. An AI caller can disclose that it's an AI, confirm it has the right person, ask what's changed, present one written offer, and record an outcome such as accepted, declined, callback requested or opted out. Complaints that need judgement should be flagged for a person on your team.
- How much do churn prevention calls cost?
- With pay-as-you-go AI calling, roughly 14 to 20 cents per talk minute for US and UK numbers, and calls that don't connect have no minute charge. A three-minute save call to a US customer costs around 40 cents. Compare that with the margin you keep from one retained customer and the threshold for calling is usually low.
Keep reading
AI Calling Campaigns
AI outbound calling: how it works, costs and rules
AI outbound calling explained: what an AI phone agent does on each call, what a campaign costs per minute, the consent rules, and how to run your first list.
ReadAI Calling Campaigns
AI sales calls: where they work and where they don't
AI sales calls pay off on warm, consented lists and fail on bought ones. The law, a worked cost per sale, and a script structure you can run this week.
ReadAI Calling Campaigns
Automated calling system: the four types compared
An automated calling system can be a robocaller, an IVR, a dialer or an AI agent. What each does when a person answers, what the law says, and how to choose.
Read