---
title: "Automatic service reports with AI · Quentin Andrieux"
description: "The technician dictates or takes a photo, the standard service report goes to the customer and the office. Demo, client case and method for maintenance and HVAC companies."
url: https://quentinandrieux.com/en/maintenance/service-report/
lang: en
author: Quentin Andrieux
updated: 2026-11-10
---

> Version texte de https://quentinandrieux.com/en/maintenance/service-report/ (the HTML page is the reference). Author : Quentin Andrieux, https://quentinandrieux.com/

[Maintenance](https://quentinandrieux.com/en/maintenance/) · Service report

Maintenance · HVAC · heating

# The service report, complete before leaving the site.

The technician dictates a few sentences and photographs the nameplate. The standard report goes to the customer and the office, with no re-typing.

[Try the demo](https://quentinandrieux.com/en/maintenance/service-report/#demo)[Get a written plan](https://quentinandrieux.com/en/maintenance/service-report/#plan)

Quentin

By the way, I've already built this for an HVAC company with around twenty technicians.

**÷2** time spent per job *as estimated by the team*

**~20** technicians equipped

**700** jobs a month

**287** quotes waiting at the start

A mobile app that reads photos, takes dictation and works offline, and an office tool that helps with diagnosis. Delivered in a two-week sprint.

What goes wrong today

## The report is where information gets lost.

- Photos pile up unsorted, then get sorted at the office.
- Brands and serial numbers are copied by hand, with mistakes.
- When quoting, nobody knows which part to replace: a technician has to go back.
- Time on site is not always recorded.
- The office re-types what the technician already wrote.
- Work still to do (a quote, a return visit) is written down nowhere, except in the technician's head.

The demo · example, fictitious data

## A technician's note, a report ready to send.

Pick a job. On the left, what the technician sends from their phone. On the right, the report the tool produces, with what a person needs to check.

Demonstration with fictitious data. At your company, the tool follows your report template and writes into your database.

How it works

## Four steps, no re-typing.

01

### Dictate and photograph

On the phone, even without signal. The technician describes the job out loud, in their own words.

02

### Read the nameplate

AI reads brand, model and serial number, and finds the equipment in your database.

03

### Write the report

In your company's format: reason, work done, parts, time, work to plan.

04

### Send and move on

The report goes to the customer and the office. If a part needs replacing, the quote is prepared straight away.

What it costs you

## Your reports have a cost. Work it out.

Technicians, jobs per week, minutes of writing and re-typing: the calculator gives you an order of magnitude in hours and euros per year.

Example · 20 technicians ≈ €159,000 a year [Slide to work out yours](https://quentinandrieux.com/en/maintenance/service-report-calculator/)

Free · no commitment

## A written plan, for your company.

Tell me which documents slow you down. I'll send back the two or three worth automating, and the shortest path to a tool that runs.

Thank you, received. I'll reply within two business days.

Rather talk it through? [Book 30 minutes](https://cal.com/quentin-andrieux/30min)

Frequently asked

## What people ask me before we start.

### Will my technicians have to change their habits?

No. They dictate and take photos instead of writing: it is quicker than before. The day on site is used to fit the tool to the way they work.

### Does the report follow our template?

Yes. We start from your current reports: sections, mandatory mentions, layout. The tool fills in your template, not its own.

### What about no signal, in a basement?

The app can work offline and sync when the signal is back. That is what we built for an HVAC company.

### What if the photo is blurry or the nameplate unreadable?

The tool does not guess: it flags what it could not read, and a person validates before the report goes out.