JEMS is a French company that specialises in Big Data, creating data platform models with a unique industrial manufacturing process. JEMS believes that it is only by coupling generative AI with predictive AI that qualified services can be generated for businesses based on their data. A joint interview with Nicolas Laroche, Managing Director, and Jacques Benhamou, President of JEMS.
JEMS« core business is creating, managing, and exploiting our clients» data assets, with the aim of making them more valuable. The objective is for this data to become a real company asset. The stakes are high, as 30 to 40% of global GDP is based directly or indirectly on products linked to data or its consumption. According to a 2022 Gartner study, 85 % of data projects failed to meet their objectives, meaning they either didn't work technically or didn't deliver the expected service. This observation is mainly linked to the fact that the world of data is perceived as a commodity, whereas it is a "core business" that requires an industrial approach.
JEMS has thus rethought the professional codes of our market, to align with those of the industry in order to optimise client support. We have set up a genuine data product factory. Our modelling method allows us to deliver different use cases without having to completely rebuild each project.
Our clients are very large companies, but also medium-sized companies (ETIs) in France, Switzerland and Europe, as well as on the African continent.
«Data is perceived as a commodity, whereas it is a core business that requires an industrial approach.»
Depending on their maturity, our clients entrust us with the entire scope of their project, either from the outset or after a certain period, which is generally 4 to 6 years. This latency period, observed in companies with a consumerist view of data, is linked to the fact that they multiply initiatives without seeing their project through to completion.
By trying to create new use cases based on previous ones, these companies make a mistake. Indeed, this procedure, which links the use cases together, makes their model complex to maintain and unusable as soon as the number of cases increases.
JEMS manages this technical debt, readjusts and recovers the entire scope. We also provide a complete turnkey solution for first-time buyers.
JEMS's advocated asset-based approach is the opposite of a consumer-based approach. Banking, energy, insurance, e-commerce, luxury, transport, aerospace or automotive industries… our universal method adapts to data from all business sectors. Only the use cases and products are tailored to each sector.
AI is a support function that facilitates certain actions. Therefore, most AI applications, such as predictive maintenance, are restricted to a very specific use case.
Generative AI is totally different as it creates content. To create this content, there are 3 solutions: use a trained model, train a new model, or do a mix of the two. In the first case, it's the ChatGPT type of model. As everyone has access to it, there is no competitive advantage for businesses. In the second case, you use a generative AI model to train it on a given scope. However, this training costs tens of millions of euros; in comparison, previous predictive AI models did not exceed a few thousand euros. There is therefore a gap because companies do not have the technical capacity to structure their data, nor the corresponding budget.
Our vision is to couple generative AI with predictive AI in order to qualify the obtained result and interact with data specific to a business area. Our generative AI model sits between a model trained on broad or public data areas, and a proprietary system that interacts with each client's data.
The utilisation of data from specific sectors offers exceptional added value compared to that provided by open-source open data (Bard, ChatGTP, Alphacode, etc.).
The world of data is rapidly evolving, but we are at the forefront of innovative services. We already have a considerable lead over the competition on the hybridization, the cross-referencing of an AI's generative capability with the scope of specialised data.
As «early adopters», most of our ongoing client initiatives or projects are subject to confidentiality clauses. Generally speaking, researching information and creating content is time-consuming, and generative AI offers a lot in terms of automation.
Attention, Generative AI is not a magic bullet either! Its current limitations are linked to the fact that it is not predictable and is based on a probabilistic approach. Therefore, if you ask a generative AI the same question twice, it will produce two different answers, not necessarily contradictory, but different, which poses a problem for approval. Generative AI therefore cannot be used for qualifying professional purposes, especially in sectors like medicine.
For businesses, the future lies in a combination of predictive AI and generative AI, which will enable the qualification of the confidence level for each response, allowing it to be used without apprehension. With our partner ecosystem, we provide this essential traceability and legitimacy for the professional world and the industry of tomorrow.
