At every stage, an engagement that goes all the way.
Nobody arrives at the same place. Some are still working out what the technology actually does, others have a tool running that holds together only because one person keeps watch. Five engagements, one per stage, each with its working framework and what you take away at the end.
Stage 01
Getting the basics right
Understand before you buy
Everyone here talks about AI, nobody knows what we are allowed to do.
Before a single line of code, the people who decide need to know what they are talking about. Not a conference: a plain account of what the technology actually does, what it costs, and what the law has required of you since 2025.
-
1
What it is
Three families, and they have nothing in common: the large general purpose model, the small specialised model, and automation with no model at all. Choosing between them is the first decision that matters, before any optimisation.
-
2
What it commits you to
The European AI Act and its timetable, the GDPR, the duty to tell someone they are talking to a machine, and what leaves your company when a text enters a prompt.
-
3
What you decide
The list of permitted uses, the list of forbidden ones, and above all who rules on a case nobody anticipated.
Stage 02
Value diagnosis
One project, one return
We have ten ideas, we do not know which one is worth it.
Most AI projects fail before they start, because they begin with an idea rather than a task. We invert that: we look at what your teams redo every day, we put figures on it, and we keep exactly one.
The framework
The map of repeated tasks
Four steps, on your figures, not on market averages.
-
1
We list tasks, not ideas
What someone redoes every day or every week, named the way they name it themselves. An idea cannot be measured, a task can.
-
2
We measure the real thing
How many times, how long, at what point it breaks and who cleans up after. No figure comes from anywhere but your company.
-
3
We put numbers on a single case
The expected gain, the method that will verify it afterwards, and the effort it takes. An indicator chosen now, not reconstructed at the end.
-
4
We write down what we will not do
This is the part that saves the most time, and the part nobody does.
Stage 03
Framing
The scope written before the code
We know what to do, we do not know how to frame it.
Framing serves one purpose: making the bad surprises visible while they still cost nothing. We write the whole system before building it, along the four acts our work is made of.
The framework
The four acts
Interfaces, data, interaction, processes. The same breakdown we use on projects.
-
1
Interfaces
One screen per real action, in the order the work happens. The scope narrows before it widens.
-
2
Data
Which entities actually exist, what links them, what must be unique, what may be missing. It is the most expensive decision to correct later.
-
3
Interaction
Where the machine decides, where it suggests, and where it hands back. That switching point is written now, not when a customer complains.
-
4
Processes
The trigger, the job, the outputs, the log. A silent automation is a dangerous automation.
Stage 04
Prototype in production
On your real data
We already had a mockup once. It never got used.
This is the core of our work: we send the engineer to where the problem is. They work at your site, on your real data, and what they ship is useful from the first week. A mockup that never touches the real thing proves nothing.
The framework
The engineer at your site
On site, in batches, with a way out at every batch.
-
1
They move in
On site, with your access and your real data. Not a fixture, not a sandbox that resembles your trade from a distance.
-
2
They ship in batches
Every week something usable, put in the hands of the people who do the work. They are the ones who say whether it is good.
-
3
We measure
The indicator chosen at step 02, before and after. With its method, not a percentage out of nowhere.
-
4
We stop or we continue
A prototype that does not move the indicator stops. That is written at the start, so it is not a negotiation at the end.
Stage 05
Industrialisation
What holds without us
It works, but only because someone keeps an eye on it.
A tool that depends on one person is not a tool, it is a debt. This step makes the system operable by your own teams, measurable in its cost, and repairable when it breaks.
The framework
The handover
Four workstreams, until you no longer need us.
-
1
Accounts and roles
Who sees what, who can do what, with an audit log of sensitive actions. This is what lets a tool leave one team and serve the whole company.
-
2
Operations
Backups, a restore that has actually been tested, an alert on failure. A job that fails must tell you, not be discovered three months later.
-
3
Cost
What each use consumes, tracked over time, and what gets removed when it serves no purpose. We only use what is necessary.
-
4
Handing over
Operations documentation, training for your teams, and one contact who knows your system when you need them.
Across every stage
What does not change from one stage to the next
Six rules that apply at the first workshop as much as at the last deployment. They come from our position memo on AI, and you can hold us to them: if we break them on your project, you have grounds to say so.Closed corpus, cited sources
The system answers from your content, says where the answer comes from, and says when it does not know.People keep what carries the value
Advice, sales, judgement, relationships. The machine routes to the competent person, it does not take their place.Data under control
Identified hosting, no training on your exchanges, ownership, export and deletion written into the contract.Replaceable supplier
The model call is abstracted, a no model fallback exists, and nothing depends on a single vendor.Measurement published with its source
A figure comes with its method and its date. Never a saving percentage we have not measured.
Where to start
Tell us where you stand, we will tell you which step to enter by.
You do not have to go through all five, nor to start with the first. Most projects enter at stage 02, because the target has not been chosen yet. One conversation is enough to find out.