Article

AI for job notes and customer updates

Short answer: have the AI turn the tech's notes into a job record, have a person check it, then draft the customer update only from that approved record.

After a service visit, the same facts get written twice: once in the job record for your files, and again in a message to the customer. An AI assistant can draft both from one set of notes, as long as a person reviews them and the bot uses only the facts it's given.

Why the record and the update belong together

The job record is for your team and the customer file, and it may matter later if there's a question about the work. The customer update is for someone who wasn't watching and may not know your trade's terms. Both come from the same source: what the tech did and noted on the visit. Write the facts once and let the bot make two versions.

Job recordCustomer update
ReaderYour team, laterThe customer, today
LanguageYour trade's terms are finePlain words, short
IncludesWork, parts, observations, recommendations, open itemsWhat was done, anything they should know, what happens next and who follows up
Leaves outGuesses and opinionsInternal notes, pricing talk the record doesn't approve, blame, anything uncertain stated as fact

A note format techs can say out loud

Most of the problem with AI job notes is the input. If techs use the same five lines every time — typed or dictated from the truck — the bot has far less to guess:

  1. Did: the work performed
  2. Used: parts or materials
  3. Saw: what they observed, in their own words
  4. Recommend: what they recommend, if anything
  5. Next: what the customer asked, what's still open, and who follows up by when

If a line is empty, the tech says "none." That tells the bot the gap is real, not forgotten.

The workflow, step by step

  1. The tech records facts as usual. Photos and job details stay in your normal tools. The tech adds the five lines above.
  2. The office pastes the notes into the record bot, with saved rules: facts only, mark gaps with MISSING, keep the tech's findings and recommendations exactly as meant, add nothing.
  3. A reviewer checks the record against the tech's notes and photos and saves it to the customer file.
  4. Only the approved record goes to the update bot. Routing the content this way does not remove the need to review the update.
  5. The reviewer checks the update and sends it from your normal email or text tool, on a channel the customer agreed to.

That's two bots, each with one job, and one person approving both.

Worked example 1: a routine lawn visit

Fictional example for illustration. The customer, notes and details are invented.

Approved job record (the only source for the update): "Weekly lawn visit, Thursday. Mowed and edged front and back. Side gate latch is broken (photo 2); crew closed the gate with a zip tie for now. Customer asked about fall leaf cleanup. Follow-up owner: office will send fall cleanup options by Monday."

Rules for the update bot:

  • Plain, friendly English, under 100 words. Customer first name only if the record gives one.
  • Use only facts in the approved record. Don't add reasons, dates, prices, advice or promises.
  • Mention anything the customer should know, as the record states it.
  • If the record assigns a follow-up, say who will follow up. If it doesn't, write "MISSING: follow-up owner" for the reviewer and stop.

Reviewed update: "Hi, your weekly lawn visit is done: front and back mowed and edged. Heads up: the side gate latch is broken, so the crew closed the gate with a zip tie for now. Thanks for asking about fall leaf cleanup — our office will send you options by Monday. — The [Company] team"

Every sentence comes from the approved record. The bot didn't quote a price for the cleanup, suggest how to fix the latch, or promise a date the record didn't give.

Worked example 2: a part on order, no date yet

Fictional example for illustration.

Tech notes: "Did: diag on no-cool call. Used: none. Saw: failed part in outdoor unit, see photos 1–3. Recommend: replace part, quoted on site, cust approved. Next: part ordered, supplier hasn't confirmed delivery. Office calls cust w/ date."

Approved record (after review): "No-cooling service call. Tech found a failed part in the outdoor unit (photos 1–3) and recommended replacement. Customer approved the quoted repair on site. Part ordered; supplier delivery date not yet confirmed. Follow-up owner: office will call the customer with a date once the supplier confirms."

Reviewed update: "Thanks for having us out today. Our tech found a failed part in your outdoor unit, and you approved the repair. We've ordered the part. We don't have a delivery date from the supplier yet, so our office will call you as soon as it's confirmed. — The [Company] team"

What made this one tricky: the honest answer was "we don't know yet." The rule "never fill in a date the record doesn't give" kept the bot from writing "in a few days." The update also doesn't name the part or explain the fault; the tech already went over that with the customer in person, and the reviewer decided the message didn't need it.

When a written update isn't the right tool

Some visits need a person on the phone first. Call, then send a short written summary of what you said, when:

  • The tech noticed something that could affect safety
  • The work wasn't finished, or the price changed from the estimate
  • The customer is unhappy, or there's a warranty or billing question
  • The news is bad, or the next step is a big decision for the customer

The AI can still help draft the follow-up summary after the call — from the reviewer's notes of what was actually said.

Review checklist

Job record:

  • Work, parts and observations match the tech's notes and photos
  • Recommendations are the tech's own, not added by the AI
  • Gaps are marked MISSING; no speculation about cause or fault

Customer update:

  • Every sentence is supported by the approved record
  • No prices, dates, diagnoses, repair or safety advice the record doesn't contain
  • Uncertain things are described as uncertain ("not confirmed yet")
  • Open questions go to the person the record names
  • Sent by a person from your normal channel, to a customer who agreed to that channel

Records, photos and privacy

  • Your field-service or job-management tool stays the record. The AI chat is where the draft gets written, not where you keep the file.
  • Paste text, not every photo. If a photo matters, describe it in the notes ("photo 2: broken latch").
  • Leave out card details, gate, lockbox and alarm codes, and personal details the update doesn't need. Check the data and history settings of the AI tool you use.

Boundaries — and the path beyond them

  • The AI doesn't diagnose, recommend repairs, price work or commit to dates. Your qualified techs and your records are the source of job facts.
  • Pricing, billing and warranty questions go to the right person on your team, not the bot.
  • Safety questions go to a qualified person.
  • In this starter workflow, nothing is sent automatically. Once the update workflow is repeatable, it can become a defined, explicitly authorized automation — known customer, a channel they agreed to, approved content only, with logs, stop rules and an off switch — where the connected service supports it. Otherwise keep it in review.

Quick answers

Q: Can AI send customer updates automatically?
A: Not in this starter workflow. Later, a defined update can be automated with explicit permission and safeguards where the tools support it.

Q: What should the bot do when notes are incomplete?
A: Mark the gap with "MISSING:" and leave it blank. Never guess parts, times, causes or who will follow up.

Q: Can techs dictate notes instead of typing?
A: Yes, if your phone or tool turns speech into text. Use the five-line format and check names, part numbers and other figures in the transcript before the bot uses it.

Q: Is it okay to paste customer details into an AI chat?
A: Use business accounts you control, share the least you need, and leave out card details, gate, lockbox and alarm codes, and personal details.

Make it better each week

When the reviewer fixes a draft, add the fix as a rule. Over time the drafts should need fewer changes.