🧪 Experiment

Real tests, verifiable results.

The Lab gathers my experiments on AI agents, automation and business tools. Distinct from the blog: these are tests, not pedagogical content.

AI agents·May 2026

Goal: Compare reliability of automatic pre-screening on 50 CVs with a given criteria grid.
Tools: ClaudeChatGPTn8n
Result: Claude is more stable on long CVs and atypical profiles. ChatGPT is faster but more opinionated.
Recommendation: Keep a human in the loop, especially at final validation. AI doesn't decide — it suggests an order.

n8n·Apr 2026

Goal: Reduce the average delay between an inbound lead and the first personalised follow-up.
Tools: n8nHubSpotClaude
Result: Average delay went from 38h to under 2h, with stable open rates.
Recommendation: Always add a 24h human review step to avoid the robotic feel.

ChatGPT·Mar 2026

Goal: Give new hires an assistant that answers the most common HR questions.
Tools: ChatGPT (GPTs)Notion
Result: 70% of routine questions (leave, benefits, tools) absorbed by the assistant in week one.
Recommendation: Connect the GPT to a living docs base (Notion / Drive) and refresh it monthly.

Make·Feb 2026

Goal: Measure the real cost of an automation stack over 6 months for a 25-person SMB.
Tools: MakeZapierGoogle Sheets
Result: Make is 3-4× cheaper at equivalent volume, with better scenario readability.
Recommendation: Start on Make unless the team has been heavy Zapier users for years already.

Next step

Turn your tools into performance levers.

A 30-minute call to identify automation, optimisation and AI opportunities tailored to your business.

What you walk away with