LAS VEGAS — T-Mobile has been directing more of its customer service calls to actual human beings. What a concept! And as a consequence, its Net Promoter Scores, which measure the likelihood to recommend a brand, are up to all-time highs.
Although it appears to customers that the mobile phone company is reverting to good ole fashion customer service, in reality, the company is employing some sophisticated machine learning technology on the backend.
Heather Nolis, a machine learning engineer with T-Mobile, said the company is using SageMaker Ground Truth from Amazon Web Services (AWS). She explained that SageMaker is helping T-Mobile to create natural language understanding models that show them relevant, contextual customer information in real-time.
“We should understand why a customer is coming to us,” said Nolis.
To help set the stage for a customer call, SageMaker uses machine learning based on thousands of previous customer calls.
Usually, the first step in creating a machine learning model to assist with customer service conversations is for human annotators to manually review thousands of conversations and add the labels required to train machine learning models. This process is time consuming and expensive. SageMaker Ground Truth speeds up the labeling of data. The first step still uses human annotators, but then SageMaker Ground Truth learns from these annotations in real time and automatically applies labels to much of the remaining dataset.
“Instead of doing labels manually, you have a user interface create labels and then automatically create learnings,” said Nolis.
This helps smooth the calling experience for customers. For example, automated phone systems often annoy people with stupid questions. A person might call the toll-free number to pay a bill. First, they’re asked to dial “2” to pay their bill. Then they’re transferred to another phone tree that asks them to dial “3” if they’re an existing customer. Why would they be calling to pay their bill if they weren’t already an existing customer?
Customers often resort to punching “0” in hopes of talking to a real person, or they shout “customer service representative” into their phones.
AWS’ Sagemaker inserts some machine intelligence into all this. For example, if a customer has recently received a message that their bill is coming due, SageMaker predicts that’s why the customer is calling. They are then connected to a customer service representative who is prepared to talk with them about their bill.
T-Mobile and AWS
T-Mobile and AWS have an ongoing relationship. Vinay Kshirsagar, a director with T-Mobile, said almost all customer-facing T-Mobile apps are hosted in AWS. A few are hosted in Microsoft Azure. “The systems behind, like payment systems, or billing systems, they are part of our data centers,” said Kshirsagar.
He wasn’t willing to discuss any of T-Mobile’s plans in terms of upgrading its network for 5G. But Kshirsagar did say that T-Mobile began its relationship with AWS for its Infrastructure-as-a-Service technology. “Then we went to containers, then we’re getting big on serverless and machine learning.”