Telenor lays the foundations of a modern data architecture

Jawad Saleemi, Director of AI & Cloud at Telenor, is a co-leader of TM Forum’s Modern Data Architecture Project. He spoke with Inform about the group’s focus and Telenor’s approach to modernizing its data architecture. With operations across the Nordic region and Asia, the company is focusing on data as it implements an AI First program. Our conversation has been edited for length and clarity.

Q: In my reporting on telco data architecture modernization, I’ve learned that developing a data architecture is not quite the same thing as democratizing data but that democratization can be a driver of modernization. Is that how you view it?

A: Modernizing data architecture – data mesh, data fabric, you name it – these are just the ways to achieve something, and one of the goals is democratization of data. What it actually means is that you can make data easy to access, discover and utilize regardless of where in the organization you are. Because for a long time, data has been sitting in closed systems and silos… And with this ever-increasing production of data – and a lot of new potential usage of data, especially with AI and advanced analytics and BI [business intelligence] – there’s a growing need for it to be easy to connect data from all these different previously siloed verticals.

Q: How big of a driver is the move toward an AI-driven, intent-based networks?

A: It’s a big driver, and I would say that it’s a prerequisite. Our traditional, old-school data architectures were built around centralized data, data warehousing and centralized teams working with data – and only centralized people with competence to look at data and understand it. So, we’re moving away from that towards more distributed, always accessible, always available data and data products. The modern data architecture is the platform to enable it.

Q: Where is Telenor on this journey?

A: I work in the Telenor group, and Telenor is quite big. We have a lot of different business units with presence in the Nordics and in Asia. From our group perspective, what we are trying to achieve is how to find synergies and some homogeneous methodologies and ways to work with data to achieve the North Star ambition that we have. We are running the AI First program in Telenor, and the ambition is to enable AI and use AI as a horizontal layer, enabling basically every aspect of what we do. To achieve that, data is a very, very strong prerequisite. We’re working on high-level design of our data strategy to enable data which is AI friendly, AI enabled and AI ready. Data comes first. There is not much the algorithms can do if they don't have data.

We are working on a strategic level to enable this kind of data mindset, because it’s not just technical; it’s also cultural. It’s how you organize teams. It’s your structures within your organization to enable this kind of data ownership and clear ambitions with data. That’s where data mesh and data fabric and other methodologies and other architectures come in.

Q: Are you moving to a data lakehouse, which combines the analytics and storage. Is that a first step, or is that a later step? And how does it relate to data mesh and data fabric?

A: If you think about data warehousing, it was more concentrated around structured data. When you’re storing data, you need to know what kind of structure you’re storing it in, and that limits a lot of possibilities because you have to clean the data with the ETL [extract, transform, load] pipelines and then process that for more curated data. But with data lakes, you can basically just dump in anything you would like: unstructured data, images, text data, service documents, whatever. There was a gap between these two approaches because in the data lake world, you could put all this data in but we didn’t have the guarantees that the data warehousing provides in terms of the consistency of the data… A data lakehouse combines the capabilities of both.

Last year, we made a strategic partnership with Google. So, GCP [Google Cloud Platform] is our analytics platform of choice. But Telenor is quite big, so we also have other hyperscalers as well. But the overall principles are the same, and that is to enable us to get the maximum value out of data by following these overarching principles: data lakehouse, distributed data ownership and bringing software engineering principles and practices into data. You might have heard about DevOps and DevSecOps? Now, we’re moving towards DataSecOps, where we can bring the software engineering principles to data and data transformations.

Q: How does this relate to the data mesh concept?

A: Data mesh advocates for distributed data ownership and federated governance with a strong central infrastructure, and then working with data as a product. But if you go into the very theoretical details, it can get a bit complex. So, from our perspective we look at it as something where we can get some inspiration where it makes sense for us, but we don’t get too deep down into implementing it on all levels.

For example, we have a lot of focus on data as a product. When we work with data, we want to make sure that the data that is produced is trustworthy, scalable, reproduceable and modular, which means that you can pick one data set and then you can combine it with something else to produce a new data set, which can then be combined with something else and in a more meshed way. But most importantly, every data product has a clear owner. If you are looking at some data set and let’s say you have some issues with quality or interpretation, you know exactly which team is responsible for that data product, and it comes with some SLAs [service level agreements].

Q: Are you starting small with specific use cases? I know that Telefonica, for example, is starting with operational support systems (OSS), particularly inventory and assurance?

A: Absolutely, not particularly just on OSS or inventory for us, but we are starting small with potential data products which are tangible and which add business value. Because that’s also one of the requirements of data products: They have to have some value for them to exist... And that value is business value. It has to answer some business question.

Q: What does this mean for governance?

A: Again, we have quite different architectures and ecosystems in Telenor because it’s quite big, so where we are moving more towards a data mesh kind of ecosystem, we have federated governance. That means we have value streams, and the representatives from these value streams work with a centralized governance team, which is very lean. Together they come up with the overall governing principles to set guidelines, best practices and the ‘golden path’ – the overall strategy of how to work with data. But we are also in an early phase, so I wouldn't claim that we have done it. We are learning, but it’s important to start somewhere – start small and then learn and build upon it.

Q: Does the primary value of this transformation lie in increasing automation and cutting costs, or is it in increasing revenue because you can deliver new products? I guess it’s likely both?

A: It’s both, and there are a lot of domains, right? There’s the customer domain, the network domain and the people domain – our own employees. There is a lot of potential in improving and enabling all these domains with better data. You can improve customer satisfaction and engagement, and that will, in turn, increase revenue and loyalty. So, all these things are connected, and it’s the same on the network side. If you have better automation, you have reduced opex and better services in terms of flexibility, coverage and utilization. That then gets reflected in better customer service, which can bring more revenue. So, I think all this is a part of a bigger ecosystem.

A: How closely tied are your efforts around modernizing data architecture to autonomous networks?

A: Autonomous networks depend on AI, and then AI has dependency on data. To be able to produce the data which the AI algorithms can use, we need to enable or initiate these kinds of initiatives like modern data architecture. It’s absolutely foundational.

Q: What are the biggest challenges you’re facing in modernizing data architecture?

A: It has two parts: one is technology, and the other one is people. And the technology is the easy part. We know how it works; we can make it work the way we want. But how do you structure the organization? How do you associate proper ownership to the relevant teams? How do you structure the domains, and what kind of governance model do you have? These are the challenging parts, especially the governance. If you go with completely distributed governance with no central or semi-central governance, then it will result in total anarchy.

Q: What’s the role for standards?

A: To be able to connect data, the systems have to be interoperable. And to be interoperable, they need to have some open standards that they should follow: a common language, common format, common ontology, and common platforms or a common marketplace where they can be discovered and connected. Otherwise, we’re back to the siloed approach. TM Forum’s Modern Data Architecture group is working on setting the foundations of the modern data architecture, including some of the elements like data products and data contracts, and data platforms and capabilities.

Once we have good building blocks – if we make good Legos – then you can use them to build something. As long as we do our job well building these building blocks that are standardized, interoperable, have proper documentation and have proper specs, then the consumers using them can build upon them... Our group is mostly focusing on the foundations of the data architecture but not so much on the utilization. Other groups within TM Forum, like the Autonomous Networks Project, can then use the foundation.