 Live from San Francisco. It's theCUBE. Covering Google Cloud Next 2018. Brought to you by Google Cloud and its ecosystem partners. Hello everyone, welcome back. This is theCUBE live in San Francisco for Google Cloud big event here called Google Next 2018. Hashtag Google Next 18. I'm John Furrier, Dave Vellante. Bringing down all the top stories, all the top technology news, all the stuff that they're announcing on stage. Coming through the executives, the product managers, customers, analysts, you name it, we want to get that signal and extract it and share that with you. Our next guest is Dan Herron, he's the product manager for Cloud AI at Google and Dialogflow with a hot product here under his purview. Thanks for joining us. Good to see you. Yeah, excited to be here. Yeah, we're just bantering off camera because we love video, we love speech to text, we love all kinds of automation that can add value to someone's products rather than having to do a lot of grunt work or not having any capability. So we're super excited about what you're working on and the variety of things. This one's the biggest. Dialogflow, talk about the product. Sure, yeah. Yeah, so Dialogflow, it's a platform for building conversational applications, conversational interfaces. So it could be chat bots, it could be voice bots. And it started from the acquisition of API AI that we did a year and a half ago. And it's been gaining a lot of momentum since then. So last year at Google Cloud Next, we announced that we just crossed 150,000 developers in the Dialogflow community. Yesterday we just announced that we now cross 600,000. Oh, I'm back up, slow down, I think I just missed that. You had what and then turned into what? Say it again. So it was 150,000 last year or over 150 and now it's over 600,000. Gradually, that's massive, that's traction. It's very, very exciting. Forex. And yeah, we're still seeing a lot of strong growth and with the new announcements we made yesterday, we think it's going to take a much larger role, especially in larger enterprises and especially in sort of powering enterprise contact centers. Natural language processing, also known as NLP for the folks that know the jargon or don't know the jargon. It's been around for a long time. There's been a series of open source, academia has done it. Just never, ontology has been around. It's just never cracked the code. Nothing has actually blown me away over the years until cloud came. So with cloud, you're seeing a rebirth of NLP because now you have scale, you got compute power, more access to data. This is a real big deal. Can you just talk about the importance of why cloud and NLP and other things that were, I won't say stunted or hit a glass ceiling in big capability. Why is cloud so important? Because you're seeing a surge of new services. Yeah, yeah, sure. So there's two big things. One is cloud, the other is machine learning and AI. And they kind of advanced speech recognition, natural language understanding, speech synthesis, all of the big technologies that we're working on. So with cloud, there's now sort of a lot more processing that's done centrally and there's more availability of data that you could use to train models and that feeds well into machine learning. And so with machine learning, we can do stuff that was much harder to do before machine learning existed. And with some of these new tools, like what makes Dialogflow special is you could use it to build stuff very, very easily. So I showed last year at Google Cloud Next how you build a bot for an imaginary Google hardware store. We built the whole thing in 15 minutes and deployed it on a messaging platform and it was done. And so it's so quick and easy, anyone can do it now. So we're going to do an ask the cube bot, take our transcriptions, have canned answers, maybe down the road, you automate it away. Yeah, yeah, yeah. They're going to kill our job. Yeah. No, it's pretty awesome. What's interesting is it's shifting the focus from kind of developers and IT to more to the business users. So what we're seeing is a lot of our customers, one of the people that went on stage yesterday in the Dialogflow session, they were saying that now 90% of the work is actually done by the business users that are programming the tool. Really, so it's a low code type of environment? Yeah, you can build simple things without coding. Now, if you're a large enterprise, you're probably going to need to have a fulfillment layer that has code, but it's somewhat abstracted from the NLU piece. And so you can do a lot of things directly on the UI without any code. So I can get started as a business user, develop some function, get used to it, and then learn over time and add more value, and then bring in my real hardcore devs when I really want some new functions to come. Right, right. So, yeah, what it handles is understanding what the user wants. So if you're building a cube bot and it needs, what Dialogflow will do is to help you understand what the user is saying to the cube bot. And then what you need to bring in a developer for is to then fulfill it. So if you want to, for example, every time they ask for cube merchandise, you want to send them a shirt or a toy or something, you want your developer to connect it to your warehouse or whatever. Well, they could say, give us the best blockchain content you have. Right. There it is. Yeah, so how would we go about that? We have all this corpus of data that we ingest and we would just, what would we do with that? Take us through an example. So you would want to identify what are the really important use cases that you want to fulfill. You don't want to do everything. Right. You know, you're going to spread yourself and it won't be high quality. You want to pick what are the 20% of things that drive 80% of the traffic and then fulfill those. And then for the rest, you probably want to just transition to a human and have it handled by human. So let's say for us, we want it to be topical, right? So would we somehow go through and auto-categorize the data and pick the top topics and say, okay, now we want to chat bot to be able to ask questions about the most relevant content in these five areas, 10 areas or whatever. Is that a, would that be a reasonable use case that you could actually