Taking a build-once approach can extract most value from our data and bring together data scientists and software architects

Mohammad Sharifan, Lead Architect Data and AI Architect, Deutsche Telekom AG, talked about adopting the right AI architecture strategy, outlining the key challenges facing large enterprise network operators.

For large enterprise Telco operators, it was highlighted that data management, privacy, and ownership must be addressed to reduce costs and boost efficiency. “There are many questions that you need to consider when it comes to data. This includes where the customer data is coming from, where it is stored, how to access it, who to give access to, and who is accountable for data in the home,” advised Sharifan.

The importance of ensuring a “build-once approach” for many different applications was highlighted by Sharifan.

“Data scientists may look to use state-of-the-art AI and ML tools to understand the data and solve a business problem. Whereas a software architect may instead consider how we share and access data. Ensuring we deliver quality data at the source is no small feat. We need to ensure a seamless knowledge transfer between subject matter experts and data scientists,” Sharifan advised.

Three pillars of architecture were identified as being key for extracting greater value from data:

• The platform in use (toolstacks and vendor solutions)
• The value, meaning, and ownership of data
• The domain (such as CPE domain or access technology)

Aligning the data platform with the domain and considering where AI will sit within a network is vital, Sharifan warned. Key questions included how to make data closer to the device, what value does AI bring, and how can this be delivered effectively to organizations?

“Telco Operators have mostly a huge legacy landscape. Therefore data accessibility and visibility become a challenge. To run and build AI on top of this big landscape, a proper architecture and collaboration is needed.,” Sharifan continued. “While AI technology is powerful, it does pose a risk without the right regulation of the network to make sure it is accessible by the right person, and data must be monitored at every layer.”

Sharifan outlined that different use cases have special architectural requirements. From real-time service delivery to network planning and analytics, each have varying needs. Sharifan said building the right architecture with AI and ML tools built-in is critical. ML frameworks should be run on the most appropriate platform for greater automation and help Proof of Concepts transition to real world deployments, Sharifan advised.

Building out an AI prototype is the least costly step of productizing, said Sharifan. It can take weeks to design it and months to launch in a production environment. “ML experts often lack the operational experience. So, it is critical that operational experts and data scientists work together to lay the foundations for our networks of tomorrow.”