Video: How to Design and Build Natural Conversations

Learn more about conversational IVR at the next SpeechTEK conference.

Read the complete transcript of this clip:

Nico Acosta: The basic building block of a conversational experience are tasks. And tasks represent the end goal of what the user wants to do. These are not intense. These are the end goal of what the user wants to do. So get car insurance. Cancel policy. Get claim status. That is the end-to-end experience of what you want to do.

Then you can train those tasks, and you can program them. You train them so the machine learning knows how to trigger and invoke a task. And you program them to design what experience you want once the task is invoked. So when you say, "Hey, I'm reaching out because I need car insurance for my new Lamborghini," Autopilot does the speech recognition, does the intent classification, and does the entity extraction.

This is where the machine learning comes in. And it knows that that sentence means get car insurance, that the make of the car is Lamborghini, and that it's a new one. And then, once you land on "get car insurance," you can decide what that experience looks like. And it can look very different depending on the task. A task can be a data collection workflow. Where you say, "Okay, I want to collect the make, the model, the year, and a bunch of like personal information" to give you the quote. It can be a system-of-record question. Like, "Hey, what's the status of my claim?"

That answer is in your database. You can go over Wikipedia and you will not find the answer. And you need to have a system of record that says, "Hey, Claim 123 is spending payment." And that's where APIs are so powerful. Where you can integrate with your systems of record, and answer those type of questions. The other thing you can do is answer knowledge questions. Of like, "Can I insure luxury cars?" Or, "Can I insure multiple drivers on the same policy?" Those are knowledge questions.

Or you can hand off to an agent if it's the case of reporting an accident, for example. Or a combination of these. And you build the tasks with seven simple, conversational actions that are fully programmable. Where you can say things, you can wait for input, you can ask questions, you can hand off to human agents. You can remember information to build context out of conversation, and you can even show images in devices that have visual displays.

SpeechTek Covers
for qualified subscribers
Subscribe Now Current Issue Past Issues
Related Articles

Video: Three Megatrends in the Speech Technology Market

Twilio Director of AI/ML Nico Acosta explores the market's three prevailing megatrends--cloud, omnichannel, and intelligence--and their impact on the customer journey and the cost of doing business in this clip from his keynote at SpeechTEK 2019.

Video: How to Circumvent the Limitations of Traditional Conversational IVR

Twilio Director of AI/ML Nico Acosta discusses the advantages of AI-driven conversational IVR platforms like Twilio Autopilot over old-school 0-9 systems for improved CX and business efficiency in this clip from his keynote at SpeechTEK 2019.

Video: How to Assess if a Conversational UI Is Right for You

Allstate Conversational Designer Katie Lower outlines working models for assessing the viability of a conversational interface with multiple teams within an organization in this clip from her presentation at SpeechTEK 2019.