tackle? Yeah, definitely. You know, there's a lot of tools some Google offer, some that others offer that can do that kind of categorization, but you would want to kind of figure out what are the important use cases that you want to fulfill and then sort of build paths around them. Okay, and then you've got ML behind this and this is a function. I can, this fits into your serverless strategy or you just announced GA today. So, right? Yeah, no. So we announced GA a few months ago. Yeah, okay, great. What we announced yesterday was five new features that help transform Dialogflow into sort of from a tool. Those features, take a minute. Sure, yeah, yeah. So, first is our Dialogflow phone gateway. What it does, it can turn any bot into an IVR that can respond within, it takes 30 seconds to set up. You basically just choose a phone number and it attaches a phone number and it costs $0 per month, zero nothing. You just, you pay for usage if it actually goes above a certain limit. And then it does all the speech recognition, speech synthesis, natural language understanding, orchestration, it does it all for you. So, you know, setting up an IVR, a few years ago it used to be something that, you know, you needed millions of dollars to set up. Science project, yeah, absolutely. Now you can do it in a few minutes. Second is our knowledge connectors. What it does, it lets you incorporate enterprise knowledge into your chat bot. It could either be FAQs or articles. And so now, if you have some sort of FAQ, again in like less than a minute, you can build it into Dialogflow without having to build intents for it. Then there are a few other smaller ones we introduced also our speech synthesis, automatic spell correction, which is really important for chat bots because people always have typos. I'm guilty just as much as everyone. And last but not least, sentiment analysis. So when it helps you understand when you want to transition to a human. For example, if you have someone sort of that's not super happy. Agents. Yeah, exactly. Exactly. And some of these capabilities were available separately. So for example, you could have built a phone gateway and connected it to Dialogflow before, but it used to be a big project that took a lot of work. So we had a guest speaker yesterday in the session for Dialogflow. And they've been running a POC with a few vendors right now. It's been going on for a few months and they told us that with Dialogflow, phone gateway and knowledge connectors, they were able to build something in a few hours that took a few months to do with other vendors because they had to stitch together multiple services, configure them, set them up, do all of that. So we use case for this just to kind of, first of all, chat bots have been hot for a while, super great. But now you have an integrated complex system behind it powering a elegant front end. I can see this as a great bolt on to products, whether it's websites or apps, how-tos, instrumentation, education, a lot of different apps. That seems to be the use case. How does someone learn more about it? How do they get involved? They go to the website, they download some code, just take us through, I want to jump in tomorrow or now. What do I do? It's a free edition, I can add to that. Exactly, yeah. So yeah, so the good news is, yeah, you could go to either cloud.google.com slash Dialogflow or Dialogflow.com. There's, if you go to Dialogflow.com, you can sign up for the standard edition, which is 100% free. It's for text interactions, it's unlimited up to a small amount of traffic. And you can even play around with a phone gateway and knowledge connectors with a limited amount without even giving a credit card. If you want cloud terms of service and enterprise grade reliability, we also offer Dialogflow Enterprise Edition, which is available on cloud.google.com. And you can sign up there. That comes with an SLA that. Exactly, an SLA and like cloud data terms of service and everything that's kind of attached with that. I'd also encourage people to check out the YouTube clip for the session that was yesterday, that was where we demoed all these new features. Sorry. What was the name of the session? Automating your contact center with virtual agents. Okay, check that on YouTube, good session. Okay, so take us through the roadmap. You're on the product, so you're a product manager, so this is, you got to decide priorities. Probably to maybe cut some things, make things work better. What's on the roadmap? What's the guiding principle? What's the North Star for this product? Yeah, so for us, it's all about the quality of the end user experience. So the reality is, there's many thousands of bots out there in the world, and most of them are not great. Most of them are not great. Most of them really suck. Yeah, if you Google for why chat bots, why chat bots fail is the first result, right? And so that's kind of our North Star. We want to solve that. We want to help different developers, whether they're startups, whether they're enterprises, we want to help them build high quality bots. And so a lot of the features we announced yesterday are kind of parts of that journey. For example, integrated sentiment analysis that lets you transition to humans, because we know we can't solve everything, so it helps you understand. Or knowledge. Automation helps to a certain point, but humans are really important at a crossover point. Trying to understand that's important. Exactly, and we'd rather help people build bots that are focused on specific use cases, but do them really, really well, versus do a lot, but leave users with a feeling like they're talking to a bot that doesn't understand them, and they have a bad experience. We take all the questions we've done on theCUBE, Dave, and put them into a chat bot. Yeah, exactly. What's the future of bots? Yeah. Go ahead, answer the question. So, I think, so we're kind of in the last year or two, we've been at an inflection point where speech recognition has advanced dramatically. And it's now good enough that it can understand really complex questions. So you can see with Google Assistant and Google Home and a bunch of other things that people can now converse with bots and get reasonably good answers back. And that just feeds ML in a big way. Right, exactly. And so now, you