SEATTLE—Comcast implemented a machine learning program that the company said can accurately predict whether it needs to dispatch a technician to a customer’s home in order to fix connectivity problems. The company developed the program internally, in part by using open-source software.
Adam Hertz, Comcast’s VP of engineering in the company’s Silicon Valley Innovation Center, said the company’s prediction program could save Comcast “tens of millions of dollars” by predicting whether the company needs to send a technician to a customer’s home in order to fix a problem. Hertz explained that some connectivity problems occur inside customer’s homes, while others occur at some point in Comcast’s network, and the company’s machine learning application can predict with more than 90% accuracy whether a technician will be able to fix the problem by driving to a customer’s home.
He said that such “truck rolls,” in cable industry parlance, are an expensive proposition, so Comcast invested in the machine learning application in order to prevent them where they are unnecessary.
Hertz made his comments here at the Mobile Future Forward event held by Chetan Sharma Consulting. He said Comcast’s machine learning application is one example of how the company is working to make its services more efficient through the use of advanced software and analytics techniques. He said the app was recently put into use by Comcast.
In order to develop the application, Hertz said Comcast first assembled a wide range of data sets across its operations, including data from calls to its customer service center and data from its network operations. The company then built a model geared toward using all the data that could accurately predict whether a truck roll would be necessary.
“We run all this data through this model, through this machine learning application,” Hertz said.
Once the machine learning application reached the point where it could predict such situations with at least 90% accuracy, the company then put the application into operation. Now when a customer calls about connectivity problems, customer service representatives check the application to determine whether a truck roll is necessary.
Comcast isn’t alone, of course, in using advanced software techniques to make its network and its operations more efficient. A wide range of other telecommunications companies are engaging in similar efforts in to reduce the costs in their services and their networks. AT&T, for example, created an internal big-data division that works to leverage data on the company’s operations to find better ways of doing things. One of the division’s findings was savings of $119 million last year through turning off power to millions of pieces of unused or unneeded telecom equipment throughout the country.
Machine learning is one of several software applications the global tech industry is developing. Machine learning has been described as techniques that give computers the ability to learn without being explicitly programmed, generally via the study and construction of algorithms that can learn from and make predictions on data. Other similar software services include deep learning and artificial intelligence.
Article updated Sept. 12 to clarify the status of Comcast's application.