Operationalizing machine learning at a telco provider

"We have hard problems to solve," said the provider's director of engineering analysis. "H2O.ai machine learning is one of the cooler things that we are working with in order to help us solve problems in a way that's really valuable for the organization."

The provider applied H2O across several use cases. To cut avoidable truck rolls, the customer-care team built a model that flags issues fixable without a technician visit, such as resetting a modem or changing a setting. Because most truck rolls are unavoidable, the team used H2O's balanced-class option and subsampling to keep the model from skewing to the majority case. The video research team used gradient-boosted decision trees to predict the popularity of a show or film 24 hours ahead, clustering users and surfacing what is trending inside each cluster in real time.

To measure customer experience across its entire footprint, the provider used H2O clustering to correlate many data sources into customer-experience groups, guiding where to add network capacity. A self-healing system feeds real-time data into a machine-learning rules engine that scores incoming events, diagnoses problems, and suggests fixes automatically. H2O made these models straightforward to put into production: the team exports them as Java code or runs H2O instances as web services.

The provider moved machine learning from analysis into live operations. Avoidable-truck-roll models reduce unnecessary technician visits, gradient-boosted models predict trending content 24 hours before it airs, clustering produces a customer-experience metric that spans the whole footprint, and the self-healing rules engine resolves system errors without a support call.

"In the future, we will see more decisions leaning on this technology," said the director, "and not just executive decisions. We are seeing more and more intelligent systems and products that leverage this technology to do things like self-healing, add better features to the product, and predict what will be trending."