Helping your business evolve with AI.
From diagnostic through to production: studies, programme ownership, business AI agents and change management.
Three pillars
Train and win people over, study and decide, build and run.
Training & change
The turn to AI happens at several speeds. We train, coach and support change over time, until the habit sticks.
- Courses and workshops, off the shelf or tailored
- Coaching for executives, sponsors and power users
- Change management orchestrated by the A.D.A.P.T. method
Consulting & studies
We start by listening to the business, on the ground. We then turn those working situations into use cases that are feasible, costed and prioritised.
- AI diagnostic and use-case mapping
- Feasibility and impact studies, with ROI figures
- Programme ownership, architecture, roadmaps
Implementation
We assemble proven building blocks rather than rebuilding everything: market standards, open source where it makes sense, at the best balance of quality, security and cost.
- Proof of concept and first use cases within weeks
- Business agents and multi-agent systems
- Go-live, with human-in-the-loop workflows
- Then we run them: SLAs, KPIs, quarterly reviews
AI diagnosticA Bpifrance-referenced expert. Several companies supported on their AI diagnostic. Engagements may qualify for co-funding — eligibility to be checked case by case.
How we workOn Behalf AI, an AI operator
Beyond development, integration, testing and go-live, we run your AI over the long term. And we commit to measurable results.
Support and maintenance under SLA, incident tracking through tickets, availability metrics.
Measuring the productivity gains and savings actually observed, not the ones announced.
Quarterly reviews: what worked, what needs to change, what we stop.
Training is essential. Getting it adopted is the real work.
AI touches every function, and not everyone moves at the same pace. A training course does not change a company: impact is built over time, with formats that fit and adoption measured at the level of individuals, teams and the company as a whole.
The A.D.A.P.T. method orchestrates change: observe, design the path, set things moving, practise on real work, measure and embed. It is what separates a series of courses from a transformation.
Map it to decide well, then build it
A two-phase approach: a diagnostic that produces a costed roadmap, then a first concrete implementation.
Implementation
Secure chatbot, a first agent or multi-agent system, and training for your teams.
Phase 1 is an AI diagnostic. As such it may qualify for Bpifrance co-funding — eligibility to be checked case by case.Our work
A few concrete implementations delivered for our clients, from secure chatbots to multi-agent systems.
Secure enterprise AI platform
An open-source solution on Azure OpenAI, Google Vertex, AWS Bedrock or OVH, with RAG, OCR and a LangChain agent stack. Under the client's control, on the client's own infrastructure, with SSO for users.
First-line legal agent
Automated detection of departures from the client's legal policy, proposing alternative clauses as tracked changes directly in the contract.
Workforce planning & job descriptions
Bulk loading, iterative prompting and AI-generated templates to bring hundreds of job descriptions up to date and into line with each other.
Buy-resell automation
A multi-agent system progressively automating every step of 3,000 transactions a year, with a human in the loop.
Proprietary knowledge base, queryable by AI
The company's entire body of business knowledge brought together in a secure in-house RAG — several hundred thousand documents — enforcing each user's own access rights in real time.
Procurement AI agents
Agents wired into buyers' reference data and tools, taking on the day-to-day analytical work and handing back control when a judgement call is needed.
Tooled-up tendering (RFX)
Framing the requirement, writing the specification, building the scoring grid and comparing the bids: the RFX chain tooled up end to end, with the decision left to people.
Secure meeting bots
Bots that join your Microsoft Teams, Google Meet or Zoom meetings, listen, transcribe, write the minutes, and can interact with internal participants during the session. Entirely in-house and secure.
Multi-agent system for office documents
Built to make everyday office tools easier to work with, while respecting the client's own style guides and templates. Fully vendor-agnostic, designed by domain experts.
Use the latest technology without giving up on responsibility
As a French and European company, we guarantee compliance with the GDPR, the DSA and the AI Act. Security is not an option at the end of a project: it is a constraint from the start.
Hosting you control
Azure, GCP, AWS, or sovereign providers such as OVH and Scaleway. On-premise deployment when the stakes require it.
Data kept apart
The RAG vector store stays under the client's control, and nothing is trained on your data. Each user's identity and permissions are propagated from your directory through delegated authentication: nobody sees what they are not entitled to see.
A framework you can trust
GDPR, AI Act and ISO 27001: compliance requirements are designed in, not added afterwards.
End-to-end security
We work hand in hand with the IT department, from the API to the desktop. On this we do not compromise.
We work across every sector and every support function, with four areas of choice
Whatever your sector and your function. Four areas concentrate our tools, our agents and our track record today, as a matter of history.
Achats
We are former buyers, so we speak your language. Training, AI diagnostics and specialised agents, compatible with your ecosystems (Copilot, Claude, ChatGPT, Dust…) and your systems (Ivalua, Basware, Coupa, Ariba, JAGGAER…).
Our AI-augmented cost killing offering sets this out in detail: how it runs, how it compares, and the success-based fee.
Finance
Financial data quality that finally works at scale: analysis of balance sheets, tax filings and annual reports (IFRS, US GAAP), management control, AI-driven dashboarding, FP&A, reconciliation across subsidiaries and figures, taxonomy.
Customer service
Ticket handling at scale, natively interfaced with the main tools on the market: ServiceNow, Zendesk, Jira. Data science on customer journeys, onboarding, reconciliation, multichannel service optimisation. And a knowledge base that keeps growing.
