Churn prediction for service businesses and agencies: which signals appear 60 days out?

What is churn prediction for a service business?

Churn prediction is reading the signals that come before a cancellation early enough to act. In a service business, an agency or a coaching program, those signals already sit in your CRM, your help desk, your billing and your call notes. Statisfy, a company that sells churn software, places the first of them 60 to 90 days before the cancellation email (Statisfy, 26 June 2026).

You searched this because a client left, and if the email landed on a morning when nothing on your calendar hinted at it, the first sign of it was already sitting weeks earlier in a system you pay for every month, in a login count that slid, in a reply that took four days where it used to take one, in a champion who stopped turning up to calls, which means the question in front of you is less whether the warning existed and more who was assigned to read it, isn't it?

Do you need machine learning to predict churn?

No. A written list of signals, each routed to a named person with a defined outreach, is enough to start predicting churn. The pages that rank for this term are machine learning papers and analytics vendors, built for companies with tens of thousands of accounts and a data science team to model them.

You know your 100 clients by first name, so the model you need is shorter than a spreadsheet tab, and the nine signals further down this page are the whole of it, which leaves one practical question: which of them can you read from your own CRM tomorrow morning, before the first coffee goes cold?

What is the ONBOARD to RETAIN seam, and how do you measure it?

The ONBOARD to RETAIN seam is the handoff between your onboarding team and your account management team, and two rates measure it. The first is days of warning before each cancellation. The second is the share of cancellations where a warning signal was already sitting in one of your systems.

It meets every test of a revenue seam: capable people sit on both sides, nobody owns the handoff itself, it divides into a rate, and that rate appears on nobody's dashboard. You can put both numbers on paper this afternoon.

1. List every cancellation from the last 12 months with the date the client told you.
2. For each one, find the earliest signal in your CRM, help desk, billing or call notes, and write its date beside the cancellation.
3. Subtract the two dates: the gap is your days of warning on that account.
4. Count the cancellations where any signal existed, and divide by the total to get your share.

Run it on your last three losses and the two numbers arrive as a pair you have never seen side by side, a warning that sat for weeks next to a response time of zero, and once that pair is on the page the sentence inside your company changes from "they left" to "we had the warning and no name beside it," which is a sentence you can do something about by Friday, wouldn't you say?

What does a churn seam sound like inside a company?

A churn seam sounds like surprise after the fact. The Thirteen Seams lists three sentences owners say when Seam 09 is open:

• "They cancelled out of nowhere."
• "Nobody flagged anything."
• "We do a QBR every quarter, that's our early warning."

Each one is true from where its speaker sits, because onboarding handed the account over at day 30 with every box ticked, account management was waiting for the renewal window to open, and the months between carried no calendar entry, no report and no name, so when you hear "out of nowhere" you are hearing the exact coordinates of the long middle, aren't you?

Who owns the customer between onboarding and renewal?

Nobody owns the customer between onboarding and renewal in most service companies. Onboarding owns the first 30 days, account management owns the renewal date, and the stretch in between belongs to neither team. The thirty days before it is the onboarding handoff, Seam 08, which ends the moment this one begins. What happens in those thirty days carries straight into this one: in the Onramp report of 7 May 2025, a survey of 161 customer success and onboarding leaders, 57% of companies that reduced onboarding spend saw churn rise within six months.

Whose name sits on the account in month five, and when did that person last speak to the client without an invoice or a renewal being the reason? If the honest answer is a team name, the account has a contract and no keeper, and every signal it gives off lands in a queue that someone clears instead of a list that someone reads.

Why is a quarterly business review not an early warning?

A quarterly business review is not an early warning because it arrives every 90 days and the first signals appear 60 to 90 days before a cancellation. A review on that cadence meets most decisions after they are made.

If your QBR lands every 90 days and the signal starts two to three months ahead of the email, how many of last year's departures had already been decided by the time you opened the slide deck? The review is a good meeting for looking back over a quarter, and the decision to leave forms in the weeks between two of them, in the inbox and the help desk, where a weekly reader finds it.

How many days of warning did you have before your last three cancellations, and where was each signal sitting?

What does losing customers cost a $5,000,000 business?

Losing 20% of 100 clients at $50,000 a year costs a $5,000,000 business $1,000,000 a year in recurring revenue. Saving 5 of those 20 with 60 to 90 days of warning keeps $250,000. The table uses round inputs so you can replace each one with your own.

If 20 of your 100 clients move to leave this year, what is keeping five of them worth to you, and which five would you call first? The $250,000 is the price of weeks your company already holds and does not use, because the warning arrives either way and the only variable is whether it reaches a person who picks up the phone. The clients who leave anyway have a scheduled next offer of their own on the win back path for cancelled customers, Seam 04.

"A dollar left on the table is not a rounding error. It is dinner." Jon Weberg

How much of churn is failed payments rather than a decision?

