Companies have never had such easy access to artificial intelligence tools, but are still struggling to transform them into measurable operational gains. In Nantes, Edison AI offers SMEs and mid-sized companies a model based on the integration of agents into business processes, annual support and a guaranteed return on investment within six months.
If equipping a company with ChatGPT, Copilot or Claude licenses only takes a few minutes, changing the way it prepares its quotes, processes its customer requests, plans its operations, controls its documents or manages its activity requires more time than a software subscription.
Data is dispersed between ERP, CRM, messaging, Excel files, production tools and document bases. The teams do not all work at the same pace, access rights are heterogeneous and operational managers must be able to check what the tool does. It is between the general tool and the real work process that Edison IA intends to position itself.
Founded in Nantes by Anaïs Vivion and Julien Hervouët, the company does not only sell access to an artificial intelligence model. It markets an annual program which combines acculturation of the management committee, diagnosis, prioritization of use cases, deployment of agents and measurement of the economic impact. The name of this program: Lightning Program.
Uses that are progressing, results that remain to be demonstrated
The adoption of AI is now underway in French companies, but remains very dependent on their size. According to INSEE, 31% of companies with 50 to 249 employees declared using at least one AI technology in 2025, compared to 58% of those with 250 or more employees.
This is the paradox of the moment, leaders have access to an increasingly abundant supply, but they often lack an answer to three simple questions. Which process should be transformed first? What data should be connected? And how can we establish that time saved actually translated into lower costs, additional capacity or improved service?
This difficulty in moving from prototype to deployment is also observed in large organizations. As MAIF’s AI strategy has shown, the challenge is not to add tools but to organize use cases, data, access rights, control rules and the possibility of changing supplier.
The segment targeted by Edison IA is between 50 and 5,000 employees, it brings together already structured SMEs and ETIs, i.e. an order of magnitude of around 25,000 companies in France. At this stage of development, these organizations generally have an information system, support functions and sufficiently recurring processes to justify a cross-functional project, but not always internal data or AI teams capable of carrying it out alone.
Put models at the service of business tools
The offering is based on the Edison Hub, a platform presented as a connection layer between the company’s AI models, knowledge bases and software. The challenge is not to build a proprietary model, but to connect existing models to the information and actions that make up daily work.
Edison AI distinguishes three types of agents. Conversational agents assist employees or respond to customers. Autonomous agents chain business tasks together, according to defined rules. Application agents use company data in a dedicated interface to produce analyses, indicators or dashboards.
This distinction is essential, because an agent is no longer limited to writing a response or summarizing a document, it queries data, applies rules and can trigger an action in business tools. It is precisely this passage from the assistant to execution that FRENCHWEB.FR analyzed with regard to AI agents in companies and the Parloa case.
The company also highlights multi-model logic: ChatGPT, Claude, Copilot or another model can be used depending on the use cases, security constraints and cost. This technological independence has become a commercial argument: models are evolving rapidly and companies want to limit their dependence on a single supplier.
The value of such a system therefore lies less in the model than in its implementation: connections to software, access rights, quality of documents, validation rules and adoption by the teams. These are also the necessary prerequisites for deploying AI across an organization, beyond the first pilots.
50,000 euros per year for demonstrated results
The Lightning Program is offered at 50,000 euros excluding tax per year, after a training and diagnosis phase. Edison IA announces termination free of charge if the return on investment is not demonstrated within six months, according to the terms provided for in the contract.
This is the most unique element of its positioning, because when most publishers charge for licenses per user, API consumption or integration projects, Edison IA promises to share part of the risk linked to deployment.
The promise, however, raises the question of how to measure the ROI of an AI agent? A quote prepared ten times faster, documentary research reduced by several hours or automated administrative processing do not automatically produce savings. The gain must be able to be reinvested in more production, improved service, reduced errors, avoided recruitment or a reduction in external expenses.
The strength of the guarantee will therefore depend on the precision of the indicators retained with each client: starting point, volumes processed, value attributed to time saved, access to data, level of adoption of the teams and exclusion conditions. In other words, Edison AI will have to sell a measurement method as rigorous as its agents.
Competition on several levels
The major providers of office templates and suites, Microsoft, OpenAI and Anthropic, offer access to technologies and productivity tools. Integrators and ESNs, often partners of large clouds, can deploy more complex projects in the information system. Consulting firms, automation agencies and AI agent specialists support companies in the choice of use cases, training and first pilots.
Competition has already intensified, the DGE and Hub France IA have listed 88 solutions adapted to SMEs and ETIs in their 2026 catalog, while the Hub France IA map lists nearly a thousand French players. Offerings range from business co-pilots to knowledge management platforms, from customer relationship automation to industrial quality control.
Edison IA is seeking an intermediate position: neither a strategy firm without tools, nor a simple license publisher, nor an integrator reserved for large accounts, its ambition is to package deployment into a subscription offer, with a catalog of agents, a presence among teams and a network of managers via the Edison IA Club.
The challenge: transforming a service into a reproducible model
The market can become large if the promise is kept. Taking the annual price displayed as a reference, 100 member companies would represent approximately 5 million euros in gross recurring revenue. This is the objective set by Edison IA for 2028.
But the equation remains demanding, the company, which has around ten employees, will have to enrich its catalog of agents and recruit in deployment and commerce while avoiding each new client becoming an entirely tailor-made consulting project.
Its ability to reuse connectors, use cases and industry methods will determine the model’s margin. Conversely, if each integration requires a significant mobilization of experts, the platform risks remaining an extension of a service activity.
The issue is also commercial, SMEs and ETIs do not buy AI like they buy an office license: they commit their processes, their data and their teams. Edison IA will therefore have to demonstrate, file after file, that the announced gains are not only observed in a pilot but captured in the organization.
Edison IA announces that it has raised 1 million euros in Seed from Go Capital, with the participation of CIC Ouest and Bpifrance. The funding must be used to strengthen the development and commercial teams, enrich the Edison Hub catalog and extend the model to the national level.
