AI automation: your team gets back the hours the repetitive work eats, measured.

We find the tasks worth automating, build the agents with n8n or custom code, run them on EU infrastructure, and measure the time recovered against a baseline recorded before we start.

Tell us about your projectA 20-minute video call Estimate the hours yourself

Baseline recorded before the build
EU hosting by default
A person approves what commits you

A person at a desk reviewing work on a laptop and a large monitor — Photo for illustration.

What does an AI automation agency do?

It maps the processes that consume time, picks the ones genuinely worth automating, builds them, puts them into production, then measures what came back. The discipline has an older name: business process automation.

Depending on who you ask it is also called workflow automation. What generative models changed is the range: tasks that required reading a document and understanding it became automatable. Everything else stayed where it was: someone still checks the output, exceptions still need a route, and the process still changes twice a year.

The tools follow the job. n8n, self-hosted in the EU, when workflows touch personal data. Make when your team wants a hosted tool it can edit itself and the data allows it. Custom code when the logic outgrows both.

Where your data goes: three hosting levels, chosen per task before anything is built.

An automation that reads customer emails sends personal data somewhere. Which model sees it, in which country, under which contract, is written down at scoping, so your data protection officer reads a fact rather than a promise.

Three hosting levels for AI automation: the data each suits, where it runs, and what you give up
Three hosting levels for AI automation: the data each suits, where it runs, and what you give up	Suits	Where it runs	What you give up
Public model API	Content with no personal or confidential data	The provider's cloud, under a contract that excludes training on your data	Control over location; nothing sensitive may go there
Model hosted in the EU	Personal data and ordinary business confidentiality	An EU region, with a data processing agreement and the region pinned	Some of the newest models arrive later in EU regions
Open-weight model on your infrastructure	Genuinely sensitive material: health, legal, trade secrets	Your servers or your EU cloud account	You run it, or pay someone to: hardware, updates, monitoring

Whatever the level, a log of who triggered what with which model, and a notice telling people when they deal with an AI system: a duty under Article 50 of the AI Act since August 2, 2026, checked on EUR-Lex on September 28, 2026.

Anyone can demo. Show me production.

What is rare is a system still alive six months after launch, with someone watching it and correcting its drift. That question sorts vendors faster than any credentials slide, so we answer it ourselves: we build and operate our own products, on our own money.

Our products in production

An audit product that writes the fixesIt checks websites for compliance, security and accessibility, then drafts the corrections rather than listing problems.
A monitoring product that fights false alarmsA page can return a healthy status and still be broken for the person using it. Catching that without crying wolf is most of the work.
A lesson both taught usA tool that misfires often gets switched off within a week, and then it protects nothing.

From the task map to hours you can count.

Not a transformation program: a map, a priority order, and waves of automation with the return measured every time.

Half a day
A ranked list of what eats time.
With the people doing the work, not only those commissioning it. Yours whether or not you continue with us.
First wave
Two or three agents in production.
Wired into the tools you already run, with the starting point recorded before a line is written.
Before widening
A system that holds.
Edge cases watched in production, misfires tuned, the next wave only once this one runs clean.
At handover
Your team in control.
Documentation, a runbook for when something breaks, one person trained to operate and adjust.

How many hours could an automation give your team back?

Three settings, instant resultEstimate the recoverable time

No magic percentage. Enter your own numbers and the estimator projects what you would get back. The share that is realistically automatable is yours to set, not ours.

Gross time potentially freed up

7,5 h / week

Hours spent on this task each week

Each square is one hour. Green ones are the hours an automation could take over.

hours a month

4,6

person-days a month

person-days a year

Every figure is a team total, not per person. Conversion assumptions: 4.33 weeks a month, a 7-hour day, 11 working months a year. They are written down so you can challenge them.

A projection from what you entered, not a promise. And it is GROSS time: it does not deduct human review of the output, edge cases, maintenance, or the time the team needs to adopt the change. The net figure comes from mapping your actual tasks, at the start of an engagement.

