Advisory that can build
When a decision calls for building, our production team builds: LLM integration, agents, automations. So we only recommend what we know how to keep running in production.
AI practice
AI consulting for SMEs and mid-sized companies: advisory that makes your projects last
Models chosen on your own examples, pilots followed through to a decision, teams that become self-sufficient: a consultant from our AI practice at your side, month to month, backed by a team that builds.
Data sovereignty: hosted in France by defaultTrusted by the public and private sectorsYou own your intelligence
After the roadmap, making AI stick in the organization
AI consulting, the way Bonne Idée practices it, is an ongoing advisory engagement: we stay at your side while the roadmap is carried out.
Many organizations have an AI roadmap, sometimes a first pilot, and no use case that sticks. The blocker is rarely the technology. More often it is a call nobody makes: two pilots competing for the same teams, a provider that changes its terms, a model that gets it wrong with no one knowing who is accountable.
That is the role we take on: a regular presence, written decisions, and a production team behind us when something needs to be built.
No roadmap yet? The Consulting practice’s AI audit builds it, and AI advisory takes over from there.
Our commitments
What our AI advisory guarantees
A partner working directly with your leadership team
This partner comes from the AI practice and follows your account from the first committee meeting to the handover. You don't explain your context to a new consultant every month.
An exit planned from the start
The engagement aims to make you self-sufficient: trained in-house leads, documented decisions, tools in your name. You own your intelligence, and nothing stops the day you no longer need us.
Our method
The rhythm of an AI advisory engagement
One check-in a month to set the queue, and written decisions between check-ins.
-
1 The starting point
Your roadmap, or the one from the AI audit. We review the selected use cases, the state of the data and the commitments already made with providers.
YouShare the roadmap and the current contracts.UsThe queue of workstreams, in your order. -
2 Choosing providers and models
For each workstream, two or three options compared on your real examples: quality, usage cost, hosting, contract terms, ease of exit.
YouApprove the choice.UsA written decision, and the reasons for it. -
3 Tracking the pilot projects
A pilot has a scope, a success metric and an end date. We track it through to the decision: roll out, fix or stop.
YouName a lead for each pilot.UsA review of each pilot, with figures. -
4 Guardrails and committees
Who reviews what, which data can leave, who is accountable when the model gets it wrong. The rules are written down, and a committee decides whatever they don't cover.
YouConvene the committee when a decision is needed.UsThe usage policy and the decision log. -
5 Handover to your teams
Trained in-house leads, documentation in French, metrics tracked by your teams. The goal is to make the advisory unnecessary.
YouFree up time for training.UsAn organization that runs without us.
AI practice
Building what the advisory decides
Integrate a model, build an agent or an application, or train your teams with the Consulting practice.
LLM integration
Drafting, extracting, classifying: a language model handles it inside your tools, without exposing your data.
AI agents and RAG
Assistants that answer from your documents and cite their sources instead of making things up.
AI MVP and POC
A first version of your product, used by real users in twelve weeks, to find out whether it is worth taking further.
Business applications
A custom internal tool that replaces the spreadsheet and the re-keying, used by your teams in four months, scoping included.
BI productBIA
Our sovereign AI workspace, run locally, for regulated professions and sensitive data.
Consulting divisionAdoption and change management
Train your teams, write the usage policy, set up in-house leads: AI becomes part of everyday work.
ChatGPT
Claude
Gemini
n8n
AI references
Advice from a team that builds
Our own AI agent, CoBIos, co-built with the Orléans Commercial Court Registry.
Data and AI providers: the committee's rules
An advisory engagement touches on the choice of providers, and so on the question of where your data goes. Every decision follows the same rules as our projects: a data processing agreement (DPA), data hosted in France by default (or on your own infrastructure when the sensitivity of the data requires it), plans whose contract excludes training on your data, and a record of processing activities that the committee keeps up to date.
The GDPR and EU AI Act rules that the committee applies are detailed on our AI guarantees and data sovereignty page.
FAQ
Frequently asked questions about AI consulting
A high-volume task where a mistake costs little and the result is easy to check. The first use case is as much about building trust as about saving time. Starting with a critical case is the surest way to discredit the topic internally at the first visible mistake, and to delay every project that follows.
That is exactly what the engagement decides, one type of data at a time. Your data is hosted in France by default, or on your own infrastructure when its sensitivity requires it. Only the data the engagement authorizes goes to a provider, on a plan whose contract excludes training on your data.
Rarely, and it is the most expensive mistake we see. In the vast majority of cases, an existing model fed with your documents is enough, for a fraction of the budget and the time. Custom training is justified when the task is highly specific, the volume very high and the data plentiful, and when that is the case we say so.
For most everyday uses, what matters is knowing which AI tools you use and what for, and taking measures so that your teams learn how to use them. That is what Article 4 of the EU AI Act has required since February 2025. Its July 2026 rewrite (Regulation (EU) 2026/1744) does not require a guaranteed level of AI literacy, but real measures. Text checked on EUR-Lex on October 6, 2026. We map your uses during the engagement.
It runs month to month, as part of the Partner offer, with no commitment beyond the current month, until the handover to your teams: trained in-house leads, documented decisions, tools in your name.
An AI roadmap that isn't holding up?
A free first call, by video: what is blocking you, and what an advisory engagement would change.
Book a call