Thirty minutes with an engineer, and you'll know what AI can change in your organisation.
You talk with an engineer who has already shipped this kind of system in production, with no salesperson and no scripted pitch. He assesses your case candidly: what is worth committing to, and what can wait. If you decide to go further, we design the solution, integrate it into your tools and operate it over the long term.
Your data stays in Europe — or in Canada, if that is what your rules require.
organisations equipped with our GDPR compliance software
43 days
of work given back in a year to a consultancy we built a tool for
Since 2019
we design and operate our own products in production
2 countries
France and Canada — the GDPR and Law 25 practised daily
the essentials
What you need to know in 30 seconds
A sovereign AI agency designs, ships and operates artificial intelligence solutions. It guarantees that its clients' data stays in their jurisdiction, under their legal framework: the GDPR in Europe, Law 25 in Quebec. That data stays out of reach of extraterritorial access laws such as the Cloud Act.
DPLIANCE is a sovereign AI agency founded in Poitiers, France, in 2019 and present in Montréal. It builds document assistants (RAG), AI agents with guardrails, document extraction and business automations. It integrates these solutions into its clients' existing tools, such as ERP, CRM and business software, and then operates them day to day.
Its GDPR compliance software supports more than 500 organisations. ADEME, the French ecological transition agency, picked it out of its innovation challenge. ADEME then asked it to take the agent that pre-screens its funding applications into production.
The first conversation takes 30 minutes, over video, with an engineer, with no commitment.
Most AI projects that fail rarely die of a bad algorithm: they die between the prototype and production. The demo impresses in a meeting, then the project runs into integration with the information system, the quality of real data or compliance requirements. It then stays stuck in the lab. The budget is spent, and the gain never reaches the teams' daily work.
Second blocker: the most visible AI tools are American, and your legal counsel will eventually ask the right questions. Where does the data go? Who can access it? What happens if the vendor changes its terms or its pricing? For an organisation subject to the GDPR or to Law 25, the lack of a clear answer is enough to freeze a project. That freeze can be permanent.
Meanwhile, your teams keep re-typing invoices, hunting for the right clause across hundreds of documents and triaging a shared inbox that overflows. These are precisely the tasks a well-scoped AI absorbs. It must be designed for production from day one, with your data, your tools and your legal framework.
An AI solution creates value when it is used every day, for years. That is why the same team designs your solution, ships it to production and operates it day to day. Your investment is protected end to end.
01
Design
Your solution is designed for production from day one: your real data, your compliance constraints, your existing tools. What you validate as a prototype is what you will use every day.
02
Production
We integrate it into your information system — ERP, CRM, business software. Your teams keep their tools: the AI comes to them.
03
Operation
Sovereign hosting, monitoring, maintenance and continuous improvement. Your solution is looked after every day by the team that built it, and evolves with your needs.
An engineer reviews your need over video, with no commitment: 30 to 45 minutes are enough to scope it. We look at your data, your tools, your compliance constraints. You leave with a reasoned opinion on the right approach, realistic timelines and the prerequisites on the data side.
2
Prototype
4–8 weeks
We build a prototype on your real data, not on a sample set, with success criteria defined together before the first line of code. You see the solution working on your concrete cases, and you validate on measurable results — not on a demonstration.
3
Production
3–6 months
The solution integrates into your existing tools: ERP, CRM, business software. Your teams are supported as they take it up. Depending on complexity and depth of integration, allow three to six months from kick-off to production. You see the gains in your daily work, not just in a report.
4
Operation
Ongoing
We host, monitor and evolve your solution: fixes, improvements, adaptation to your new needs. It is the same principle as for our own products — we operate what we build. It stays performant, compliant and aligned, over time.
This method starts with a 30-minute conversation, with no commitment: you leave with a useful opinion, even if you never work with us.
Six areas of expertise we deploy and then operate in production. Each follows the same thread: a problem your teams live with, a solution that integrates into your tools, a gain you see in daily work.
RAG & document assistants
Your teams spend time hunting for the right clause in a contract, the latest version of a procedure, the applicable article of a regulation. The information exists, but it sleeps in hundreds of documents — and everyone ends up asking a colleague again.
We build an assistant that queries your business corpus in natural language. Every answer cites the exact passage it comes from: your teams verify the source in one click instead of trusting a black box. Your documents stay hosted in your jurisdiction — a European provider, Canadian options where Law 25 requires it, or your own servers.
The gain: a sourced answer in seconds where it used to take an hour of searching. Hours are given back to your teams every week, and new joiners become autonomous faster.
Who it's for: legal, quality, HR and support teams — any organisation whose knowledge is locked in documents.
Intelligence gathering, meeting minutes, the triage of incoming requests: necessary tasks that eat away at the day without creating visible value.