Compliance & risk
Turning regulation into requirements and keeping them alive: security policy, vulnerabilities, patching, logs, traceability, bastion hosts, SIEM, secrets, data leaks. NIS2, ISO 27001.
We are not auditors: we get you ready for audits and certifications, continuously.
Each specialisation runs across all three of our pillars.
AI-augmented cost killing
A close-up on one part of our procurement offering, which is not limited to it. Procurement consultants, equipped with AI agents specialised by category, segment and supplier, analyse the whole of your spend: the mapping goes further, and so do the savings it uncovers.
Loading and indexing all of your documents: contracts, orders, invoices, tenders, reports. No sorting needed on your side — we digest anything that might yield something useful. Read-only access, granted after a security review.
Every document is read, finely categorised, and the key facts extracted: renewal dates, commitments, specific clauses, history, competitive pressure, dependency. Your spend base is rebuilt at a fine granularity, with a savings-potential matrix supplier by supplier. Deliverable: a full Power BI workbook.
Run by our consultants, person to person: email, phone, meetings. You approve every approach before it is made.
Savings measured over the following twelve months. Success-based fee, capped — beyond the cap, everything is yours.
A service first
The point is to find savings — this is procurement consulting before it is an AI offering. It happens that we get there using the best tools available today, and that is what lets us go further, and faster, than a manual review.
What we commit to
Our consultants run every negotiation, you approve every approach, and the AI never passes itself off as a human. Read-only access to your documents, granted after a security review.
Two standing concerns
AI risk and AI cost are not end-of-project topics: we keep an eye on both from the first day of an engagement to the last. When the stakes justify it, either becomes a study in its own right — here is what it covers.
AI CyberRisk Assessment
An AI model is not software like any other: the risks it carries are new, and existing vendor-assessment frameworks simply do not see them.
We equip your security teams — we do not hand out audit stamps.
AI cost optimisation
In three years the market has become unreadable. Choosing a model or a plan is no longer a technical decision: it is a trade-off between performance, security, sovereignty and real cost in use.
And where the amounts justify it, we help you negotiate your contracts with the major AI vendors.
Investing well: neither too early nor too late
AI changes our relationship with time: the tools deployed in 2026 will not stay in production for ten or fifteen years, the way CRMs and ERPs once did.
A relentless cycle
Every company is affected: white-collar work first, blue-collar next as robotics arrives. An agent put into production lives for about a year — certainly not fifteen.
Striking the balance
Aim for short payback periods without sitting on your hands: some technology will always end up on top. Hence deliberately vendor-agnostic solutions, able to move to another platform — including for reasons of sovereignty.
AI that helps, not AI that performs
AI that imitates you is worth little (simulated sales emails, generated calls — AI slop). It should save you time on your analytical work, when you are alone in front of your screen.
What we believe: where AI belongs, and where it does not
Everything you do alone at a screen, without needing to talk to anyone: analysing, comparing, checking, drafting, summarising. That is where AI should act — and it will do it well.
As soon as trust is at stake — a colleague, a client, a supplier — passing AI off as yourself destroys the very thing you are trying to build: a hundred generated prospecting emails will not close a sale. Nor does it replace situational judgement, the kind that picks up the phone instead of sending a third written follow-up. That judgement is precisely what will be expected of teams more and more.
A language model handles language, nuance and instruction-following with impressive ease. It remains a statistical tool, not a brain — all the more reason to put it in the right place.
Damien Juillard — founder of On Behalf AI
A stack we follow and know well
We operate on every cloud, with all the major models, and we pick whichever building block answers the need — proprietary or open source, without dogma.
On Behalf AI is a consulting and systems-integration firm specialising in generative AI. An INSA Lyon engineering graduate, Damien has spent more than twenty years in IT consulting, on business and IT transformation projects, for over seventy clients — Elee, SamBox.io, BNP Paribas, Wavestone.
The way he works owes as much to listening as to experience: getting to grips with a business, its economics and the daily reality of the people who will use the tool, before deciding what to build. Value comes from that grasp of the client's stakes, backed by twenty years in the field.
Over ten years in finance, then IT consultant in software asset management at Elee. AI consultant since 2025.
Expertise carried by real people
Data science, Machine Learning & LLM
Selecting and implementing the models that fit the need, the security requirements and the budget.
Cloud & Infrastructure
Controlled deployments on-premise or in a private tenant — Azure, GCP, AWS.
AI Automation & Workflows
Orchestrating multi-agent systems and end-to-end automation.
Infrastructure & security
Robustness, redundancy, maintenance and long-term operations — an IT department's own criteria, treated as such rather than bolted on at the end.
At Elee, Damien ran IT before the company had a CIO: end-user estate, security, some thirty VMs, software and development. A hands-on engineer who knows the constraints large IT departments live with, and can talk to them directly.
Join On Behalf AI
Help build a growing company, in a field that shifts every month. Curious minds, innovators, pragmatic dreamers: if you share the conviction that AI drives progress when it is put in the right place, write to us.
Where do you start with AI?
A first conversation to frame what is at stake, identify the highest-impact use cases and sketch out a roadmap.
The number is shown as an image after a quick check, to keep it out of automated harvesting and the cold-calling that follows.