Failed payments account for 16% of churn at B2B companies and 24% at B2C companies, according to Stripe data reported by Churnkey (Churnkey, 14 November 2025). Business services sit at 26%. That slice is a payment problem, and it belongs to Seam 07, failed payment recovery, where retries and a named owner bring it back.

Payment delays also sit on the list of nine warning signals, which is where the two seams share a border: a card that declines twice in a quarter is a billing event to finance and a mood to the client, and the owner of the long middle is the one person who reads it as both.

How do you close the churn seam by hand?

You close the churn seam by hand in five steps, and the first is naming one owner for the long middle.

1. Name an owner for the months between onboarding and renewal, by name, with the authority to call any client.
2. List the warning signals already sitting in your CRM, help desk, billing and calendar.
3. Route each signal to a person with a defined outreach and a 48 hour clock.
4. Count days of warning on every cancellation, including the ones you save.
5. Review the list weekly in a 20 minute meeting with that owner.

The order carries the result, because a list of signals with no owner becomes one more report nobody opens, while an owner with no list spends the week guessing, and the moment both exist in the same chair the long middle has a keeper who can tell you on any Monday which three clients need a call and why. None of this saves a client whose business closed or whose budget disappeared, and it does not turn a weak result into a strong one. What it gives you is the weeks, and the weeks are where every save happens.

Which signals predict a cancellation 60 to 90 days out?

Nine signals predict a cancellation 60 to 90 days out: a usage drop, a champion going silent, a spike in support tickets, a falling NPS, stakeholder turnover, onboarding never finished, payment delays, no QBR scheduled and competitor research. Statisfy sets the usage threshold at a drop above 30% month over month and places the window where outreach changes the outcome at roughly days 0 to 45 after the first signal. Statisfy publishes no sample size, so treat the list as vendor guidance and your own days of warning as the proof.

When your champion went quiet on the last account you lost, who noticed first, and who did they tell? Read down the second column and you will find that every row names software you already run, so the prediction you came here for is already installed in your business and waiting for someone to be handed the reading of it.

Statisfy's same article of 26 June 2026 sketches the 90 days as a sequence, vendor guidance with no sample behind it, and it places the budget decision both after day 60 and at day 70, so read the days as a shape rather than a schedule:

If the decision lands somewhere around day 70 and your outreach window closes near day 45, how many of the accounts on your list this morning are already past the second date?

What are the signs a client is about to leave?

Four signs show a client is about to leave: communication turns scarce, they voice dissatisfaction with results, they question whether the service fits, and meetings turn tense. AgencyAnalytics names lack of results, too little communication and no perceived ROI as the top reasons agency clients leave (AgencyAnalytics, 28 June 2024, a source more than 18 months old).

You have heard each of these in a client's own voice, the "let me get back to you" that used to be an instant yes, the question about what exactly the retainer covers, the call where the camera stays off, and each one reached somebody on your team who filed it as a bad day instead of a date, didn't it?

What do you do in the 30 days after a signal fires?

In the 30 days after a signal fires, the owner makes contact inside 48 hours, names the change out loud and books one results review inside two weeks. The contact is a conversation about the client's numbers, not a discount, and the review ends with one agreed next result and a date for it.

Picture the Tuesday it works: the owner opens the list, sees a login count that fell 34% since the last review and a ticket left unanswered for six days, calls before lunch, hears that the client's operations lead resigned a fortnight ago, and books a session with the replacement for Thursday, so that by the time a budget meeting happens on the client's side, you are the vendor who noticed first, aren't you?

What does an AI employee do inside the churn seam?

An AI employee reads the signals every day and sends the owner a ranked list with the outreach due. Each morning it pulls usage, ticket, payment and calendar data, flags every account that crossed a threshold, drafts the first message in the owner's voice and logs days of warning on each cancellation. The owner still makes the call.

That is the job description you write when you hire an AI employee into this seam, and it comes after the hand-run version has held for a few weeks, because the AI employee follows the routing you proved, at a speed no person holds by hand, across every account at once, every single morning.

"A system built on a broken handoff simply runs the leak faster." Jon Weberg, The Thirteen Seams

What does a closed churn seam look like?

A closed churn seam looks like one system where the lead, the customer and the payment share a record, with a named person reading it. Jon Weberg built that kind of system at AI Acquisition: a $70,000,000 a year business running on the CRM Jon built, still running today. Jon served as Fractional President directing the Sales, Marketing and Media Buying Directors.

When the record of who bought, who paid and who stopped showing up lives in one place, and the person reading it sat in a seat that directed the Sales, Marketing and Media Buying Directors and met every department head, the long middle stops being a blind spot and becomes a list, much as Coach Kyle's live launch of 30 August 2026, with 412 registered, 142 live, 37 applications, 30 calls booked, 4 closes at $6,800, ran only after 20 of 57 messages with broken tracking were repaired first, since a signal that never reaches a reader looks exactly like no signal at all, which is the same move you can make on a $5,000,000 book this quarter, and the case studies carry the ledgers from the other seams where the same move was made.