What you sign for AI automation, in short.

Price: A fixed fee per wave, set after the task map. We sell a result on a scope, not days.
Measure: A baseline recorded before the build; volume, accuracy and hours recovered reported monthly.
Data: The hosting level of each task written down: which model, which region, which contract.
Human control: No output that commits your company leaves without a person approving it.
Exit: Code, workflows, prompts and documentation are yours. Maintenance afterwards is optional.

The commitments, with the limit of each

Who builds your automations, and who measures them.

Three senior engineers, two of whom have worked together for more than twenty years. No data team to hire on your side: the people who map your tasks are the people who build and run the agents.

Laurent, founder and CTOMaps the tasks with your operators, sets the baseline and the accuracy target.
Clairmont, technical leadBuilds the agents and the integrations, and sets up the logs and the hosting.
David, frontend leadDesigns the screens where your people review and approve what the agents propose.

Tasks, tools, AI Act: straight answers before you start.

Which tasks does AI actually handle well today?

Four families hold up in production:

Sorting and routing what comes in: messages, forms, scanned documents.
Drafting first versions of quotes, standard replies and meeting notes that a person then reviews and signs.
Summarizing long documents such as contracts, reports and regulations.
Searching a body of knowledge in plain language, so the right document surfaces even when nobody remembers its name.

Outside those, results get thin fast.

What should never be automated without a human in the loop?

Anything that commits the company: accepting a return outside policy, granting a discount, approving a refund, rejecting an application. Those calls weigh customer satisfaction against margin, and a model has no business context to make them alone.

It can propose, with its reasoning and its confidence level, and a person approves. That takes under a minute and prevents the expensive kind of mistake.

How do you prove the time saved is real?

By recording the starting point before building anything: how long the task takes today, who does it, how often. Without that number the promise cannot be verified, and unverifiable projects get quietly dropped.

We then report three figures monthly:

volume handled automatically
accuracy against a target set during scoping
hours recovered compared with the baseline.

Why do AI pilots stall before production?

A demo only has to look plausible; production has to be correct.

Most stalled projects started from the technology rather than a named task, so nothing was measurable; or fed a model scattered, contradictory data; or treated adoption as a consequence, so the tool ran correctly and people ignored it.

The way around is unglamorous:

one precise task
a baseline recorded first
a person on the decisions that carry consequences
the next wave only once the current one holds.

Which tools do you build with: n8n, Make or custom code?

Whichever the task and the data call for, and the choice is written down with its reason.

n8n self-hosted in the EU when workflows handle personal data or must run on infrastructure you control.
Make when a non-technical team needs to edit its own workflows and the data allows a hosted service.
Custom code when the logic is complex enough that a visual workflow would become the thing nobody dares to touch.

We do not resell licenses, so the recommendation carries no commission.

Does the EU AI Act apply to our automations?

Most business automations, such as sorting requests, drafting replies or summarizing documents, fall under the transparency duties rather than the heavy obligations: people must know when they are dealing with an AI system.

An automation that screens job applicants or scores creditworthiness is a different matter, since those uses are listed as high-risk, with obligations from December 2, 2027.

We map each use case at scoping, and your counsel confirms the classification.

Do you offer AI training for our staff?

No. We build, run and measure the automations; we do not sell training courses. The person who owns the system on your side receives the documentation, a runbook and a handover session, which is what they need to operate it and adjust it.

And when the question is where the AI is allowed to run?

Lead and modernize

Someone senior to decide, on a bounded mandate.

Closer to your case

AI under GDPR and the AI Act
Your counsel needs the answers to be in the architecture.

All services

Name one task that eats your week. We tell you if it is worth automating.

20 minutes to check whether it is a good candidate. If the honest answer is that a simple script would do it, we say so.

Tell us about your project

A 20-minute video call