Using the AI Churn & LTV tool

The Churn & LTV tool lives on the "Churn risk" tab of AI → Re-engagement (the old /app/ai/churn-ltv link still gets you there — it now redirects). You don't feed it anything customer-specific: you set a minimum risk score and how many customers to pull back, and it scans your own paying customers and returns a ranked list — each row with a risk score, an estimated lifetime value, a suggested next service, and a short reason to reach out. It's read-only and review-only: nothing here sends a message, books anything, or enrolls a customer in anything on its own. You take the list and decide who to follow up with, and how.

Scanning your customer base

Open the Churn risk tab, set your minimum risk score (0–100; it starts at 40) and the maximum number of customers to pull back (up to 50), and tap Find at-risk customers.

Behind that button, the tool reads your shop's own completed payments and completed appointments — nothing from any other shop, and nothing that hasn't actually happened. For every customer who's completed at least one payment, it works out how long it's been since their last visit, how often they've come in, and how much they've spent, and combines those into a 0–100 risk score: the longer since their last visit, the higher the risk, and that risk is weighted up for customers who've spent more or visited more often, so a high-value regular who's gone quiet ranks above a one-time walk-in from years ago. Only customers who clear your minimum score show up at all — a brand-new customer who just visited isn't "at risk," and someone who's never completed a paid visit doesn't have enough history to score.

Alongside the risk score, it estimates a lifetime-value figure from what each customer has actually paid so far, plus a modest projection based on how often they've come back — capped so one unusually large purchase can't blow the estimate out of proportion — and it works out a suggested next service based on what they've already had done, following your shop's usual progression from window tint through PPF, ceramic coating, and ongoing maintenance. All of that — score, LTV, and suggested service — is plain arithmetic over your own records, not AI, and it comes out the same whether or not your workspace has live AI output turned on.

Reading the results

Each flagged customer shows their name (linking to their profile), their last visit date, the estimated LTV, a color-coded risk badge, and a "Next: [service]" badge. Underneath is a one-line reason — that line is the only thing Claude actually writes here. It's handed the already-computed score, last-visit date, LTV, and suggested service for that specific customer and asked to phrase one short, plain-language reason to reach out; it isn't asked to and doesn't get to pick the score, the LTV, or the service — those are already locked in before the model sees anything.

Re-running the scan won't just show you the same names every time: customers who've already gotten a real outbound text or email from your shop recently drop out of the list automatically, so you're not repeatedly reminded to contact someone you already contacted. If nobody clears your threshold, that shows as a clean "no at-risk customers right now" state rather than an error — a reasonable sign of healthy retention, not a broken run. Lower the minimum score if you want to see further down the list.

There's no Send or Draft button on this tab — unlike some of the other Re-engagement tabs, it doesn't hand you a ready-to-send message. You take the name, the reason, and the suggested service back into whatever outreach flow you already use.

Things to know

• Plan requirement: Pro — it isn't available on lower tiers.
• Stub mode: this is a Pro feature, and on Pro it drops to stub mode once your shop has used its AI allowance for the current 30-day period (AI Booster credits can cover it). Then the tab shows a stub-mode notice, and every flagged customer gets a generic, formulaic reason line instead of one phrased by the model. The score, LTV, and suggested service are unaffected either way — they come straight from your data, not from the AI.
• Needs at least one completed (paid) visit: a lead who never converted, a no-show, or a canceled appointment won't appear here no matter how long ago they came in — the tool only surfaces customers with real history to win back.
• The suggested next service only tracks a handful of service categories toward the tint → PPF → ceramic → maintenance progression. Work outside that (flat glass, removals, and other miscellaneous categories) doesn't move a customer along it, so the suggestion can undercount what they've actually had done.
• Results are ranked by risk first, then by estimated value, and capped to however many you asked for — raising the "max customers" field is how you see further down the list, not lowering the risk threshold.

Frequently asked questions

Q: Does this tool contact my customers for me?

A: No. It's read-only and review-only — nothing here sends a message, books an appointment, or enrolls anyone in anything. It just gives you a ranked list to work from.

Q: What plan do I need to use it?

A: Pro.

Q: What makes a customer show up as "at risk"?

A: They need at least one completed, paid visit, and their risk score has to clear whatever minimum you've set (it starts at 40). The score itself rises the longer it's been since their last visit, and rises further for customers who've spent more or visited more often — a slipping regular ranks above a one-time visitor from years back.

Q: How is the estimated LTV calculated — can I trust the number?

A: It's built from what that customer has actually paid you so far, plus a modest, capped projection based on how often they've come back. Treat it as a rough "worth winning back" indicator, not a guarantee of future spend.

Q: What happens on Pro once my shop's AI allowance is used up?

A: You'll see a stub-mode notice, and the one-line reason for each customer becomes a generic, placeholder sentence instead of one phrased by the model. The risk score, LTV, and suggested service are unaffected — those are computed from your real data either way.

Q: I ran a scan and got zero results — is something broken?

A: Probably not. It means nobody in your customer base cleared your minimum risk score, which is a decent sign of healthy retention. Lower the minimum score if you want to see further down the list.