Plan 1
Mapping & ROI
€1000 to €2000 VAT not applicable
The starting point for almost every project. I don't build anything: together, we identify what takes time and where it's worth optimizing.
AI for SMEs, mid-caps and freelancers in France
Every week you redo what an AI would do in minutes.
AI isn't a shortcut it's a new way to move forward. Every company follows its own path. Together, let's identify which tasks to automate, make the right decisions, and evolve your processes so you can move faster without losing control.
- Per week per engineer
- 10 h
- First specification draft
- 4 min
- Of data, measurement and ROI
- 9 years
Instant AI answer No sign-up
Describe the process that takes up most of your time, in one sentence.
I break it down into tasks, show you where AI changes the economics of the process, and give you one action to take within 24 hours. You can then ask me your questions.
Your sentence is kept neither on this site nor by the models that answer you. What happens to your sentence
They trusted me
01 _
The problem is not your tools. It is your methodology.
In most companies, small or large, the lost hours go unseen: the same search redone, the same copy-paste, the same review, every week, by people paid to think who spend more time writing and formatting than on the issues that matter to your business.
-
The hidden cost
Nobody has costed this time
This time is buried in the working day: nobody sees it, so nobody decides to optimise it.
-
The false start
AI everywhere, but no methodology
Licences for everyone, a trial that impresses in a meeting, but no process actually defined.
-
The ceiling
Automations that break at the first surprise
A fixed rule cannot read a badly worded email or an off-format quote. An AI model can: that is where it pays.
02 _
Every company will run on AI.
The question is no longer whether, but where it will make you money. My process always starts with observation, never with the tools to roll out.
-
Process first
I start from how the work is actually done. Not from a tool to roll out.
-
The numbers decide
Current cost, frequency, errors, delay. If the numbers do not hold, I do not build.
-
It runs for real
You leave with tools that run, and you know how to evolve them.
03 _
Two ways to spend the same AI budget.
Many companies pay for subscriptions and trials, then get the invoice without knowing what it brought in. I do the opposite: we put numbers on it first, you spend afterwards.
-
Consuming AI
Paying per use but optimising nothing.
- Counting the requests sent, not the hours saved
- It feels like saving, but AI tool costs explode
- The demo pleases, but no workflow runs on its own
-
Making AI pay
Every euro committed must come back.
- A gain measured in hours and euros
- A cost known before anything starts
- A workflow that runs on its own
- Consuming AI, the slide into the trough
- Making AI pay, the climb to the plateau
- The cycle as it unfolds with no one at the wheel
-
Technology trigger
A new model comes out. Everyone talks about it.
-
Peak of expectations
Everyone is an expert. Nobody is in production.
-
Trough of disillusionment
The invoice arrives. The pilot stops.
-
Slope of enlightenment
The survivors learn what actually holds.
-
Plateau of productivity
Boring. Profitable. Everywhere.
04 The proof R&D laboratory
Nine days for a specification. Today, a first version in four minutes.
- 6 h of research
- 3 h of data entry
- 4 h of writing
- D+9
- 40 s
- 4 min
Three engineers qualified compounds one by one: document research, regulatory data, consolidation, then writing. The same gesture on every project. I kept the expertise and removed the handling.
Today, 6 steps, 14 h per specification After, 5 steps, 4 min to the draft
- 01 Before ·A compound to qualify lands on the deskStarting point
- 02 Before ·An engineer goes looking for the documentation, source by source After ·An assistant that knows your files pulls out the useful data, written the way you write itManual, 6 h Bespoke skill
- 03 Before ·Everything is copied by hand into a spreadsheet After ·A regulatory agent that searches the sources your way, and cites every referenceManual, 3 h RAG
- 04 Before ·The specification is written from a blank page After ·A writer that produces your format, with your references, and flags what it is unsure ofManual, 4 h Writing agent
- 05 Before ·An expert reviews it, right at the end of the chain After ·Your expert validates, corrects, and the system learns from every correctionThe expert decides Human validation
- 06 Before ·The specification goes to the client After ·The same specification, out on the day it was asked forDelivered D+9 Delivered D+1
- 10 h
- Recovered per week per engineer
- ×3
- Engineers concerned
- D+9 → D+1
- Time to deliver the specification
- 0
- Lines without a verifiable source
What my clients say.
