Elad Nafshi, EVP and CNO at Comcast tells Contributing Editor Annie Turner how the operator built the US’ smartest network with intelligence deployed at its edge, within feet of its customers’ premises, and where it’s going next.

Elad Nafshi is Executive Vice President and Chief Network Officer at Comcast, a role he’s held for almost five years, although he been at the company for more than 20. He is participating in the Keynote CxO Panel at FutureNet World North America which will discuss Building the AI-Native Operator: How Telcos Remain Relevant Beyond Connectivity? You could say this describes his mission over the last few years.

Nafshi explains that COVID helped push Comcast towards building its virtual, digital network platform, based on real-time telemetry because everyone suddenly staying at home changed network traffic patterns and drove a massive rise in traffic levels. The first few weeks of the pandemic created an increase in demand for capacity equivalent to what Comcast would expect in three years – and the operator had to find ways to operationalise it.

Technicians couldn’t attend Comcast’s facilities or people’s homes to turn up capacity, “So we took a proof of concept for a platform, which became Octave. It recognises different conditions and demands in different premises using billions of data telemetry points, multiple times a day, as the basis of real-time decisions,” Nafshi continues. “The aim was to find the most efficient way of delivering the capacity each customer needed. It’s transparent to the customer, improves the reliability and resiliency of the service, and increases network capacity by 35%.”

The rise of AIOps

The next challenge was building the logic behind those billions of data points that the network generates and collects daily. “This is where our AIOps platform started to come together,” Nafshi says. It learned to identify certain failures’ patterns and Comcast deployed automation to enable self-healing wherever possible. Today roughly 77% of platform alarms are completely self-healed via this AIOps platform. He adds, “Every day that gets better and even higher because as the analysis becomes more sophisticated, we’re able to cover more use cases”.

Even so, with a physical network comprising more than 1 million miles of fibre and cable, across 39 different states and to 65 million premises, “There’s a lot of opportunities for fibre to be cut,” he observes. To address this, Comcast built a device called XMF, which monitors over 1 million kilometres of fibre across the entire network using AI to detect any interruption. It doesn’t have to be a cut – other issues impair services intermittently, and can be identified too – such as squirrel chews (see Building the right models below).

Faults are detected “within seconds and pin-point accuracy on the map,” according to Nafshi. “We can then recommend to our technicians what do they need at that location – and that’s how AIOps expands and adds use cases,” he states.

Power in the hands of customers

Accurate power monitoring has become another game changer. He says, “We built logic – an algorithm called GOLD (Global Outage, Localized Detection) – to look at battery backup in locations that have suffered a power cut to see if it’s draining and how fast.” The batteries can last up to four hours, but the speed of decay depends on how many affected homes have their own emergency power supply, and how well the power grid matches Comcast’s network topology. Accurately monitoring the power situation means Comcast knows not to send out technicians unnecessarily to faults caused by power failures.

Further, Comcast puts “the same wisdom…we expose to our divisions into the hands of customers” with real-time updates in their app which differentiates Comcast from its competitors, Nafshi says. The operator also has a network agent that can talk to customers, answering queries about disrupted services, based on real-time visibility of what’s going on in the network. Alternatively, the agent can be proactive, warning customers before a problem manifests.

He continues, “The trick we’re figuring out here is how to build more deterministic agents from the wisdom of our incredible experts who are training the models, relative to what it means on our network. Then we look if we have the digital topology, digital infrastructure, to turn all that into action.”

Building the right models

These differentiating capabilities rely on having useable data in the right place, “100%,” he agrees, “but it’s also about being able to build the models – it’s not a case of asking ChatGPT. You need to build unique logic for these types of things, and the data we train them on, to build AI operation models that are unique and relevant to us.”

Comcast does not build its own models for generic things such as general networking information, Nafshi clarifies, “but we create them for stuff you can’t find on the larger models”. An example is training a model to identify specifically when a cable has been chewed by a squirrel, as mentioned.

“We have AI capabilities deeper than any other telecom company through our DOCSIS 4.0 FDX investment…which we leverage to deploy multi-gig symmetrical services,” he says. “With it, we’re deploying neural processing units (NPUs) in our devices that sit feet away from your home. Those AI chipsets constantly scan the quality of your service…so I know that Annie is on this node, on leg number two, three amplifiers down the cascade, and that 100 feet to the right is where that impairment is.

“We have millions of customers benefiting from that level of reliability, driven by our AIOps platforms…AI enables us to offer a step function improvement in our overall service reliability to the point this is our competitive advantage against fibre. An all-fibre network is passive. You don’t know where the issues are and you don’t know what the issues are versus having that 24×7 spectrum analyser sitting literally within feet of your home, constantly monitoring your experience in real time.”

Improving network reliability

Nothing daunted, Nafshi is determined to improve the resilience and reliability of fibre too. In September, Comcast began rolling out AI-based sensing technology that can recognise events like someone climbing an antenna tower or a train passing close by through the vibrations they cause on the fibre. It flags up unexpected activity, giving Comcast the option of dispatching a crew in case anything goes wrong.

The operator is also pushing intelligence into the home. Its award-winning Xfinity Wi-Fi 7 gateway devices optimise Wi-Fi’s performance on the fly, from the basement to the attic and the porch.

Nafshi says his focus is to lean ever deeper into the tech, leveraging more advanced models, more self-healing use cases and mitigating the effects of physical events, investing in “the absolute best, most reliable, fastest connectivity platform in the world with the lowest lag. This is very important for the most advanced services.”

Enabling advanced services

 Live sport is such a service. People watch sport on many platforms and it generates the highest traffic levels, peaking with Amazon Prime’s NFL Thursday Night Football*. YouTube TV, Peacock, Netflix and other streaming services also provide sports’ coverage, all of which has created new usage patterns on the network.

Comcast began to develop then deploy the Virtualized Cable Mode Termination System (vCMTS) to replace the physical CMTS back in 2018, adding edge compute no more than 5 milliseconds of network delivery time away from customers’ homes.

At the end of September, Comcast announced it is working with software house Fastly and NBCUniversal to leverage that vCMTS and deliver live sports plus other high-demand events on the Peacock streaming service. “It enables us to provide the highest quality of delivery because we don’t need to traverse the World Wide Web to get there,” Nafshi says. “It goes straight to that edge compute with the lowest lag, without all those wasted hops”.

Cybersecurity is another advanced service. In June Comcast launched a nationwide security product for small businesses by embedding AI technology on its edge compute platform. It spots data patterns that indicate cyberattacks, malicious data and viruses. He says, “Customers don’t have to download or configure anything. We already have many thousands of customers on this [service] and in the last 30 days, we protected them 3.2 billion times from attacks leveraging that edge compute platform.”

The power of the platform

“You are seeing how the network becomes a platform,” Nafshi sums up. “It enables connectivity but so much more. The agentic side creates a whole new generation of experiences for customers connected to our network, like live sport and cyber protection, but we’ve got lots more coming”, especially in B2B and B2B2C, as demonstrated by the partnership with Peacock.

“We can apply the same logic to a gaming company,” he says, noting that Grand Theft Auto VI is due on 19 November and expected to create extremely high traffic levels.

Nafshi concludes, “We’ve made tremendous progress, but we have so much more to do and so many opportunities for our customers, I’m super excited about what’s coming. Recently we showed a network expert agent at the SCTE TechExpo 26 , trained by our brightest and best…and are already 90% accurate using that analysis, even with the size of our network. Give me a few more months and we’re going to be as close to 100% as you could possibly get.”

* Stands for National Football League, meaning American football, not soccer