Council · AI Governance
Governing your agents
Your teams deploy AI and agents in days. Your governance takes weeks, and no one has a complete list of what's running. In six weeks, we'll implement a system that keeps pace: a registry of your AI systems and agents, qualification of each one against the AI Act, an approval process that doesn't hinder business operations, and an AI committee that's sustainable.
Starting from €4,000 excluding VAT for inventory and qualification · Written proposal within 24 working hours, without obligation.
The problem that all large organizations face
A Copilot agent created by an account manager, a custom GPT shared within a department (OpenAI announced its withdrawal by December 2026), AI embedded in a vendor's software, a pilot program launched by R&D: each of these systems entails your responsibility, and most have never been declared. The AI Act now mandates an inventory, risk level classification, and user training. Governance platforms automate this work, but they automate a process that, for most of our clients, doesn't yet exist.
- Knowing exactly which AI systems and agents are in service or planned, who is responsible for them, and what data they process.
- Assess each system against the AI Act and your own rules, and prioritize those that present a risk.
- Give the trades a simple declaration process, with a response in days rather than weeks.
- Set up an AI committee that decides, documents and reviews, without becoming a bottleneck.
Our four-step method
Six weeks, four to six days of intervention spread out, and an operational system at the end, built with the departments that actually deploy the agents.
Inventory
To record everything that is happening, including what no one has declared.
- Interviews by management and short questionnaire to AI ambassadors and business experts.
- Extraction of existing administration consoles: Copilot and Copilot Studio, Claude, ChatGPT, Gemini, Mistral, Dust, Dataiku depending on your equipment.
- Review of supplier contracts to identify the AI embedded in your software.
- Deliverable: the AI register, with owner, purpose, data processed, provider, status and review date.
To qualify
Classify each entry in the register and know what to do with each one.
- Risk level as defined by the AI Act: prohibited practice, high risk, transparency obligation, minimal risk.
- Associated obligations: documentation, human supervision, informing people, training users under Article 4.
- Internal three-verification grid, applied in the workshop with the system owners.
- Deliverable: the qualified register and the list of systems to be processed as a priority.
Submit
Design the "I want to deploy an AI or an agent" journey.
- Ten-question declaration form, completed by the profession in less than fifteen minutes.
- Self-approval rules for minimal risk cases: this is what avoids the bottleneck.
- The others will go through the AI committee, with roles, decision criteria and target deadlines.
- Lightweight tools, integrated into your existing tools: SharePoint or Notion lists, Forms forms, Power Automate approval workflows. If the volume warrants a dedicated platform, we will draft the specifications and help you choose, independently.
Start
Set up the committee, the standard file and the rituals, then train those who will carry them out.
- The AI committee was set up and engaged in its first two sessions, with the first real cases being addressed.
- Typical file per system: form, evaluation, decision, conditions, review date.
- Quarterly registry review ritual and removal of obsolete systems.
- Two-hour session for committee members and AI liaisons from the departments.
Why a firm rather than a platform?
AI governance platforms are useful when the process already exists and the volume of systems justifies it. Before that, they simply add another tool to feed. Our approach is to first build the system with your business processes, within your existing tools, and connect it to what you already have.
- A system that fits within your existing tools, without new licenses or dependencies, and that your teams know how to develop.
- A connection with your charter of practices, your AI ambassadors and your adoption roadmap.
- Developing users' mastery of AI, as required by Article 4 of the AI Act, is our core business: the registry directly feeds into your training plan.
- Experience in organizations with high requirements: banking, energy, aeronautics, space, defense.
Two ways to start
A two-day entry point to quickly obtain a qualified register, or the complete mission to install the device from end to end.
Inventory and qualification
A one-day inventory with your management and IT teams, followed by a one-day qualification workshop with the system owners. You leave with the qualified inventory, priorities, and a device recommendation.
€4,000 to €6,000 excluding VAT
Two days, depending on the size of the organization and the number of departments
Governing your agents
The four stages of the method over six weeks: registration, qualification, validation circuit equipped in your tools, AI committee installed and facilitated, training of members and referents.
From €12,000 excluding VAT
Four to six days of intervention spread over six weeks
What you get
- The register of your AI systems and agents, qualified under the AI Act and your internal rules.
- The declaration and validation process, its self-approval rules and its tooling in your tools.
- The AI committee: composition, mandate, decision criteria, standard file, review schedule.
- The committee members and trained AI mentors, and a trajectory to continue with a part-time AI leadership if you wish it.
Measured results
More than 30 organizations supported, 4,850 professionals trained and an average satisfaction of 9,4/10. See our references →
FAQ
What exactly does the AI Act require of an organization that uses AI?
For a user of AI systems, the obligations revolve around three things: knowing which systems are being used and what level of risk they belong to, respecting the obligations specific to that level (transparency, human supervision, documentation for high-risk systems), and ensuring that the people who use them are trained, under Article 4. The mission covers these three aspects.
Do we need an AI governance platform?
Not initially. The system we install integrates with your existing collaborative tools and is sufficient for the vast majority of organizations. If your infrastructure comprises several hundred systems, or if your audit obligations require it, we will draft the specifications and help you choose a platform, independently of any vendor.
How do you find AIs that no one has declared?
By combining three sources: interviews by management, administration consoles of the tools in place (Copilot agents, personalized GPTs, Claude projects, Dust or Dataiku spaces) and the review of supplier contracts to identify the AI embedded in your software.
Who should sit on the AI committee?
A sponsor from senior management, the IT or security department, the legal or compliance department, the data protection officer, and two or three representatives from the business units that deploy the most. The committee only deals with cases that don't require self-approval, which keeps it lean.
How long does the mission take?
The entire mission takes six weeks, with four to six days of intervention spread out over time. The "Inventory and Qualification" entry point is completed in two days, generally one week apart.
How much does it cost ?
The two-day entry-level service is offered between €4,000 and €6,000 excluding VAT, depending on the size of the organization. The full service starts at €12,000 excluding VAT. A detailed proposal will be sent to you within 24 business hours after an initial discussion.
The AI Systems Governance Kit
Before entrusting us with the task, or to get started on your own: our Word-format kit for inventorying, qualifying, and governing your AI and agents within your collaborative tools. The kit includes the register, the declaration form, the AI Act qualification grid, the three verdicts, and the standard case file, just as we implement them for our clients.
- The register has fifteen columns, with one example row filled in.
- The declaration form in ten questions
- The risk level qualification grid and the decision tree
- The verdicts, the self-approval criteria, the standard file, and the AI committee
Receive the kit
The download link will be sent to you by email.
Align your governance with the pace of your deployments
Describe your context in a few lines: organization size, AI tools in place, sector, regulatory deadlines. We will get back to you within 24 business hours with a proposal. You can also book a 30-minute consultation directly.
Do you prefer to call or write? +33 1 89 16 48 05 or hello@nextstart.aiYour contact details are only used to reply to you.
NextStart is a training organization registered under number 11756140675. This registration does not constitute state approval.
