AI Agents in a 2-Person Firm: SAM Cut Admin from 6 to 2.5 Hours
In short
A documented two-person French retirement advisory, SAM, used three AI agents to cut administrative time per file from 6 to 2.5 hours (−58%), lift conversion from 28% to 41%, and reduce per-file costs by 30–35%. Running costs: €650/month; setup €3,000 and 60 hours. The case shows AI agents work in micro-business when the process is standardized before the technology.
A Micro-Business as Blueprint
SAM (Services et Assistance pour la Retraite) is a two-person French retirement advisory founded in 2020. The case was published by the French government program France Num in May 2026; the firm was also a finalist in the CPME SME association's 2025 AI award. The founder built the system with his then-15-year-old daughter using off-the-shelf no-code tools.
Three AI Agents, One Clear Process Model
The decisive sequence: the founder standardized each file process before automating anything. Every file follows the same step sequence. Three clearly separated agents sit on top of that standard process.
- Collection agent: requests documents, sends reminders via email/SMS, checks incoming documents, and flags gaps.
- Writing agent: drafts letters to pension funds, summaries, and client statements.
- Tracking agent: maintains the log, delivers status updates, and flags stuck files.
What the Numbers Show After Six Months
The documented results are notable for a micro-business: less administration, faster document collection, higher conversion, and lower unit costs.
- Administrative time per file
- 6 h → 2.5 h (−58%)
- Time to collect main documents
- 6 weeks → 3.5 weeks
- Incomplete files after 15 days
- 30% → 8%
- Conversion rate (lead → client)
- 28% → 41%
- Customer satisfaction
- 4.2 → 4.7 out of 5
- Cost per file
- −30% to −35%
- Automated reminders
- 70%
58%
less administrative time per file after six months
Costs and Setup: Realistic, Not Mythical
Running costs are about €650 per month for AI licenses. On top came roughly €3,000 for server setup and about 60 hours of configuration work. That is an order of magnitude a two-person firm can bear—if you honestly account for your own time.
Not a pure no-code promise
This case does not show that you can simply buy software without process thinking. Important letters still get human approval, and the AI has clear boundaries. That is exactly what makes it scalable.
Six Practices That Made the Difference
- Standardize the process first: every file follows the same step sequence before automation.
- Human-in-the-loop: important letters still get human approval.
- Define clear limits: what the AI may do autonomously and what it may not.
- Measure the baseline: record administrative time, costs, and conversion rate before the change.
- Data hygiene: central instead of scattered data, one file per client.
- Data protection from day one: encryption, access controls, consent evidence, audit trail.
For Swiss leaders, the cultural shift matters as much as the technology. The Swiss AI Podcast discusses why AI in SMEs is a management topic, not an IT project.
Transfer to DACH: Data Protection and System Integration
In Switzerland, revDSG applies as soon as personal data is processed—and GDPR may apply for cross-border processing. In the SAM case, encryption, access controls, consent evidence, and an audit trail were built in from day one. For a Swiss SME, that means checking whether processing happens in Switzerland or the EU and whether the agents can connect to existing systems such as Abacus, Bexio, or Microsoft 365.
Your First Step
Do not start with a big AI project. Start with a single, well-bounded administrative process. Measure the baseline, standardize the step sequence, define clear limits, and keep a human in the approval loop. The full case study is documented at ai-first.company for reference.
Frequently asked questions
- Can AI agents really work in a two-person business?
- Yes. The SAM case shows a two-person French retirement advisory cut administrative time per file from 6 to 2.5 hours, with about €650 monthly license costs and 60 hours of setup work.
- What does a comparable AI agent setup cost?
- In the documented case, about €3,000 for server setup, roughly €650 per month for licenses, and around 60 hours of configuration. For Swiss conditions, add potential adjustments for revDSG and local system integration.
- Which processes are best to automate first?
- Standardizable, repetitive administrative processes with clear inputs and approval points—for example document collection, reminders, and status tracking. Highly individualized advisory conversations are not a good starting point.
- Do you need an in-house IT team?
- No. In the SAM case, the founder built the system with his then-15-year-old daughter using no-code tools. Process clarity and realistic boundaries matter more than programming skills.
- How do you measure ROI?
- Establish a baseline first (time per file, costs, error rate, conversion rate), then compare the same metrics after a suitable evaluation period. SAM measured administrative time, file completeness, and conversion rate, among others.
- Do AI agents in Switzerland require revDSG compliance?
- Yes. As soon as personal data is processed, revDSG and possibly GDPR apply. Encryption, access controls, consent evidence, and audit trails should be planned from the start.
Sources
Chris Jon Graf · AI Strategist & CEO
For over 15 years, Chris Jon Graf has built productive Swiss software platforms and advised executive teams on the strategic deployment of artificial intelligence. Would you like to evaluate this topic for your company?