Finding at-risk customers in RealGreen before they cancel

How do you find at-risk customers in RealGreen before they cancel? You score five signals your database is already collecting: renewal silence, payment lag, complaint history, shrinking service counts, and disengagement. In our experience, most cancels are not surprises: by the time a customer calls to quit, the data has been announcing it for months, and the cancel call is just the paperwork. This post walks through the whole system: a simple scoring model your office can actually maintain, and a specific intervention for each risk tier. Analyzing why customers already left is a different exercise, and a valuable one: we covered it in what your cancel reasons are trying to tell you. This is the before picture.

What signals does RealGreen already have?

You do not need new software to see risk. You need to look at fields you already have, together instead of separately:

• Renewal behavior. A customer who has not responded to a renewal letter or prepay offer that most of the base has already answered.
• Payment lag. Not delinquency: drift. A customer who paid in 12 days last year and 38 days this year is telling you something before the AR report does.
• Complaint history. Any service complaint in the last twelve months, weighted heavier if the resolution visit never happened.
• Declining service count. Dropped the seventh app, canceled the aeration, said no to the renewal upgrade. Shrinking customers are rehearsing the exit.
• Silence. No portal logins, no email opens, no replies all season. Disengagement is weaker as a single signal but potent combined with any of the above.

None of these alone predicts much. A late payer with no other flags is probably just disorganized. The combinations are where prediction lives.

Do you need a data scientist for this?

No, and be suspicious of anyone selling you a churn-prediction model with machine learning in the pitch before your office has mastered a points system. A weighted checklist gets you most of the value: complaint in last 12 months unresolved, 3 points; resolved, 1 point. Payment lag doubled year over year, 2 points. Dropped a service this season, 2 points. Renewal outstanding past your median response date, 3 points. Zero engagement all season, 1 point. Score every active customer weekly from your synced data. Then cut the list into three tiers: 0-2 healthy, 3-5 watch, 6+ at risk. The exact weights matter less than you would think. What matters is that the list exists, refreshes automatically, and someone owns it.

Two honest limitations. First, a scoring model built on dirty data scores the dirt: if half your cancel reasons are coded "other" and complaints live in a spiral notebook, fix the inputs first. Second, expect false positives. A third of your "at risk" tier was never leaving. That is fine: the interventions below are cheap and none of them damage a healthy customer.

What should each tier get?

Watch tier: automated warmth

Trigger: customer crosses into the watch tier. Action: they are enrolled in a light-touch sequence: a genuinely useful seasonal advisory, a service-satisfaction check-in with a reply path, suppression from upsell campaigns for 30 days so the next thing they hear from you is not a pitch. No discounts. Discounting at the first wobble trains your base to wobble.

At-risk tier: a human

Trigger: customer crosses 6 points, or lands any single catastrophic flag (unresolved complaint plus outstanding renewal, say). Action: a task lands on a named person's desk: call within two business days, script built around listening, with authority to fix problems on the spot. Log the outcome. Math: 2,400 customers × 8% flagged at-risk ≈ 190 accounts a season. If half of those would actually have churned and calls save a third of them, that is roughly 32 saved customers × $620 average program ≈ $19,800 in retained revenue for a few hours of calls a week. Illustrative numbers: plug in your own base, churn rate, and program value.

If the customer cancels anyway despite the call, the moment-of-cancel play is its own discipline with its own script: see cancel-save automation. And if they still leave, the clock starts on win-back timing. Risk scoring, cancel-save, and win-back are three stations on the same assembly line.

Why is prevention so much cheaper than replacement?

Because a replacement customer costs real acquisition money and starts at year-one economics, while a saved customer keeps compounding. Retention is the highest-margin marketing you can do, which is the argument of our broader piece on retention automation. The at-risk list is just retention with a targeting system: instead of treating all 2,400 customers as equally likely to leave, you concentrate human attention on the 190 who are actually drifting.

There is also a compounding intelligence benefit. Every logged save call teaches you why customers drift before they cancel, which is earlier and more honest data than exit interviews. After two seasons of this you will know, for your market, whether the leading indicator is payment lag or complaint recurrence, and your weights get sharper.

How do you keep the list alive?

The failure mode is not building the list. It is the list dying in week six when the season gets busy. Three rules keep it breathing. The score must refresh without human effort, off your nightly sync, because any list that requires a manual export dies by June. The at-risk calls must belong to one named person with a weekly number reviewed in your regular meeting: "how many at-risk calls happened, what did we learn, what did we save." And the tiers must gate your other marketing: an at-risk customer should be automatically suppressed from aggressive upsell sequences until they stabilize, which is a routing job your automation layer should handle without anyone remembering to do it. None of this plumbing is exotic: we can automate pretty much anything off the nightly sync, and connect RealGreen to whatever platform your office already lives in, native connection, API, or Zapier.

The takeaway: your database announces most cancels months in advance: late payments, quiet complaints, shrinking programs, unanswered renewals. A weekly score, three tiers, and one owned call list turn those announcements into saves instead of exit paperwork.

Your first 30 days

1. Pick five risk signals you already trust in your data and assign simple point weights to each.
2. Build the weekly scored list off your synced RealGreen data, split into healthy, watch, and at-risk tiers.
3. Wire the watch-tier sequence: seasonal advisory, satisfaction check-in, 30-day upsell suppression.
4. Assign at-risk calls to one named person with a two-business-day standard and a logging habit.
5. Review saves, misses, and false positives monthly, and adjust the weights once a quarter, not weekly.