Generative artificial intelligence will come into focus in 2023 among the Data departments of French companies. It is not surprising that the subject remains at the heart of their concerns for the current year.
“The tsunami hit at an unprecedented speed for a technology that has quickly become one of the priorities of data departments,” immediately indicates the barometer published by Quantmetry and Capgemini.
Data boosted by the rise of generative AI
Many CDOs declare that generative AI has been a boost and an accelerator. The study confirms this: 80% of respondents believe that data is gaining priority within their company with the rise of generative AI.
A boon probably for leveraging efforts around the various fundamentals of a Data strategy, such as data quality, governance or scaling up. According to the barometer, GenAI is not the only priority for Data departments – even if it does occupy a preponderant place.
Four major areas of work are identified regarding generative AI. This will involve quickly launching new experiments. The goal is simple: “to meet expectations and form initial convictions about the value of uses and technological choices.”
Last summer, 25% of respondents had already launched at least one AIGen experiment and more than two thirds of the working groups. The movement has significantly accelerated and uses mainly revolve around RAG (Retrieval Augmented Generation).
Train to make interaction a skill
The next wave of use cases will undoubtedly target office applications, including Microsoft’s Copilot tools. Their massive deployment is not a given, however, and many practitioners wonder about the real ROI of these solutions.
Data departments are also keen to train users, training being essential for large-scale deployment. The objective: “treat interaction with generative AI as a skill to be widely mastered.”
But increased use of IAGen also requires a risk assessment and an adaptation of the operational model. Managers are indeed expressing fears about the reliability of the models. They will have to work on it, just like how to move from PoC to production.
The “respondents share several questions about the elements to integrate in order to be able to move on to first uses in production in 2024. On the technology side, while the time has come for testing, the evolution of MLOps platforms and tools is only just beginning among the most mature players,” reports the barometer.
The challenge of impact at scale
But achieving these different ambitions for generative AI requires a sufficient level of maturity. Behind the glitter, we find foundations, and not all of them make the profession or directions dream.
However, according to the study, “data departments are still struggling to break their current glass ceiling and have a strong impact on the scale of the company and its business model.” Data Offices must change scale.
However, many of them are prisoners of a “data DOer position.” This results in a bottleneck and difficulty in moving away from “tactical” and non-strategic uses for the company.
Three major challenges must be met for Data Offices: being part of the strategic challenges of their company; drive the data department through value; multiply the impact of data at scale (with Data Mesh).
Succeed in the Data Mesh by moving away from it
The Data Mesh speaks to Data departments, or at least its main principles. Nearly 80% are starting to deploy them to meet this challenge of scaling up impact. However, these initiatives suffer from a lack of coherence.
Data Mesh projects “still too often emerge in an uncoordinated manner and without a common trajectory with stakeholders”, notes Jonathan Cassaigne, Director of Expertise at Quantmetry
And the best way to succeed in the Data Mesh is perhaps to move away from it. Among the best practices is that of distancing oneself from the theoretical framework (“often deemed unattainable”) in favor of accessible victories around data domains and data products.