To generate «business» use cases from data, several strategies exist. We talk about Data Mesh, Data Driven or else of Data-centric. But what approach to aim for? If the approach Data-centric remains the most relevant, it is entirely complementary to the approach Data Mesh. Joint interview with two data experts at JEMS.
Hello gentlemen, can you introduce yourselves?
Pascal DURY: I am Vice President of Data Management and Governance. I support our clients with everything related to establishing data governance organisations.
Anthony LIbor Hello, I am Vice President for AI and Analytics offerings and I help our clients with their IT and Data transformation, to serve their new data-driven use cases.
Could you remind us of the difference between Data Mesh and Data-centric ?
Anthony : The approach Data Mesh, it's a breakdown of a data platform by business domain. It's very useful in large organisations where an iterative approach is necessary, and where one is driven by business use cases. The Data Mesh facilitates the distribution of responsibilities as, when faced with a data domain, it is possible to define owners and processes. The approach Data-centric comes in addition.
Pascal: To build on what Anthony is saying, Data Mesh, takes up the design elements Domain-Driven Design. This is not new; it's a rebranding of things that have existed for a few years already, namely identifying data and organising it according to a domain-driven vision, where each domain is responsible for its scope (HR, marketing, production, etc.).
What is an approach Data-centric ?
Pascal: The approach Data-centric considers that data is not just a consumable, but a company asset. It considers that each company possesses a reusable data asset that is independent of the uses for which it is intended. It aims to standardise information so that it can serve all types of uses, including data, dashboards, performance, and predictive use with AI, etc.
Is it necessary to have an approach Data Mesh then Data-centric Or the reverse?
Pascal : We are complementary. Let me explain: in JEMS terminology, to be Data Driven, it's about making data-driven decisions; I take data at its source, apply it to my use case, and consume it. And I'd do the same thing again tomorrow for another use case. That's good but inefficient in the long term because data is seen as a consumable rather than a key company asset. If you adopt an approach Data Mesh without any approach Data-centric, the limitations will be the same.
Therefore, one must be Data-centric then Data Mesh and avoid a consumerist approach as much as possible Data Driven.
We create the data asset which will then be consumed in the use cases, more or less organised by domain. If your data asset is huge, having a single team to manage it becomes very difficult and we will lose agility. We will therefore distribute the asset across the different domains. We will have objects that are cross-cutting between one or two domains, or even across all domains, and others that are mono-domain: this is the approach Data Mesh.
Anthony : I don't entirely agree. When building a data platform, you are necessarily driven by use cases from the business departments, mainly for funding reasons. You will organise the data in the usual way, in layers, following an organisation by business domain. So it's similar to the approach Data Mesh !
On the other hand, this investment effort needs to be made right away to conceive of elements that will have maximum reusability, and that can themselves be enriched by other fields. And there, yes, I agree with you: we need to have a D approachData Centric From the outset, it's the best possible strategy.
In terms of cost, what is the best approach?
Pascal: So, there will be an extra charge for switching from Data Mesh, then Data-centric. A surcharge which is the same as when you go from Data Driven Oh Data-centric, it's the same thing.
We have examples of companies that have taken this approach and realised that, in the long term, the approach Data Driven didn't suit and who have taken up their entire data architecture again. I'm thinking of Renault, who rebuilt their entire data architecture using an approach of Data-centric from which they build their Data Products.
What final piece of advice could you give?
Pascal The Data Mesh [FR]: exige de poser les bases solides d’une gouvernance de la donnée [EN-GB]: requires laying the solid foundations for data governance. Saying that professions take responsibility for organising themselves as they see fit is planting the seeds of anarchy within the data system. In 3 years, we'll go back to where we started because it doesn't work.
The Data Mesh, This is the application of federated governance. Federated governance means that at the central level, a framework is defined, the rules of the game: what can be done and everything that is not allowed. Once defined, domains use these rules and apply those that concern them, adapting some to their specificities, and we have something that will work. It's fundamental.
Anthony: The evolutionary trend of IS, with the emergence of Microservices architectures, is to adopt a modular and elemental approach to the components we implement to serve typical uses such as AI or digital applications that will consume data from data platforms.
My advice is to adopt a marketplace vision for your data assets, where you will build modular data stores that will serve these uses.
But to have the best Time to market, this work of capitalisation and reusability of the asset must be carried out by identifying the right company objects.
« The The product has a limited lifespan. Your asset data does not, it is immortal.»
