AI for SMEs with technicians: what is possible today, and what should you watch out for?

At every trade fair and in every newsletter, the talk is of AI and ChatGPT. Meanwhile a technician retypes their reports in the evening, and a French-speaking customer waits for a version they can read. You want to know where AI helps with that today, and what you had better not do.

Published by Abdelhay Rahou (opens in a new window)5 November 20269 min read

AI for SMEs with technicians: where it helps today

The AI everyone is talking about, such as ChatGPT, is a language model: software that reads text and suggests a text. So AI for SMEs with technicians fits best with the writing that comes back every day:

Turning notes into a report. “Boiler cleaned, pressure topped up, radiator valve leaking” becomes a report the customer understands.
Summarising. A long email exchange or the history of an installation in a few sentences, before you call the customer.
Translating. A report or an email for a French-speaking or English-speaking customer.
Preparing an email. Explaining a finding, proposing a second visit, answering a customer’s question.
A first version of a work instruction or checklist, based on your own notes.

A language model writes fluently, including about what it does not know. It can add a measured value, a replaced part or a promise that appeared nowhere in the notes, and quote a price or a standard that is wrong. So AI helps with the form. The content comes from the technician who did the work.

Where fixed rules work better than AI

Much of what gets left undone in a service company is not writing. A reminder to the customer the day before the visit, a quote that needs following up, a draft invoice after a finished intervention, a notification to the warehouse: a fixed rule is enough for these. When this happens and that is true, do that.

A rule does the same thing every time, and afterwards you can see why it did or skipped something. A language model writes a different text for every request. For a proposal that someone reads through, that is not a problem; a step that must run the same way every time is something you set down in a rule.

Task	What fits	Why
Turning the technician’s notes into a report	AI, with a read-through	Writing; only the technician knows the facts
Translating a report for a French-speaking customer	AI, with a read-through	Language work; check the technical terms
Reminding the customer of tomorrow’s visit	A fixed rule	The same step at a fixed moment
A draft invoice after a finished intervention	A fixed rule	The amounts come from the hours and parts
Following up a quote that has been open for a week	A fixed rule, then a person	The rule creates the task, the salesperson calls
Granting a discount or a credit note	A person	A decision about money
Deciding whether an appliance is safe	The technician	A judgement on site

Which follow-up steps to set up as a rule first is explained in automating processes.

What do you not paste into a public chatbot?

Whatever you paste into a chatbot goes to the provider of that service. What they do with it is in their terms: read what they say about retention and about using your data to train models, and let your staff work with an account whose terms you know.

Do not paste in:

names, addresses, phone numbers and email addresses of customers;
access codes: the code of an alarm or a gate, where a key is kept;
prices, margins and quotes, and what you agreed with a customer;
anything about the occupants: who needs care, when someone is not at home.

To have a text rewritten, “customer A, flat, heat pump” is usually enough. Remove names and addresses before you paste.

Customers’ names, phone numbers and email addresses are personal data, and the GDPR requires you to process only the personal data needed for your purpose. That is how the European Data Protection Board explains it in its guide for SMEs (in French). Whether, and under what conditions, you may use an AI service with personal data is something to check with your data protection officer (DPO), if you have one, or with your legal adviser.

A person reads it through, and a person sends it

A proposal from AI is a first version. Before a report or an email goes to the customer, someone who knows the work reads it through:

Is every fact right compared with what the technician did and found: the work, the parts, the measurements?
Does the text promise anything you did not agree: a deadline, a price, a warranty?
Is there anything in it that should stay internal, such as a remark about the customer?
Have the technical terms been translated correctly?

Whoever reads it through also sends it. That way, for each document, there is one person who knows what the customer received.

Starting with AI in your SME: one use, six steps

Choose one writing task that comes back often. For example the reports of maintenance visits, or translating reports for French-speaking customers.
Set down what must not go in. The list above, and how you remove names and addresses.
Write a fixed instruction. How you address the customer, the fixed parts of a report (problem found, work carried out, solution, advice) and what the text must never add.
Test it on a few old reports. Put the proposal next to what the technician wrote at the time. What did you have to correct?
Agree who reads through and who sends, for each type of document.
Look back after a month. What was corrected every time, and does it work for your team? Adjust the instruction, or stop.

