Comcast turns up AI and ML for network insights and to improve customer experience

Comcast is using artificial intelligence and machine learning to proactively find problems before customers call in to customer service. (Pixabay)

Comcast is tapping into artificial intelligence (AI) and machine learning (ML) to gain valuable insights across its networks, and to provide a better customer experience. It has also come up with an internal program to make sense of all of the buckets of data that its ML and AI systems are gathering.

RELATED: At SCTE-ISBE Cable-Tec Expo, Comcast's Werner and Zelesko talk virtualization, culture and the edge

Previously, Comcast's Tony Werner and Matt Zelesko spoke to FierceTelecom about how company's virtualization efforts were based on the deployment of its next-gen X1 video platform. Comcast's use of AI and ML were also founded on the back of X1.

With X1, Comcast was able to establish complete telemetry and visibility into the platform, establish incremental roll outs and roll back to previous versions if there's a problem.

"We're now seeing that carry into the network as well, and so things that we've learned in X1 are now kind of permeating everything," said Comcast CTO Zelesko.  "I think AI and ML is one part of that and it really helps us with a lot of the telemetry and visibility.

"So it's being able to detect things before customers tell us. The cool thing about AI is you just take a whole bunch of data, throw it into the soup pot, and AI comes out with the insights and the connections."

Werner, president, technology, product, Xperience organization at Comcast Cable, said Comcast took a lot of the same algorithms and systems that the company built for the X1 voice controlled remote and applied them to other areas.

"The good news is we've got a large group of PhDs that work on this for us," Werner said. "The same systems that we built for our voice remote, actually we were able to apply to several other things. So it's powering our AI bot right now, our CX (customer experience) layer that we put through everything, which is actually answering a lot of customers issues with the AI engine.

"Above and beyond that, we are applying it to several things in the network today so that we can quickly diagnose problems and start to even correct them before they become customer-facing."

While the promise of AI and ML have been long-standing, there needs to be coherent approach to all of the data that is generated. Werner said a group of Comcast engineers came up with a program they named "CXels" to address gleaning the most relevant data.

"The metaphor is just the same way that there's millions of pixels on the screen to make up a picture, there's millions of CXels in each area to make up the customer experience," Werner explained. "What we've done is we've divided our customer experience from the network side into eight different groups.

As examples, Werner said one group could be all of the DOCSIS services on the network while another group would be WiFi. The product owners in each of the eight different groups are responsible for curating the important information that comes out of signals and probes.

"So the team that owns the WiFi will listen to say thousands of points of data, but these are the three things that really determine what's important," Werner said. "And then each of those is supplemented with an AI engine that looks across this and says 'Out of all this data, I can tell you when these two things go wrong, the customer's not going to be happy.' Then you take these eight groups together, and they bubble up to an AI engine that goes across all eight of them that says 'Here's what this customer experience has been,' and then that feeds into our internal tools."

Comcast correlates that information with explicit indications from customers, such as customer calls after they've ran speed tests.

"Chances are if they're running a speed test, they're not happy with the networking performance they're getting," Zelesko said. "So we correlate all that internal stuff with those external indicators to really determine what has the best correlation and what makes sense.

"We're finding things that we never would have found before through the application of machine learning and AI. I think it's making us a lot more proactive."