Somewhere in your records is a woman in Needham whose furnace you serviced two Octobers ago. She was glad you came, paid without blinking, and said she'd call you next time. Then life happened, and so did yours. You haven't thought about her since, and odds are she hasn't thought about you either.
Multiply her by a couple hundred. That's the quiet pile most shops are sitting on: people who already trusted you with their money, scattered across old invoices and job histories, unbothered for a year or more. Chasing a brand new lead means buying ads, fielding tire-kickers, and proving yourself from scratch. Reaching back out to someone you already took care of skips every bit of that.
Here's the part that used to make this a chore. Combing through a year of records by hand is the kind of job that never gets done, because there's always a truck to load or a quote to send. This is where AI actually earns its keep. It can read your history, pull the people who've gone quiet, sort them by what they bought, and hand you a short list worth texting. You stay in charge of the message itself.
Pull the list of who you haven't seen
Most of what you need is already in your invoicing or scheduling software. QuickBooks, Jobber, Housecall Pro, or even a spreadsheet you've kept by hand, almost all of it can export to a file. You want a plain export with the basics: name, phone or email, what the job was, and the date you last did work for them.
Hand that file to an AI tool and ask it a plain-English question. Something like, show me every customer I haven't done a job for in the last twelve to eighteen months, newest at the top. In a few seconds you get back a list that would've eaten an evening with a highlighter.
Tell it what else matters so the list is useful, not just long:
- Anyone with no visit in the last 12 to 18 months
- What they last bought or had done
- Roughly what they spent, so you know who's worth a phone call instead of a text
- Whether they ever left a complaint, so you lead with a little more care there
One honest warning. If your records are messy, the list will be messy too. AI reads what you give it, so a customer entered three different ways might show up as three people. Give it a quick human eye before you trust it.
Group them by what they last bought
A blast of the same text to everybody reads exactly like a blast of the same text to everybody. The win here is sorting people into a handful of buckets so the note actually fits the person getting it.
Ask the AI to split your quiet list by the kind of work you did. For most trades you'll end up with a few natural groups:
- Annual-service folks who've missed their yearly visit (heating, AC, chimney, alarm)
- One-time repair customers who might have a second thing by now
- Big-ticket installs from a few years back, getting close to a check-up or a warranty window
- Seasonal work, like gutters, irrigation, or a spring opening
Each bucket gets its own short message. The furnace-tune-up crowd hears something different from the one-time drain-clog crowd, and both can tell you wrote it for them.
Write a text that sounds like you remember them
The whole point is that you already earned this person's trust once. Your message should sound like a human who did their furnace, not a marketing department that bought their number. Short, warm, and easy to ignore without feeling nagged.
Try this shape. You name yourself and the work you did, give a reason you're reaching out now, and make the ask small:
Hi Mrs. Alvarez, it's Dave from Dave's Heating. We put your furnace in a couple winters back. Figured I'd check in before the cold sets in and see if you want me to swing by for a tune-up. No rush at all, just didn't want you stuck without heat in January.
Compare that to the version every one of us ignores: BOOK NOW, limited fall savings, call today. One sounds like a neighbor. The other sounds like spam.
Four quick rules keep these on the right side of the line:
- Use their name and the actual work you did, never a generic greeting
- Give a real reason for the timing, not a fake discount deadline
- Keep it under about four sentences, because they're reading on a phone too
- Make the no-cost exit obvious, so nobody feels cornered
- Have AI draft it, then read every one yourself before it goes out
Time it to the season
Timing is half the reason an old customer says yes. A tune-up offer in the week they were already starting to worry about it lands far better than the same offer in a random month.
New England hands you an easy calendar. Reach out to heating customers in early fall, before the first real cold snap sends everyone scrambling. Catch the cooling crowd in April and May, when the first warm day reminds them the system sat all winter. Gutters and chimneys fit the stretch before the leaves and the snow, and anything outdoors wants that short window after mud season.
Have the AI flag who's due based on the season and when you last saw them. You can tee up the fall batch in August and let it sit ready, so you're not writing texts at nine at night in October when the phone's already ringing.
Where AI gets this wrong
This tool is good at reading and sorting, and it's honestly not bad at drafting a friendly note. It's useless at knowing which of these things are true, so a few jobs stay yours.
It can't tell who moved, sold the house, or passed away. A year-old list will have a few of those, and the only fix is a human reading each name and a soft message that doesn't assume too much.
Left alone, it will happily invent a warm detail, like claiming you remember a dog you never met. Don't let it make up anything personal. If you don't actually remember the specifics, keep the note simple and true.
And it should never hit send on its own. Batch-texting on autopilot is how a nice gesture turns into a complaint, or worse, a run-in with the rules about texting people you haven't heard from in a while. You approve each one, or at least each small batch.
Worth doing this week
You don't need a new app or a system to start. You need one export and an hour.
- Export your customer list from whatever you use to bill or schedule, even if it's a plain spreadsheet
- Ask an AI tool to pull everyone you haven't served in 12 to 18 months and group them by what they last bought
- Pick the one group that fits this season, and have it draft a short, warm text you'd actually send
- Read them all, fix the names, cut anything that isn't true, and send ten by hand to see who answers
- Keep the list, because next season this becomes a thirty-minute habit instead of a project
If wrangling exports and getting the grouping right isn't how you want to spend that hour, that's the sort of setup we handle at Massachusetts AI Agency. Either way, the goldmine is already in your files. The only mistake is leaving it there.