Le premier dashboard Looker Studio fiable et qui ne plante pas toutes les 2 secondes.
Grâce à l’ETL créé par Loïc, nous avons enfin un reporting multi-régies fiable.
Les rapports automatisés par mail avec recommandations IA ont changé notre pilotage.
Seven steps, from the real process to the measured gain.
A frame I have followed for nine years on data and measurement, now applied to AI automation. Every step produces a deliverable you keep.
-
Process discovery
Understanding how the work is actually done.
People, information, decisions, tools, exceptions, volumes. The floor, not the org chart.
-
Process economics
Putting a price on every step.
Time spent, frequency, errors, delays: every step deserves a cost in euros.
-
Mapping the AI opportunities
The right technology, step by step.
For each step, choose between keeping it human, setting an automatic rule, or calling an AI agent (skill, search in your documents, connection to your tools).
-
Prioritising by ROI
A priority queue, not a wish list.
What pays the most with the least risk goes first.
-
Architecture
Decisions written down.
Which AI model (sovereign, frontier…), at what cost, with which data, and where a human must validate.
-
Going to production
It runs at your place, and you know why.
The system runs at your place, your teams know how to operate it and evolve it.
-
Measurement
The proof, before and after the setup
Money saved, minutes gained, errors avoided, delays cut: measured before and after the setup.
Four layers, one single system.
Tools are an end, not a necessity: ideally, first define the right architecture that stacks on what you already have, with every extra tool having to make the case that it is indispensable.
Layer 1
Your tools and your data
The starting point: I first read your data and your tooled workflow without touching anything; we connect the other tools the agentic workflow needs, read-only first, then with write access once the trust is there.
Layer 2
The shared memory
Your standards, contracts, history and procedures become a base the AI can query, always citing its source.
Layer 3
The domain systems
One AI agent per task, able to communicate: one extracts the data, another writes, a human validates.
Layer 4
Where it runs, and the proof
On your premises, on a high-performance cloud or on a sovereign cloud, your choice. Access, history and gain measurement in one place.
Ten families of work already built, with their lead agent, their sub-agents and their memory.
Four hats. One single contact.
No prototypes or trials abandoned after a month: tools you use every day.
-
Embedded consultant
I work with your CRM, your data, your workflows. Production systems your teams own.
-
Data and acquisition consultant
Finding and qualifying prospects, watching the competition, measuring what works (scoring, enrichment, outbound sequences). Without spending your days on it.
-
AI builder
Agents, bespoke skills, MCP integrations, multi-model pipelines.
-
Embedded trainer
Workshops, leadership sessions, side-by-side support until you build without me.
What those four hats produce, family of work by family of work.
No imposed toolkit. No default model.
-
Your tools are the starting point
I build inside the tools you already use, not around them.
-
Fitted to your situation
Team skills, budget, security requirements, maturity. The solution respects all four.
-
A model chosen per use case
The AI model is chosen for the task: its reasoning depth, its speed, its cost and how it respects privacy.
-
Everything is handed over
Documentation, training, handover. Your teams own what is built.
Prices known before we start.
Ranges, depending on the number of processes, the tools to connect and your regulatory constraints. The diagnosis is deducted from the build if you go ahead.
Plan 2, most requested
Building an AI system
€2000 to €10000, VAT not applicable
From drafting the process to its deployment and launch. The range depends on the number of integrations and the level of regulatory constraints.
Track 03
Embedded AI partner
€6,000 to €12,000 per month, VAT not applicable
For companies with more than one process to rework that want to build the skill in-house rather than outsource it forever.
Looking only to measure your sales and campaigns (tracking, attribution, reporting)? That is the measurement step of the method, also offered on its own because it is my original trade :)
First step
One process, twenty minutes, a costed answer.
You give me the repetitive task that costs you the most time today; by the end of the call, you know whether AI can reduce it, and at what price.
Awards
- 2022 Prix du numérique Salon de l’entrepreneur de l’Allier
- 2021 Finaliste eCommerce Speed Up Challenge Cloudways