“What is the best government? The one who teaches us to govern ourselves,” replies Goethe. Applied to data governance, this teaching promotes decentralization and the empowerment of professions.
Because if data is everyone’s business in the company, in practice this involves a transfer of responsibilities and therefore so-called federated governance. This new model is notably promoted by Data Mesh.
The federation, a meal built together
The concept developed by Zhamak Dehghani, a former consultant, however, reflects a trend already underway in certain organizations, for example at Pierre Fabre, a pharmaceutical group. Anthony Asso, its former Chief Data Officer, now interim manager, led this initiative.
“Data Mesh is not a revolution. It is above all an evolution of practices,” he declared during a round table organized by Le Wagon For Business. To describe it, he opts for an analogy.
“It’s a bit like a family meal. A few years ago, we had Christmas dinner at our parents’ house just by putting our feet under the table. Now everyone participates, by bringing dishes or cooking. The Data Mesh, and the role of the Data Office in it, is to organize the meal. We are drawing a line under ultra-centralized governance, which ultimately only produced rules.”
Federated governance preserves rules of course, as underlines Mehdi Benabderrazik, Lead Data Governance for Technip Energies. Its function is to “put in place rules, processes, roles, responsibilities so that data is understood, identified, defined, of quality, accessible, under certain conditions, and protected.”
Data Domain Owner and Data Steward in the field
The Chief Data Officer does not (or no longer) decree from his ivory tower. He co-pilots with internal partners as close as possible to the producers and consumers of the data. Two roles are therefore key in a decentralized organization: the Data Domain Owner and the Data Steward.
The first is the owner of data attached to a perimeter, a domain. The second fulfills more operational missions, for example by assisting the Owners in defining the data, improving its quality, etc.
This approach and these functions are also found at Club Med, says its group CDO, Siddhartha Chatterjee. “We no longer use the term Data Mesh. On the other hand, we base ourselves on its principles of data organization,” he says.
Operationally, the data was therefore distributed between domains, then those responsible for these assets were designated. “I don’t use the word owner, which suggests that an individual or business owns the data. It is the company that owns it, not its employees,” he emphasizes.
Roles, but with responsibilities
These roles also come with responsibilities. Having Domain Owners is not enough, warns the CDO. “We must also organize and define the missions of all those involved who work with the data, producers and consumers alike.”
Since the creation of the Club Med Data Office, less than two years earlier, Siddhartha Chatterjee has worked first on the organization, then on the creation of the community ensuring governance. “It is important to find allies, people who understand the subject and/or are interested.”
Chafika Chettaoui experienced these transformations at the governance level from the inside, within L’Oréal, Suez and now Axa France. The subject was “rather taboo” 10 years earlier. For what ? Because perceived as strictly administrative and a cost center.
Perception has changed, thanks in particular to artificial intelligence, which has become an even more concrete subject for professions since the democratization of generative AI by ChatGPT.
“AI has made it possible to understand the contribution of data exploitation. The awareness is more tangible. It is easier now to send messages on fundamentals such as governance.”
AI, the driving force behind federated governance
Because without implementing these foundations, the creation of efficient AI models is out of reach, insists the Chief Data Officer of the French insurer. And the executive committees of companies, interested in the potential of generative AI, are sensitive to this discourse and constitute strong sponsors. This support is essential.
Indeed, all the participants underline that data governance is a complex pillar, which requires organization and management over time. AI is a lever for promotion, but not the only one. For Chafika Chettaoui, it was also necessary to rethink governance in order to connect it to use cases and therefore to value.
This trajectory towards “governed decentralization” is the one that it is adopting as part of Axa’s new Data strategy – currently being deployed. In this context, Data Leaders have been appointed, “guarantees of the temple as close as possible to the profession”. They play the role of local Chief Data Officers, ambassadors and relays.
Coming from the profession, to guarantee their legitimacy and understanding, Leaders combine governance and value management. “It is important to have the same person to address both objectives in synergy. In my opinion, this is the best way to embody governance by value.”
A Data Office facilitator and conductor
Club Med is moving towards a comparable model as part of a more global change management policy. To manage both governance and projects – aimed at designing Data Products – the Group Data Office decentralizes through the creation of teams of champions all over the world.
“Our central role will be to help them. Being a facilitator and an accelerator is the main role of the Data Office,” explains Siddhartha Chatterjee.
What about the premises? At Technip Energies, in terms of governance, this is the area that prioritizes the data to be entered in the catalog, as well as the use cases. These priorities are defined according to the specific objectives of the Data Domains and their strategic challenges.
“Which use cases to select, what data to put in quality, priority, etc… It is really up to these areas to choose, 15 in total. Ditto on the financing part. The budget must come directly from the business and the profession,” explains Medhi Benabderrazik.
Governance also designed for AI
This goes for any Data Product, simple dashboard or AI model. All products must therefore be integrated into the Data Catalog deployed within the company. This product approach also requires the dissemination of an adapted culture.
Training and acculturation, for example of Data Leaders and Domain Owners, must be directly involved, ideally with the assistance of human resources. “Without HR, there is no transformation. It is therefore extremely important to get them on board,” notes the CDO of Axa.
This human aspect is just as crucial when it comes to generative AI, if only to underline the criticality of data quality, essential for the creation of efficient models.
“Comex asks a lot of questions about AI Gen. They are overwhelmed with information and need acculturation, particularly on the prerequisites for its deployment. We must first reposition the Data Office and secure Data governance. Only then will it be possible to start implementing generative AI adapted to the business,” concludes Anthony Asso.