know, Dialogflow introduced speech recognition in November. We just introduced speech synthesis yesterday. And so we're now, you know, we're looking to empower all of our developers to build these amazing voice-based experiences with Dialogflow. Give an anecdote or an experience that the customers had where you guys are like, wow, that blew me away. That is so cool, or that was just so technically amazing, or that was unique, and we've never seen that coming. Give us some, share some color commentary around some of the implementations of the bot world and the Dialogflow's impact to someone's business or life. Yeah, sure. So I think, so yesterday, the Ticketmaster team was showing how, you know, how they look at their current IVR that's kind of based in the old world where you have to give very short responses like yes or no, or like San Francisco, California. And because it's built on these short responses, it's kind of a guided IVR, it takes 11 steps. Which is an IVR again? Oh, integrated voice response. Okay, got it. Or interactive voice response. Yeah, one of the two. It's a system that answers the phone. I don't think you should get the jargon right. Okay. So now with something like Dialogflow, they can go and build something like that. Instead of 11 steps, it takes three steps. So because someone can just say, you know, I'd like to buy tickets for so-and-so and complete the sentence. And the cool thing is sort of the example that they gave was a recording that I made with them about a year plus ago. And the example was, you know, I'd like to book tickets for Chainsmokers. And then they were showing it yesterday in the conference. And they were like, oh, we know why you chose it. It's because the Chainsmokers are performing at Google Cloud Next. It's like, yeah, it's probably just a funny coincidence. So they've deployed this now or they're in the process of deploying it? They're in the process of deploying it, first for customer service and at a later stage, it's going to be for sales as well. Yeah, because the IVR for Ticketmaster today, I know it well, I'm a customer, I love Ticketmaster, but you're right. It tells you what you just asked them pretty well. But it really doesn't quite solve your problem. Yeah, I mean, they recognize the sales one was built a long time ago, but they're kind of overhauling all of that. But I'm excited to see it because there's a good point of comparison. You know, you're always a good reference point that you understand. And the takeaway that I'm getting, Dan, is the advice you're giving is nail the use case, narrow it down, and then start there. Don't try to do too wide of a scope. Exactly, exactly. Handle, the most important thing is delivering great end user experiences because you want people to really enjoy talking to the bot, right? So in surveys, people say 60% of consumers say that the thing they want to improve most in customer service is getting more self-serve tools. They're not looking to talk to humans, but they're forced to because the self-serve tools. They suck so bad. Yeah, they're terrible. And it's terrible. And if I can get it quickly self-serve, I'd love that every time. I serve myself gas and a variety of other things. Airport kiosks have gotten so much better. You know, I don't mind those anymore. Okay, one quick follow up on Dave's point about making it focus. I totally agree, that's a great point. Is there a recommendation on how the data should be structured on the ingest side? What's the training data? Is there a certain best practice that you recommend on having a certain kinds of data? Is it Q and A? Is it just text? Speaks is why, is it a blob of data that gets parsed by the engine? I mean, take us through what's on the data piece. Yeah, so that really changes a lot. So depending on the specific use case, specific company, specific customer, so someone asked in the audience yesterday, asked the guest speakers how many intents they've built in Dialogflow. And each one of them had like very different answers, right? So it depends a lot. But I would say the goal is to kind of focus on the top use cases that really matter, build high quality conversations and then build a lot of intents and text examples in those. And when I say a lot, it doesn't actually, we don't need a lot because Dialogflow is built on machine learning. So sometimes a few dozen is enough, or maybe like a couple hundred if you need to. But we see people trying tens of thousands or like we don't need that much data. And then for the other stuff that's not in your core use cases, that's where you can use things like knowledge connectors or other ways to respond to people rather than sort of manually building them or just divert them to human associates that can fill those questions. Great job, Dan. So you're the lead product manager on? I'm the lead product manager at Dialogflow Enterprise Edition. And there's a large team kind of working with me on the part. We don't talk about that. What are the products you own? Oh, I'm also product manager for Cloud Speech to Text and Cloud Text to Speech. Awesome. Well, great to have you on. Thanks for sharing. Super exciting. Love the focus. I think it's a great strategy of having something that's not a one-trick pony bot kind of model. Having something that's more comprehensive, seeing that's why bots fail. But I think there's a real need for great self-service. It's the Google way. Get your search result, get out quick. Get your results. I mean, this is a Google ethos. Get in, get your answer. Yeah, we're all about democratizing AI. So now with Cloud Speech to Text, Cloud Text to Speech, it put the power of Google speech recognition, speech synthesis in the hands of any developer. Now with Dialogflow, we're taking that a step further. And anyone can build their voice box with ease, what used to cost like millions of dollars or need special expertise. All right, Dan Herron, who's the product manager for the Dialogflow Enterprise Edition and doing Cloud AI for Google is to bring you all the best dialogue here in theCUBE. Doing our part. Soon we'll have a CUBE bot. You can ask us any question and we'll have a canned answer from one of the CUBE interviews. I'm Dave Vellante. He's here with me, I'm John Furrier. Thanks for watching. Stay with us. We'll be right back.