As the largest cable TV company and largest home Internet service provider in the United States, Comcast boasted a customer base of over 22.5 million TV subscribers and 25.1 million broadband subscriptions in 2017 (Kafka). With this enormous scale comes vast amounts of data—data that Comcast is only recently leveraging with machine learning algorithms to both launch new products as well as to dramatically improve customer experience. While large tech companies and specialized startups may be more well-known for their machine learning efforts, these innovations can be even more critical for incumbents like Comcast. Though live TV remains the primary way consumers view video content, the threat of “cord cutting” looms large as people shift from cable to online streaming options like Netflix (Fortune). Successfully navigating this changing landscape will be key for Comcast to stay relevant.
Historically, one of Comcast’s biggest fundamental challenges has been customer service. The cable industry as a whole is admittedly notorious in that regard, but Comcast may represent the worst of the worst—it recently ranked dead last in a study of NPS scores of 300 companies across 20 industries (Temkin). Accordingly, machine learning represents an unparalleled opportunity to improve both accurate detection and resolution of customer issues while simultaneously reducing the cost to serve (H2O). Comcast has launched a program that can predict with more than 90% accuracy whether a technician should be dispatched to a customer’s home in order to fix connectivity problems. Because some issues occur inside homes while others occur in Comcast’s network, the machine learning algorithm uses past data to determine what solution path will most likely resolve the problem and is expected to help the company save millions of dollars in costs (FierceVideo). In the future, Comcast will also deploy self-healing networks which use data to detect and remediate failures such that network issues can be resolved without any need for human intervention at all, and oftentimes before they actually impact performance (H2O).
In addition to improving customer service, Comcast is developing a suite of new products to personalize user experience. By combining historical data with real-time streaming elements, Comcast can reliably predict the popularity of a particular TV show or film 24 hours in advance, providing personalized recommendations to different viewing audiences (H2O). Furthermore, using natural language processing, the Comcast Labs team recently developed a voice remote that allows consumers to navigate and select shows using live speech (H2O). Similar to other voice command devices like Google Home and Alexa, the voice remote faces the familiar challenges of interpreting a users’ requests, many of which can be variable in form (e.g., questions around what to watch vs. direct commands) or may have nothing to do with TV programming at all (e.g., checking the weather) (Rao et. al).
Over the next 5-10 years, the vision for Comcast is to use machine learning to position itself as not just a cable company, but a credible player in the smart home space. With its strong foothold in internet, entertainment, and security services, Comcast could leverage its data to cross-train these products to better understand how each individual customer lives. Already, in its emerging security business, the company has developed an Xfinity Home Camera that features smart thumbnails that automatically zooms into moving objects rather than showing a static frame, providing users useful live feeds (Hall). When combined with inputs from its voice remote, the data from its cameras could prove a powerful way to understand customer habits and behavior, providing the backbone of its holistic smart home vision.
Overall, it seems Comcast is well-positioned to improve its existing services using machine learning. I believe the company should focus on first building a product that is backed by outstanding customer service in order to re-gain the trust of consumers. The data suggests that Comcast is well-positioned to improve customer satisfaction using machine learning, but its newer offerings leave me with questions: Does Comcast’s entrenched position give it an advantage when it comes to releasing new home-related products, or is its image inextricably tied to cable? Although Comcast has the advantage of being installed already in 20 million homes, will it be able to compete in the smart home space against more sophisticated competitors like Google and Amazon? It may be an uphill battle, but Comcast’s substantial investments in AI and machine learning in recent years indicate that the cable behemoth is willing to take on these challenges in hopes of retaining its position as the largest media/telecom company in the U.S. (748 words)
“Comcast’s Machine Learning App Could save ‘tens of Millions’ of Dollars in Truck Rolls.” FierceVideo. September 11, 2017. Accessed November 14, 2018. https://www.fiercevideo.com/cable/comcast-s-machine-learning-app-saves-tens-millions-dollars-truck-rolls.
“Estimates of Cord Cutting Are Exploding.” Fortune. Accessed November 11, 2018. http://fortune.com/2018/07/24/cord-cutting-comcast-netflix/.
Hall, Tonya. “How Deep Learning and Artificial Intelligence Power Comcast’s Voice Remote.” TechRepublic. Accessed November 14, 2018. https://www.techrepublic.com/article/how-deep-learning-and-artificial-intelligence-power-comcasts-voice-remote/.
Kafka, Peter, and Rani Molla. “Comcast, the Largest Broadband Company in the U.S., Is Getting Even Bigger.” Recode. April 27, 2017. Accessed November 14, 2018. https://www.recode.net/2017/4/27/15413870/comcast-broadband-internet-pay-tv-subscribers-q1-2017.
“Net Promoter Score Benchmark Study, 2017.” Temkin Group. Accessed November 14, 2018. https://temkingroup.com/product/net-promoter-score-benchmark-study-2017/.
“Operationalizing Machine Learning at Comcast.” H2O. https://www.h2o.ai/wp-content/uploads/2017/03/Case-Studies_Comcast.pdf.
Rao, Jinfeng, Ferhan Ture, and Jimmy Lin. “Multi-Task Learning with Neural Networks for Voice Query Understanding on an Entertainment Platform.” Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining – KDD 18, 2018. doi:10.1145/3219819.3219870.