An example: reports after a day of maintenance

A technician from a heating company services four gas boilers in a day. For the last one, he notes: “boiler cleaned, burner checked, flue gas measurement ok, pressure topped up, bedroom radiator valve leaking, advice replace”. At the office, the planner has that note rewritten into a report with the problem found, the work carried out, the solution and the advice. She does not paste in the name and address. The customer is French-speaking, so she also asks for a French version.

The proposal says that the valve “was replaced”. The technician only gave advice. The planner calls him briefly, corrects the sentence in both languages and sends the report herself. The reminder for next year’s maintenance runs through a fixed rule, which sends the customer an email the day before the visit. No AI is involved there.

How it works in Digitalnatie Service Suite

The AI report assistant is an optional, paid module that is quoted separately: the pricing page lists it under “Optional, quoted separately”. Digitalnatie sets it up for you.

Where: on the intervention record, in the report tab. The technician fills in the report itself in the technician app: the work carried out, the problem found and the solution.
What it does: rewriting, summarising, translating and preparing an email to the customer.
What it does not do: it never sends anything automatically and never writes in the report. A person always reviews and confirms the result before it is used.
Setting it up: the assistant needs a configuration that cannot be done in the application. As long as it is missing, the assistant says that AI has not been set up yet. The assistant’s provider is set in a settings tab that only the system administrator sees. Ask Digitalnatie.
Sending stays a separate step: the report only goes out when someone sends it, or marks it as delivered.

The automations are rules. In the work menu, the administrator builds them with blocks for what triggers the rule, the conditions, a wait and what the rule does, for example an appointment reminder or a draft invoice after a finished intervention. A test shows on a real record which path the rule follows, without sending or creating anything, and the history lists every run with its status. The video, in Dutch, shows such a rule: a confirmed quote becomes an intervention to schedule, and the planner gets a notification. No AI is involved.

Short video · 0:30

Elke keer dezelfde handeling, waarom nog manueel?

Is een offerte bevestigd, dan staat de interventie klaar om in te plannen en krijgt de planner een melding. Eén keer ingesteld, daarna gaat het vanzelf.

Video in Dutch, with Dutch captions

All short videos (in Dutch)

Good to know:

Among the integrations there are cards for Claude, ChatGPT, Gemini and a custom MCP server. You can connect and test them, but Service Suite does not yet send any AI requests through that connection.
Scheduling, invoicing and sending messages are done by your staff or by the rules you set up. A draft invoice from an automation is never posted or sent.
Whether your data trains AI: according to the frequently asked questions, Digitalnatie does not use customer content for its own purposes and never sells it. Hosting, sub-processors and the data processing agreement are covered in data in the cloud.

What a good report for the customer contains, with or without AI, is explained in the intervention report to the customer.

Frequently asked questions

What can AI do today for an SME with technicians?

Mainly help with text: turning notes into a report, summarising, translating and preparing emails. What the technician did and found must come from their own notes, and someone reads the proposal through before it goes to the customer.

Is ChatGPT safe for businesses?

That depends on what you paste into it and on the terms of the account you use. Read what the provider does with your data, do not paste in customers’ names, addresses, codes or prices, and ask your DPO or legal adviser whether you may use an AI service with personal data.

Do I need AI to automate reminders and follow-ups?

No. An appointment reminder, a follow-up or a draft invoice follows a fixed rule: when this happens and that is true, do that. Such a rule does the same thing every time, and the history shows why it did or skipped something.

Is the AI report assistant included in the Service Suite subscription?

No. It is an optional module that is quoted separately and that Digitalnatie sets up for you. Without that configuration, the assistant says that AI has not been set up yet.

Does the AI report assistant send reports or emails itself?

No. It helps with rewriting, summarising, translating and preparing an email to the customer, but it never sends anything automatically and never writes in the report. A person reviews the result, and the report only goes out when someone sends it.