In Context

The next-generation communications blog from Nexmo

< Back

Bots and AI: Great Bot Use Cases in Production Today

September 7, 2017 Published by

Every bot developer must contend with the limitations that are inherent in such a relatively new technology. But innovators always find a way to transcend tech limitations and build something remarkable. I asked Microsoft Technical Evangelist Martin Beeby and CTO/Co-Founder of Opearlo Oscar Merry to share examples of great use cases they’ve encountered or built for voice and text bots.

What Are Examples of Good Bot Use Cases Today? (Full Transcript)

Sam Machin (Nexmo Developer Advocate & Alexa Champion): What do we think is a great use case where we are today? Given the state of the tech and what the limitations are. What have people seen that’s a really nice use case for bots, both voice and text today?

Oscar Merry (Co-Founder & CTO at Opearlo): I think from the voice space, I think one trend that we’ve definitely identified is around content. So, complex applications that have a lot of different functionality on maybe mobile or web, probably aren’t best suited to transition to a voice app right now. However, if you’ve got a lot of content and if you want to build a voice app as a content delivery channel, we find that really works well. Because often people are using their Amazon Echo or Google Home from the comfort of their own home, maybe when they’re relaxing on the sofa and chilled out. And so if you’ve got a large amount of content that you want to get out to your users and it’s something that your users could enjoy in the comfort of their own home, then a voice app could actually be a really good way of getting that content out there.

if you want to build a voice app as a content delivery channel, we find that really works well

Sam: Yeah, and it’s a fairly simple interaction, isn’t it? You just got to play, pause, next-type commands.

Oscar: Exactly. Which makes it a lot easier as you’re navigating things like the voice interaction model design. And especially if you’re a big brand or a big organization looking to dip your toe in the water in the bot space, building a content app is a lot easier than integrating with your existing web or mobile product.

Sam: So not necessarily entertainment content, but maybe more informational content, things like recipes, that kind of thing? Is that something you think that’s…

Oscar: Exactly, yeah. Exactly.

Martin Beeby (Technical Evangelist at Microsoft): Yeah, we’ve worked with a number of different companies all trying to achieve sort of different sorts of things. And one of the bots that I worked on last November was with Sage. And they have these huge accountancy packages like Sage One and Sage Line 500 and so forth. And they’re using the bot to kind of interface or be the interface across all of their different product lines as a sort of, a simple-to-use API or system. And one of the things that their users really like is this ability to quickly check on things, like who owes them money, what their expenses are, or adding an expense very quickly. So it’s a number of very specific commands that you can give the Pegg bot, the Sage Pegg bot. But they’re sort of these snippets of information and data which actually are quite difficult to get from other systems.

It’s quite cumbersome to go into Sage just to find out who owes you money. But being able to ask this—your sort of personal assistant—this Pegg bot for Sage like, “Who owes me money?” and it just tells you how much people owe you and so forth, very quickly and getting that information quickly. I still sort of think about it as a way of discovery of company-wide information.

So if you can ask very specific questions and then the bot somehow surfaces that information from your company, then that’s a really helpful kind of bot. And I think it’s a really great interface for those sorts of snippets of information from your organization, helping organizations understand themselves.

Sam: Give me a specific…I was assuming a bit like the weather forecast model, isn’t it? What is the weather like? It’s that there’s a very specific question.

Martin: I mean, that’s a very consumer-focused kind of use case, weather and Obama’s birthday and the things you used, but things like which person owes me the most money this month? What’s my account and what’s the accountancy of my company look like at the moment? Has Martin Beeby looked at that document that I sent over to him yesterday? Who have I been dealing with mostly in email this month? Those sorts of organizational questions are where we’re seeing lots of people use bots, trying to get data and information and insight from their businesses into a bot.

Sam: You think that actually that maybe it’ll end up asking for output rather than input. So, it’s a case of you ask it a fairly specific question and it provides you that information rather than you capturing, inputting data into a system or controlling a system. It’s just about querying a system right now?

Martin: Most of the bots that I have used are being or built have been providing either information or accepting small amounts of information. One of the challenges in collecting information from bots is being able to make sure that you’ve got the right piece of information. The work that we did with Sage on getting expenses, for example, it was quite difficult when a user would say, “I’ve got an expense for food. It’s £500.”

When they say that, they might say in lots of different ways and being able to extract the £500 and extract the category was really tricky. And it didn’t matter how much sort of artificial intelligence training we did, we still would get these edge cases where people would use the system in very different ways.

And so when you get really complicated inputs, multi-variable inputs, then it actually becomes very difficult to manage currently with bot technology. So I think that tends to mean that people are using bots more for information and they’re asking their bot to provide them with information. And that’s where we’re seeing the sweet spot from an enterprise point of view is that people are able to use these bots to extract information from their businesses.

that’s where we’re seeing the sweet spot from an enterprise point of view is that people are able to use these bots to extract information from their businesses.

Oscar: I think just building on that as well, one of the challenges that we see is whatever the information that the user wants to extract from that business, there’s an exercise to make that content kind of bot ready or voice ready. Because you might have your existing APIs that you use to query your accountancy data, for example, but the way that data is returned it might not necessarily fit for a voice response through Alexa or Google Home.

So there’s definitely an exercise in all the projects we do of making the content voice-ready so that it makes sense to hear it back from Alexa. Because one of the things is that she does actually speak very, very slowly. And if you have a long response, your users are just gonna get quite bored and drop off and not come back.

Sam: Yeah, we’ve all seen at hackathons and things when they query the data or something and it reads out to ridiculous a number of decimal points or something. It’s like, “the temperature is 72.493761812 degrees.”

And, yeah, it’s that kind of stuff, that sanitizing and making it… I guess, it’s natural language. It’s not natural, it’s processing. It’s a form of taking structured data and building natural language. But I’ve often found it gets repetitive. If it always tells you exactly the same thing, you want a few mixes in there. You don’t want it to always read you the same thing with the same inflection. You wanted to be able to just mix it up a bit to keep it kind of feeling more human, doesn’t it?

Oscar: Yeah, and I think another great way to do that is to add contextual information from a different source as well. So something as simple as pulling your data from maybe a weather API or something about the local area and varying the content based on that each time, it just adds a bit variability. It makes it more as if it was a human telling you that they’re able to add that additional context. And that’s a great way to build more personality into the experience.

[Editor’s Note: Watch the full one-hour discussion on the state of AI bot technology.]

Categorised in: , , ,

This post was written by