How I work

Every engagement starts from the same question: what is eating your time — or about to cost you a hire?

I don't start by asking where to put AI. I start with the business problem and define what better looks like, then choose the right mix of process changes, custom software, automation, and AI. The technology follows the problem.

A fully loaded admin or coordination position in Switzerland runs CHF 100'000+ a year, before you've found the right person, trained them, and hoped they stay. The ladder below is built against that number: a fixed-price diagnostic that finds the automation, sprints that ship it, and support that keeps it running.

1 — The Diagnostic

Two weeks · CHF 3'900, fixed

We spend half a day together at your site. I map how work actually flows through your business — which documents exist, who touches them, where the hours go. Quotes, delivery notes, incoming mail, product data, the CRM nobody updates: the real processes, not an org chart.

Then I build. One quick win ships during the diagnostic itself — a working automation in your business, not a recommendation to build one.

You end the two weeks with:

one automation already running,
a map of your workflows and documents,
a prioritized roadmap: each candidate automation with its effort and its impact, in plain language.

If we stop there, you still have a working improvement and a plan you could execute without me.

Book the first call

2 — Implementation sprints

Fixed scope · on quote

One workflow at a time, picked from the roadmap. I define the scope with you before we start — what gets automated, what "done" looks like — and then I build it, personally.

Training isn't a separate line item; it's how delivery ends. Your team uses the new workflow, with me in the room, before I move on. A sprint is finished when the workflow runs in your business — not when a document is handed over.

3 — Ongoing support

Monthly · on quote

Automations are not fire-and-forget. Tools update, file formats drift, the business changes, and the owner who was promised "set and forget" ends up debugging alone.

That's what this is for: I keep what we built running, keep improving it, and stay one message away when something new comes up. Priority access, monthly, cancel when it stops earning its keep.

In every engagement

I deliver personally. The person you meet is the person who builds. For heavier software architecture I bring in senior collaborators from my network — and I tell you who is doing what.
Adoption, not just installation. I don't leave until your team actually uses what was built.
Your existing tools first. The goal is not fifty new licenses. I build on what you already pay for wherever that's honest — and tell you when it isn't.
Named tools, no mystery. I use custom development where that is simpler, and established tools where they already do the job — Claude, Google Workspace, Infomaniak, Fireflies, Slack, ClickUp, HubSpot, and Atlassian's Jira and Confluence. You'll always know what runs your automation and what it costs.

What I don't do

Reports without implementation. If you want a strategy deck for the drawer, I'm the wrong person. I only take engagements where something gets built.
Regulated-data sectors. Medtech, biotech, pharma, finance — patient data and heavy compliance deserve specialists. I'll say so in the first call and point you elsewhere.

Where your data goes

It's the first question Swiss owners ask about AI — often the only one that really matters to them: what happens to our data? Here is the plain answer, the same one you'd get in a meeting.

The mechanics, without mystery. When an automation processes one of your documents — a quote, a letter, a product sheet — the relevant text is sent to an AI provider's service, processed, and returned; the result is stored in your systems. I use business/API plans and settings under which your data is not used to train their models. That's a contractual and configuration point, not a hope.

Only what the task needs. An automation that files your mail has no business seeing your accounting. Every connection is scoped: which system, which data, read or write — decided explicitly, not defaulted. The narrower the access, the smaller the question.

Honest about who the providers are. I build with the tools that fit and I name them: Claude (Anthropic), ChatGPT (OpenAI), the Google stack — US companies that often provide the strongest models, and I won't pretend otherwise. The safeguards are real rather than rhetorical: data-processing agreements, minimization, and your sign-off before anything runs. Where a Swiss or European option does the job — Infomaniak for hosting and services, Mistral for AI, or self-hosted tooling — I use it. No sovereignty theater in either direction: I won't wave a Swiss flag over a US API, and I won't choose a worse tool just to wave one.

nLPD-aware by design. Before any automation runs, you get a one-page data map: which data, from which system, goes to which provider, for what purpose. If a workflow touches personal data — employees, customers — it's designed so your obligations under Swiss data-protection law are met. And if something shouldn't leave Switzerland, it doesn't.

Where I stand myself. This site and my email are hosted at Infomaniak, in Switzerland. The company has been in Lausanne since 2016. Not a marketing point — just where things actually are.

The same discipline applies to this website itself — the privacy policy is short enough to actually read.

FAQ

We tried ChatGPT and it didn't stick. Why would this be different?

Because a chat window isn't an automation. Most stalled AI attempts I see put the burden on people to remember to use a tool. I build workflows that run inside the processes you already have — the mail arrives, the document is processed, the CRM updates — whether anyone remembered or not. A stalled attempt is a good starting point, not a bad sign.

Which tools do you use?

The ones that fit, and I name them. I often build a small custom module, then connect it to what you already run — Claude, Google Workspace, Infomaniak, Fireflies, Slack, ClickUp, HubSpot, Atlassian's Jira and Confluence, or your ERP. I have more experience with some providers than others and I'll tell you which.

What happens to our data?

Short version: nothing leaves your business without you knowing where it goes and why. The long version is written in plain language — see "Where your data goes" above.

Do you do training?

Yes — inside every engagement, never instead of one. Your team is trained on each automation as part of its delivery. I don't sell courses.

Start with the Diagnostic

Thirty minutes on the phone first — no charge, no deck. Bring the process that eats your week; I'll tell you honestly whether the Diagnostic is worth your money.

Book a call