Artificial Intelligence

How to use machine learning to improve customer service

At the 2017 Google Cloud Next conference, the company hosted a session on how companies can use their machine learning tools on Google Cloud Platform to streamline their customer service efforts.

cservice.jpg
Image: iStockphoto/Jirsak

One area of business that stands to benefit from machine learning is customer service. Technologies like natural language processing and sentiment analysis can help companies better understand how to respond to customer feedback and questions.

At the 2017 Google Cloud Next conference in San Francisco, Apoorv Saxena, a product manager on the Google Cloud team, hosted a breakout session explaining how organizations can use Google products to improve their customer service. He began by outlining the two different approaches that companies can take to machine learning.

The first approach, Saxena said, is to use pre-trained models in the form of APIs to handle the machine learning needs of the company. For example, Google offers APIs such as its Cloud Vision API, Cloud Speech API, Cloud Jobs API, Cloud Translation API, Cloud Video Intelligence API, and the Cloud Natural Language API.

SEE: Why machine learning and data analysis are critical to Google's success in the cloud

The two most relevant APIs for customer service are the Cloud Speech API and the Natural Language API.

Cloud Language can transcribe text in over 80 languages and detect inappropriate content. Users can then analyze the text output with other machine learning APIs from Google. In customer service, one would use this to transcribe audio from customer service calls or voicemails.

The Natural Language API lets you extract entities from text, such as a person, place, or thing. It then gives extra metadata, pointing the user to a Wikipedia page for that entity, if one exists. It can also analyze sentiment (from customer feedback, for example), or analyze syntax so the user knows how the words depend on each other.

However, if a company needs a more unique solution, they would have to build a custom system using TensorFlow and the Cloud Machine Learning engine. They would start by training the system with examples of content, including classifiers. Then, they would be able to serve it in production.

As a case study, Saxena brought up Dan Nelson, the head of data for online supermarket Ocado, to explain how his company has implemented machine learning. Nelson said that Ocado gets roughly 2,000 emails into its contact centers a day, ranging from refund requests, to general feedback, to website trouble, and more.

Starting with the Natural Language API, Nelson said they were able to label message with tags such as "Feedback" or "Positive." However, they then built a custom solution with Cloud Machine Learning and TensorFlow to get more detailed filtering. It sits outside of their storage layer and allows the company to more effectively triage their customer service requests, Nelson said.

As a result, Nelson said that Ocado was able to respond to urgent emails four times faster, and saved money on headcount in the contact center. Nelson recommended that companies investigating machine learning would define their success criteria early and be careful to set control groups and perform a lot of testing.

The 3 big takeaways for TechRepublic readers

  1. Google has been increasingly investing in machine learning, opening up new tools for customers to use to improve efforts in customer service.
  2. Google customers can approach machine learning with pre-built APIs, or use TensorFlow and the Cloud Machine Learning platform to build a custom solution.
  3. Customers should define their criteria early and test often when working with machine learning.

Also see

About Conner Forrest

Conner Forrest is News Editor for TechRepublic. He covers enterprise technology and is interested in the convergence of tech and culture.

Editor's Picks