We design agents that carry out these tasks within a defined frame: a precise scope of action, every decision logged, human validation on critical steps. You can trace every action an agent takes. Your teams keep the decision; the agent prepares the ground. It is the approach that convinced ADEME: pre-analyse the files so a human can decide.
The gain: every file arrives pre-analysed and traceable, and your teams decide on prepared work — faster, with fewer things slipping through.
Who it's for: back offices, support teams, departments that process requests at volume.
Invoices, forms, supporting documents: every incoming document means manual re-entry — slow, costly, and a source of errors that propagate down the chain.
The AI reads your documents, extracts the useful fields and feeds your accounting and business software directly. Ambiguous cases are flagged rather than guessed: a human decides, and the system improves.
The gain: data entry disappears and transcription errors with it; every processed document is traceable, and your teams check instead of copying.
Who it's for: accounting, sales administration, any department that handles incoming documents at volume.
A shared inbox overflows, a CRM is rarely up to date, and documents are anonymised by hand before every share. These tasks all depend on someone being available.
We automate these chains end to end: message triage, reply drafting, CRM updates, document anonymisation. The automation integrates into your existing tools — your teams keep their software, the AI comes to them.
The gain: hours recovered every week, a CRM finally up to date without effort, and tasks that no longer stop when someone is away.
Who it's for: SMEs and mid-sized companies that want quick gains without overhauling their information system.
A classic A/B test takes weeks, a lot of traffic and a team to analyse it. As a result, most websites show the same message to every visitor, for lack of a way to test.
We build websites whose content automatically adjusts to each visitor's profile, and that continuously test their variants. This is fully automated A/B testing, with no third-party cookies and no compromise on GDPR. It is the expertise we are developing within the i3 Impulsion project, the insurance sector's applied research hub.
The gain: every visitor sees the message that speaks to them, every tested variant is measured, and conversion improves continuously.
Who it's for: e-commerce, insurance, services — any website where conversion matters.
Measuring an audience today often means choosing between two flawed options. A consent banner cuts your data with every refusal, while other tools send your figures outside your jurisdiction.
Mirage Analytics, our product compliant with the CNIL exemption, breaks that dilemma. It measures audience without cookies or a banner, hosted in France, designed and operated by our team. We use it ourselves, and it runs in production on e-commerce, industrial and institutional websites.
The gain: you measure your whole audience, beyond the visitors who clicked "accept". That compliance rests on architecture, not on a checked box.
Who it's for: e-commerce, industry, institutions — any website that wants complete, compliant figures.
Do you recognise one of these cases in your organisation? Describe it to an engineer: they will tell you frankly what is feasible, how long it takes, and what it requires on the data side. Discuss your use case
sovereignty
Your data in your jurisdiction, by architecture
The Cloud Act allows US authorities to demand access to data held by American companies, even when that data is hosted abroad. In other words: a server's location does not protect your data if its operator is subject to US law. The problem is the same for a company in Poitiers and for an organisation in Montréal.
Our answer is not a geographic slogan, it is an architecture rule. Your data stays in your jurisdiction, under your framework: the GDPR in Europe, Law 25 in Quebec. The hosting is chosen accordingly: a European provider, Canadian options when your framework requires it, or your own servers.
Choosing a sovereign architecture means keeping your freedom. You choose your models, your hosting and the future of your solution, without depending on decisions taken elsewhere.
Hosting chosen for your framework
Your solutions run with a European provider such as Scaleway. Depending on your needs, they also run on Canadian options when Law 25 or your internal requirements call for it, or in your own infrastructure. Either way, you know exactly where your data is and who can access it. Our own products are hosted on Scaleway, in Europe: we apply to ourselves what we propose to you.
French or European models
We build on French or European AI models. The semantic analysis of our product Complio, for example, runs on Mistral. We can also deploy them directly on your servers. You keep the freedom to choose your vendor and your pricing, without depending on a choice made on another continent.
GDPR & Law 25, operationalised
Compliance is not an annex to the contract: processing records, DPIAs in Europe or PIAs under Law 25 in Quebec, traceability of the AI's decisions. Our GDPR compliance software equips more than 500 organisations — we build compliance into the architecture, not into paperwork after the fact.
Two territories, two compliance frameworks
Founded in Poitiers in 2019, DPLIANCE now works on both sides of the Atlantic. Wherever you are, you work with a team that practises your legal framework every day — not a team discovering it with your project.
Poitiers — France & Europe
Your solutions are designed and operated from France, compliant from the design stage, hosted within your jurisdiction. Since 2019, our compliance software has supported more than 500 organisations, in mainland France and its overseas territories alike.
Montréal — Quebec & Canada
Florian Gadal, our CTO, is based in Montréal. We build Law 25-compliant solutions for Quebec and Canadian organisations, with hosting aligned with your framework and the same support as in Europe.
