A plumber is under a kitchen sink replacing a supply line. To shut the water off he walks past the electrical panel in the garage and notices a breaker with a brown scorch mark around it. He mentions it to the homeowner on the way out: "you might want to get someone to look at that." The homeowner nods, thanks him, and forgets. Three months later an electrician from a different company replaces the panel. The plumber's company has an electrical division. Nobody in the office ever heard about the breaker.
That is the leak this post is about. It is not a missed call, an unsold quote or a lapsed maintenance plan. It is a job that a qualified person in your own uniform already found, standing in the customer's house, and then lost somewhere between the driveway and the office. As Gideon Wafula, AI Automation Engineer, I have started seeing this come up in almost every conversation with multi-trade operators this year, and it is one of the cleanest automations I know of to build.
The home services market is consolidating into multi-trade shops. Industry outlooks and M&A reports this year keep describing the same formation pattern: HVAC plus plumbing plus electrical under one brand, often through acquisition, with the investment thesis resting heavily on selling more trades into the same customer base. The CRM vendors have noticed too; almost every home-services software roundup published this year leads with some form of "sell more to the customers you already have."
The trouble is that acquiring a second trade does not create cross-selling. It creates the possibility of it. The actual cross-sell still depends on a technician in one trade noticing a problem in another, saying something, and that observation surviving the trip back to someone who can book it. In most shops I talk to, that chain runs on memory, a free-text notes field nobody reads, and the occasional technician who happens to text the office manager. When it works, it works because of one person. When that person leaves, it stops.
Vendor blogs publish a lot of confident conversion percentages for post-job upsells. I would treat those numbers as marketing rather than benchmarks. What the operator commentary agrees on, and what matches what I see, is simpler: the moment right after a job is finished is when the customer trusts you most, and an offer tied to something a technician physically saw in their home lands very differently from a generic seasonal email.
This is the biggest leak by far. Technicians are paid and measured on their own job. Writing up something outside their trade is unpaid, takes time at the end of a long day, and requires typing into a phone app with dirty hands. So they say it out loud to the customer and move on. The fix is not a motivational speech. It is making the capture take under a minute: a single button, a photo, and a voice note. No typing, no category dropdown with forty options.
Some shops do get technicians to write things down, usually in the job notes. Then the job closes, the invoice goes out, and the note is archived with it. There is no queue, no owner, and no deadline. A note that says "saw corroded anode on WH, prob 12yrs old" in an HVAC ticket is invisible to the plumbing dispatcher. This is a routing problem, not a people problem.
"You should get that looked at" is not a follow-up. Customers are distracted when a technician is leaving; they are thinking about the invoice and the dog. What they need is a short written message the same day, with the photo, a plain explanation of why it matters, and a way to book. Then one reminder, spaced properly. Then silence.
If you do not count observations, you cannot tell which technicians are finding things, which trades generate the most handoffs, or how many turn into booked work. Without that, the whole thing lives or dies on the owner's enthusiasm in a Monday meeting. With it, you can see in a month whether the program is real.
Before I quote this for a client, I ask for an afternoon of exports:
I build this in n8n sitting between the field service platform (ServiceTitan, Housecall Pro, Jobber, or whatever the shop runs), SMS and email, and a simple opportunity table. The governing rule: the automation may structure, route and remind, but it never diagnoses. Only a technician's observation, and later a qualified visit, defines what is wrong.
A mobile form or a dedicated SMS number where the technician sends a photo and a short voice note. The workflow attaches it to the current job automatically using the technician's schedule, so they never type the address or the customer name. If it takes more than a minute, adoption dies within a fortnight.
A model transcribes the voice note and maps it onto a short fixed list of observation categories the business owns: electrical panel concern, water heater age or condition, visible leak, duct or airflow issue, missing safety device, and so on. It extracts the trade, the location in the home, and the technician's own words. It does not add severity language the technician did not use. If the note is ambiguous, the category is "needs review", which is a correct output.
The structured opportunity lands in the queue of the department that can act on it, with the photo, the original voice note and the customer history attached. Each item gets an owner and a follow-up deadline. Anything untouched after that deadline escalates to the service manager. Single-trade businesses route to a vetted partner instead, under a written referral agreement.
Within a few hours of the job closing, the customer gets a short, plain message: what the technician noticed, the photo, why it is worth checking, and a booking link or a reply-to-book option. The model writes the phrasing from the structured fields; it never invents a risk level, a price or a deadline. Safety-critical categories, like signs of overheating at a panel or a gas smell, use a fixed, reviewed script that is identical every time.
If there is no reply, one reminder goes out about a week later, and then the sequence ends. The opportunity stays open in the queue for a human to decide on. Endless nudging is how you turn a helpful observation into a reason to block your number, and it also runs straight into the consent rules I covered in the outbound compliance post.
Every opportunity writes back to the customer record with its outcome: booked, declined, not reached, or not needed after inspection. A weekly report shows observations per technician, handoffs per trade, and conversion to booked work. Many shops choose to pay a small spiff on booked handoffs, and the report is what makes that fair and automatic.
This sits naturally next to two builds I have written about before. The aging equipment replacement pipeline works from install dates in the database; the cross-trade handoff works from eyes on site, and the two feed each other. The neighbor offer expands a visit outward to the street; this one expands it inward to the rest of the house. If you run all three, you are selling almost entirely to people who already let you through the door, which is the cheapest revenue a home services business has.
Running costs typically land around 30 to 120 USD per month, scaling with the number of jobs and messages, not the size of the customer database. The one-time build is mostly spent on the unglamorous parts: agreeing the observation categories, writing the safety scripts, and connecting to the field service platform. See my automation services page for how I scope that.
Baseline four numbers before launch: observations captured per 100 jobs (this moves first, usually within two weeks), share of observations with an owner action inside the deadline, handoff-to-booked conversion, and revenue from booked handoffs. The honest caveat: this makes a real observation travel reliably from the driveway to a booking. It does not make technicians look. A crew that never glances outside its own trade will produce an empty queue, and the report will show you that very clearly. That is still useful information. It is just a training problem, not an automation one.
Gideon Wafula builds custom AI automation systems, n8n, WhatsApp, Voice AI, and more.
See Services →