Login
Register
Dawn Bushaus
Orange creates an internal marketplace for data products
As part of an initiative to democratize data and modernize data architecture, Orange is developing an internal marketplace for data products on Google Cloud Platform (GCP). Frederic Reboulleau, VP of Data and Artificial Intelligence, and Olivier Simon, VP of Smart Networks and Data, spoke recently with Inform about Orange’s data modernization efforts.
Orange has been working with Google on data transformation since 2020, and in April 2024 the companies announced an expansion of their collaboration to deploy AI, including GenAI, closer to Orange’s operations and those of its customers by working with Google and other partners such as Open AI and Anthropic. Included in these efforts is a data democratization program that has been underway for about three years.
“Democratizing data means making data accessible to our employees, regardless of their technical expertise,” says Reboulleau. “It means enabling informed decision-making across the organization, breaking silos and making sure that we have transparency for fostering a culture where the data is treated as a shared resource. So, data democratization is really key and fundamental from Orange’s perspective.”
Responsibility for data strategy and architecture is distributed between the group’s CTO office and another division focusing specifically on data and AI, but these teams work closely with each other. Network and operational support system (OSS) data is overseen by group CTO Laurent Leboucher, while overall data and AI transformation is the remit of Chief AI Officer Steve Jarrett. Groupwide data and AI initiatives are often co-sponsored by both groups.
Democratizing data is particularly challenging for Orange because the company has operations in 26 countries across Europe, Africa and the Middle East. This means navigating multiple sovereignty and privacy laws.
“In France, you cannot export your data outside of the country, or at least some parts of the data like topology and CDRs [call detail records],” Simon explains. “You have the same in Poland [and] in Senegal and other countries. So, the way we manage that is to create a single platform definition – single code – but deployed with instances in different countries to manage the regulation.”
Network complexity is also a big challenge. “In our network domain we were facing many silos as well as silos inside the countries because Orange is quite complex,” Reboulleau explains. “Even inside countries, there are lot of silos, so building data governance and data products was really fundamental from our perspective.”
Much of the complexity stems from the fact that the network has many sub-domains, Simon adds. For example, the teams managing the radio access network (RAN), transport network and core network all have their own data, data models and processes.
“It used to be like this for very good reason. I mean, it was just impossible to manage the whole thing without creating borders,” says Simon. “But now, if you want to leverage multi-domain processes that are super important for monitoring, then we need to break these silos.”
He adds: “The network domain is probably the hardest that we have because of this complexity of sub-silos and also because the volume of data is huge. Consider that in Orange we generate around 1 petabyte of data every day… So, it’s a huge work, and I think we will continue [with data democratization] for probably three or four years before it’s business as usual.”
The process of democratizing data in Orange started with the creation of data products and enforcing business ownership of those products, which includes responsibility for maintaining the quality of the data products. To do this, Orange developed an internal marketplace for data products on GCP and created a KPI for the number of data products available.
Each division of the company reports its number of data products quarterly. As of the end of Q1, 33 data products had been created, with a target for 2025 of 143.
“This KPI shows the level of maturity for data democracy,” says Reboulleau. “Then, we moved more strategically to track where products have the most value creation.”
That second KPI for measuring value is helping to solve the challenge of convincing business stakeholders of the need to create and own data products, Reboulleau adds. Potential consumers of the data products can also see a quality score for each.
“The data and AI team were convinced that we need to have data products. The main challenge was to get all the stakeholders online – stakeholders like the CFO and business owners – because we need to prove the value creation, the return on investments, to them,” Reboulleau explains.
Indeed, as is the case for most communications service providers (CSPs), the cultural change required to modernize data architecture and create a data marketplace has been tougher than technological change at Orange.
“The need for a modern data architecture is well understood now. However, the need for a good data catalog, marketplace, etc., is generally well understood by IT people but not always by data owners who must allocate resources to manage data quality and have to manage this priority among others,” Simon explains.
An important goal of Orange’s efforts to democratize data is to support use cases that rely on AI to improve customer experience and cut costs. This includes, for example, reducing the amount of time that service is unavailable, cutting the time it takes to fix faults, and optimizing the network.
“All our use cases were passing through value creation,” says Reboulleau. “So, we were very selective and really focused…on the big use cases.”
On the network side, the team started with smart capex. “A rule of thumb for the value that we can get with AI is to first target the most demanding areas in terms of capacity needs,” Simon explains. “And as you can imagine, network capex could be multiple-billions every year for a group like Orange.”
Smart capex brings value by enabling more educated decisions about where to put new sites and add fiber capacity. “Instead of having a general knowledge, you have very detailed knowledge cell by cell, address by address,” Simon explains, adding that this can result in millions of euros in savings.
Another valuable use case that relies on AI and data democratization is field operations, which represents a significant cost annually for a large operator like Orange. One example where improvements can be made is in validating fiber deployment and repairs. Field reps are instructed to take photos when they complete a job, which results in a large amount of unstructured data. This data can be managed and analyzed using AI to control quality.
Finally, Orange has targeted the network operations center (NOC) as a high priority use case for AI, where it can be used to analyze the thousands of alarms that are generated during network monitoring. Using AI helps with incident detection, root-cause analysis, remediation and routing to the right teams for intervention.
“The main tool that we are using now is AI, because we used to do all these jobs with static rules, a lot of tests and so on… But we kind of reached the limit and it was difficult to go further in both efficiency and quality of service,” says Simon.