ADEME, the French ecological transition agency, receives a large number of aid applications that are ineligible or incomplete. Reviewing them takes up its caseworkers' time. We designed a screening agent that automatically analyses each application, in an explainable, traceable way. It checks eligibility, notably through the SIRENE registry API, and the completeness of the file. Caseworkers receive pre-screened files, and applicants can check their eligibility before submitting. The prototype, tested on around a hundred cases, was named winner of the Dis-ADEME challenge. ADEME then asked us to take it into production.
We also take on focused data assignments. We led a data analysis and conversion rate optimisation project for a national automotive retail network. For an e-commerce website, we improved web performance: faster pages, for visitors and for search rankings alike.
That standard comes from the founders' background. Before DPLIANCE, Hichem was an external consultant for LVMH and Axa, and Florian built the applications of the Barreau du Québec. Depending on the project, the team expands with proven data experts.
frequently asked
Frequently asked questions
How much does an AI project cost?
The cost depends on three things: the state of your data, how deeply the solution integrates into your information system, and your compliance constraints. We price each project step by step: the prototype first, then production. This keeps you in control of your budget. A first conversation with an engineer is enough to get a range suited to your case.
How long does an AI project take?
As a rule, allow four to eight weeks for a prototype built on your real data. Production, integrated into your tools, then takes three to six months. Precise scoping shortens the timeline: that is what the first conversation is for. ADEME's screening agent went through exactly that before the agency asked us to take it into production.
What is an AI agency?
An AI agency designs, develops and deploys custom artificial intelligence solutions: document assistants, agents, data extraction, automations. At DPLIANCE, the work goes all the way to daily operation. We build and operate our own Data & AI products, and we apply the same standard to the solutions we build for you.
AI consulting firm or AI agency: which one do you need?
An AI consulting firm helps you set a strategy: it delivers recommendations and a roadmap, rarely software. An AI agency designs and builds the solution itself. At DPLIANCE, the same team does both and stays afterwards. We scope your use case, build the solution, ship it to production and operate it day to day. If what you need first is strategic clarity, a consulting firm can be enough. If you need a solution running in your tools over time, that is our job.
AI agency or freelance developer: which should you choose?
A freelance developer is a good choice for a script, an exploratory prototype or a short assignment. It is lighter to engage, and often cheaper. The limit appears after delivery: hosting, monitoring, fixes, evolutions, compliance. Everything then rests on one person, with the risk that they become unavailable when you need them. An agency like DPLIANCE commits a full team: data scientists, developers, compliance expertise. It stays after production to operate the solution. For a project meant to run in your tools for years, that continuity weighs more than the daily rate.
Why choose a sovereign AI agency?
Sovereign does not mean your data has to move to Europe. It means your data stays in your jurisdiction, under your framework: the GDPR in Europe, Law 25 in Quebec. That framework keeps it out of reach of extraterritorial access such as the Cloud Act. We build on French or European models. We choose the hosting accordingly: a European provider, Canadian options when your framework requires it, or your own servers. You keep the freedom to choose your models, your hosting and the future of your solution, by architecture, without depending on decisions taken elsewhere.
What if the prototype does not work?
It happens. That is precisely what the prototype is for: finding out before you invest in production. Success criteria are defined together before we start. If the results fall short, you know within a few weeks, at a limited cost, and you decide with full knowledge. This staged approach exists precisely to remove that risk before any heavy commitment.
Who owns the code and the models?
You do. The code developed for your project belongs to you, as do the models tuned on your data. If you ever decide to bring operations in-house or change providers, you leave with your solution. We operate your solutions because it is our job — not because you are captive. That, too, is the sovereignty we stand for.
Is our data used to train models?
No. Your data serves your solution and nothing else. We rule out any other use: training models for other clients, or passing data to third parties for that purpose. We use French or European models, hosted with a provider chosen for your framework or deployed on your own servers. Your data stays within the frame we define together, a contractual frame.
Do you work in Quebec and Canada?
Yes. Florian Gadal, our CTO, is based in Montréal, and we were hosted by the Digihub in Shawinigan and Le Camp in Québec City. We build Law 25-compliant solutions for Quebec and Canadian organisations: records, PIAs, data governance, with the same support as in Europe.
Here is exactly what happens when you write to us. We reply within 24 business hours to set a time. You then spend 30 minutes over video with an engineer, with no salesperson and no automated email sequence.
30 minutes with an engineer who has already shipped AI solutions to production — they understand your need technically, not just commercially.
A reasoned opinion on the right approach for your case, realistic timelines and the prerequisites on the data side.
Zero commitment: you leave with a useful scoping, whether you decide to work with us or not.
