Speech Technology Magazine


How Machine Learning and AI Can Enhance Virtual Agent Training (Video)

USAA's Brett Knight discusses how USAA has used machine learning and AI in virtual agent training at the enterpise level in this clip from SpeechTEK 2018.
By The Editors of Speech Technology - Posted Aug 10, 2018
Page1 of 1
Bookmark and Share

Learn more about virtual agents at the next SpeechTEK conference.

Read the complete transcript of this clip:

Brett Knight: Regarding AI and machine learning, artificial intelligence , I’ll share my experience from an enterprise perspective on what we've found and what has worked. With recognition and understanding, machine learning has really helped us out there.

There are some great tools and features out there to help with building up a corpus of data without a whole lot of user input, not a lot of us scratching our heads and just saying to people, "Hey give us some words to use." We can bring data in from chat and we have some tools that will actually build our training corpus for us. What are all the different ways you could ask, "What's my checking account balance?" We have data for that, and we have tools to process that data and build our corpus.

For simple responses, there's some machine learning that can help with that. It relies heavily on you having data sitting around that you're ready to put in front of your customer, or your member. If you do have that, that's awesome--maybe there are some machine learning techniques that can help you build those simple responses, not dialogue, but here's a question and here's your answer.

I'm actually more excited about the next two items, and I think they’ve shown more benefit to incorporating some advanced AI and machine learning. Analytics is not exactly machine learning, but when we're looking at more automation and ways of looking at how our conversations are performing. Instead of our data scientist folks writing a bunch of rules, they've come up with some really neat ways to view our performance and our data in some new ways without a lot of manual process.

Again, there’s more work to be done there. But the user interface is probably not as intuitive. When you have frequently asked questions or hints when you pop up a virtual agent, being more relevant and timely on the suggestions to help members phrase that question. Sometimes, they just come in, they need help, but they don't know how to express it. So we've looked at the question of if we even need hints. There are those users that are kind of like deer in the headlights, where they don't really know what to type and he hints/FAQs.

Because you have data coming in about what those questions are and how you're answering them, some great machine learning opportunities arise. And even the positioning of your virtual agent factors in--where is it, when does it come up, when does it engage? Again, you've got data and you build into that feedback loop to help influence how you get better at positioning your virtual agent. You should even look at what order the hints come in. Don't just expect AI and machine learning to answer all your questions for you. Look at the engagement and keeping that engagement happening.

Page1 of 1