At a glance
- Company
- SAMRetirement advisory firm for individuals and business owners
- Size
- Micro and small (1–49 employees) The official case page describes the firm as having two people.
- Industry
- Professional services
- Where and when
- France. Six months of use, published May 2026.
- Who produced the figures
- The company itselfThe figures are the firm’s own, published by France Num, the French government’s programme for the digital transformation of small firms. The firm was a finalist for the 2025 AI prize of the small-business federation CPME.
What they did
SAM advises individuals and business owners on retirement: pension rights, the timing of retirement, buying back contribution periods. The founder set it up in 2020. He describes the work before the change as run “by emails and Excel files”, with a risk of forgetting things, constant chasing and documents arriving in every format.
The idea came from his daughter, then 15, who watched him work during a school holiday and offered to help him simplify it. Together they built a system from ready-made parts:
- an automation tool that needs no programming, running on the firm’s own server;
- a library of the firm’s reference documents, forms, letter templates and common questions, which the AI draws on when it writes;
- several general AI models.
The work is split between AI “agents”, each with one job:
- Collecting. It asks the client for documents, sends reminders by email or text, checks what arrives and reports what is missing.
- Writing. It drafts letters to pension funds, summaries of the client’s situation and explanations for the client.
- Tracking. It keeps a log, updates the status of each file and raises anything that is stuck.
When a file is opened, the client fills in an online questionnaire. From then on each file follows a set process, with steps, expected documents and target dates.
The numbers
After six months of use:
- Administrative time per file: from about 6 hours to 2.5 hours, a 58% cut.
- Time to collect the key documents: from 6 weeks to 3.5 weeks.
- Files still incomplete after 15 days: from 30% to 8%.
- Prospects who became clients: from 28% to 41%.
- Client satisfaction: from 4.2 to 4.7 out of 5.
- Cost per file down by an estimated 30–35%.
- About 70% of reminders are sent by the AI without anyone stepping in.
What it cost:
- About €3,000 to set up the server.
- Nearly €650 a month in AI licences.
- About 60 hours of initial set-up work.
What the case shows
- A two-person firm did this with ready-made tools, including an automation tool that needs no programming.
- The process was redesigned first. Every file became the same sequence of steps. That is what made it possible to hand steps to AI.
- The gain was in the paperwork around the advice. Chasing documents and drafting standard letters went to AI. The advice stayed with the adviser.
- The running cost is real. €650 a month is a large sum for two people. The published account says the gains more than cover it.
- Sensitive data needed its own rules from the start.
What this says about the practices
Practices the company applied
- Redesign the workflow instead of adding a tool to it Each client file was turned into a process: steps, documents expected, bodies to contact and target dates. The AI acts at each step and flags when a person is needed. Do next · Effort: high (months, several teams)
- Keep a person checking anything that goes to a customer or carries legal weight The adviser approves important letters, and can correct how the AI has classified a document or change a reminder. Do first · Effort: medium (weeks, a few people)
- Know the rules that already apply to your AI use The files hold career records, health data and income. The firm set up encryption, access controls, access logs, consent records and an audit trail to meet data-protection law. Do first · Effort: low (days, one person)
- Measure time saved per task against a baseline The firm reports each result against the figure before: hours per file, weeks to collect documents, share of incomplete files. Do first · Effort: low (days, one person)
- Put your own records in order before you ask AI to predict from them The firm names centralised data, one file per client in place of scattered files, as one of the choices that decided the project’s success. Do next · Effort: medium (weeks, a few people)
- Set the limits before the tool acts The AI sends reminders unaided and raises an alert when a person is needed. The adviser approves important letters. Do next · Effort: low (days, one person)
Sources
- France Num (26 May 2026). Dans les Alpes-de-Haute-Provence, une entreprise de conseil retraite déploie des agents IA pour automatiser le traitement de ses dossiers. Case page, in French. francenum.gouv.fr