Data Mesh: How decentralization democratizes data use across the enterprise

First, a quick recap of the who, what, and why of data mesh: the notion was conceived by Zhamak Dehghani of ThoughtWorks as a way to improve data accessibility and simplify organization in centralized, monolithic data structures. The explosion of data during the COVID-19 pandemic accentuated these needs, and data mesh has now evolved from a concept into a distributed architecture model to make data usable by everyone across an organization.

Johan Picard of Google has a few ideas: “Part of it has to do with simple data maturity. For twenty years, companies have created and centralized their data on platforms, data lakes, data warehouses. This is what made it possible for them to exploit their data. But the domains had to wait in line to access that data, causing backlogs and bottlenecks. 

“Some went so far as to create their own shadow IT departments, because the IT teams understand the technology but not the quality of the data. And the reverse is true for the domain. But now, with the cloud, technologies are evolving and enabling data to be decentralized to the domains. It’s important to remember that this is an inspirational approach. Every company is different and must aim for decentralization in its own way, taking into account its own context, speed and ambitions.”

At Orange, Brice Miranda experienced this phenomenon of data concentration and fragmentation. “Our historical data platform wasn’t self-service oriented, so many of our teams created their own mini data platforms; we ended up with between 40 and 60.” The operator built a platform to serve their domain teams with Google Cloud Platform. “And now, we’re entering into the logic of data democracy”.

Before creating his company at the end of this year, Jean-Christophe Brun was CTO of Carrefour. “In 2015, we knew we needed to transform our vision of data, to center our retail activity around data. Until 2018, we were building an on-premise platform to become more data-centric. And in 2018, we became a Google partner to do more with algorithms. Even before we learned about the notion of data mesh, our objective was to multiply business uses. What I like about data mesh is that it allows us to formalize things we hadn’t standardized, to put words behind the principles.”    

Jean-Noel Lucas at L’Oréal began the same process three years ago. Jean-Paul Agon, then CEO, set the course: to be the champion of beauty tech. To do so, we needed to orchestrate the data, and break the functional silos; identify data stewards, agree on a common language and structure the data, to build a global reference system. And this required resources: “You have to buy time to make the data available. We evaluated the governance debt (two or three years) and estimated the resources needed to manage the company around the data. We dedicated a team of 100 people to prepare the data. We brought new skills into the group. We also do a lot of repeated evangelizing to retain the budget that allows us to make this effort. Because L’Oréaliens are not conceptual. We have to translate the concept into concrete terms”.

L’Oréal lists 50 use cases that tap into product data. From product portfolio management, to distribution improvement, to inventory management. “With a data as a product strategy, data is distributed to the different domains and its preparation is entrusted to each brand’s specialists, and it solves the bottleneck problem”, concludes Jean-Noel.

Even though Orange has a data studio for consulting and Collibra to search the data, it’s not enough: “When we make data deposits, we end up with duplicates and that poses problems. If you ask for customers’ broadband equipment, you get 800 answers”. Is the customer equipped with fixed, mobile, fiber? The tool doesn’t know. Has the customer given his consent? We don’t know that either. There is no single source of truth or data that can be trusted. As a result, POCs are being conducted on these questions (“What is an Orange customer?” “What is permission?”).

Orange is a little ahead of L’Oréal, where 40% of data is available in the cloud. “As a prerequisite, data must be available and discoverable,” says Jean-Noel. “We have to go the last mile in business intelligence.Then, we need to move up the domains in the data value chain, with data consumption as a performance criterion (KPI), to get closer to business use cases. Behind that is the data culture challenge, of getting people to adopt more standardized solutions and see real value.”

“At Carrefour, we’ve chosen to make our data product owners responsible,” explains Jean-Christophe. “They understand the use of data and ensure its communication: ‘Data products must be used. Data product owners must be interested in the use of their products.’ In the past, the business intelligence department reported to the domains. But now, the domains themselves are doing it, with the support of BI. Sixty percent of the people trained by Carrefour in data came from the domains.”

It’s a back-and-forth process, explains Johan Picard, between analytics and operations: “The data must move up into analytics and back down into operations. In the past, databases couldn’t integrate analysis, so we created the data lake. In the new paradigm, we’re closer to the data sources and we have a better understanding of it. As a result, we no longer need to separate operations and analysis. Some of our customers have put them on the same plateau. SAP specialists sit side by side with data specialists. Brice Miranda agrees and adds, “We have to erase the old boundaries between analytics and operations and go to the source.”

For Jean-Noel Lucas, “Operational staff must understand what data they rely on (by thinking about their processes) and strive to improve data quality.” For example, to supply stores and better manage inventory, L’Oréal has added a layer to PCM. “We want to ‘API-ze’ our systems.” L’Oréal has worked extensively on product data in order to make a 360 view available to retailers and consumers. To better manage their supply chain, L’Oréal needs to better segment its catalogs. And the new organization makes it possible to be more reactive, to place production orders for products on a weekly basis, when demand increases.

Carrefour has gone the furthest in converging analytics and operations: in this new world, analytics becomes an real time operational tool, at the same level of service as operations. “The product sheet must record all data, including sales figures. This allows for substitution scores between products, in case a product is missing. This was unthinkable three years ago,” concludes Jean-Christophe.