The Future of Voice [Entire Talk]

Who you are defines how you build. What I'm going to do today is I'm going to walk you through my journey in building. First off, maybe a minority of the time, I'll talk about building BlueKai, the acquisition by Oracle and how that went. Then I'll talk about building Voysea and the acquisition by Cisco, and then I'll talk about my five rules of thumb for building a business and for building a team and we'll have a conversation. Here's the goal. Don't wait till the end to ask questions. If you got questions, ask them. If they fluster me, I'll tell you, I'll answer them later. So don't worry. If I make it through this whole thing without a question, I would have failed. Don't let me fail. So I will start now by talking about kind of what was really interesting about building Blue Kai. Blue Kai taught me one rule of starting companies that I have. I don't know if it's 100% true, we could have this debate, but my belief is that if you're going to start a B2B company, you need to be an expert in the area you're going to disrupt. When I say expert, I mean to say that if you go out to a handful of CEOs in that industry, they know you. They know you by name. Because when you're disrupting the B2B industry, the way decisions are made are quite complex and they're unknown to you from the outside. Everything seems simpler from the outside. And if your competitor is started by a CEO who comes from in the industry, who, who has the Rolodex of the top CEOs who can make decisions much faster, who can learn much faster because they don't make all the stupid mistakes outsiders do, they will beat you. B2C companies are different because everybody's a consumer and everybody can self introspect and figure out things on their own and you're usually creating these new categories that didn't exist and there is no expert who's more expert than you. So that's my idea of B2B. I could be wrong. Glad to debate it. When I started BlueKai, I was in the ad tech industry. I knew a huge number of the buyers in the industry, the ad agencies, the publishers. I knew them well, they knew me. We were at conferences together, on panels together, so on. And one insight I had back then was that people, this was back in 2007 and there was this huge growing digital ad business where dollars were converting from the kind of TV offline world onto digital very fast. And they were all being placed according to contextual rules. So if you're IBM, you bought the Wall Street Journal's technology page to put an article because the technology buyers read the technology section of the Wall Street Journal. It was all place based. And we had this idea that, hey, instead of serving ads to a page, why don't we serve it to a person based on the knowledge of the person. That was insight number one. Insight number two was there's 300 ad networks on all these publishers, all competing to kill each other in this ad ecosystem. I didn't believe I could build another ad company that would be better than Google or like at that time, Yahoo and Facebook. And I didn't want to build a company that would become number four. Who the heck wants to do that? I wanted to be number one. So I said, why don't I build a company that would supply the data for every ad company out there and supply the data technology and never compete in the ecosystem with the ad players. It's a really good decision because what we did is we started out, got together a really good team and we went up. Within the first few months we signed up eBay, Expedia, Cars.com to supply us their entire real time data live. And then I built an auction marketplace. And think about these advertisers. At that time, everybody was advertising demographically. And you go up to them and say, hey, I have real time data from ebay, Expedia, every search, every, every click, you can bid on it in real time. The second they leave Expedia and ebay, you're going to be able to show them an American Airlines ad to someone who just looked for a trip to Hawaii. The ad buyers loved it, all the portals loved it. And we just started gaining liquidity very fast. And if you remember about 2007, 2008, 2009, the web started to become very sticky. You'd see a product ad and you'd see it. I really apologize for that. So that kind of changed happened. So that was we landed our commerce buyers, we then landed our strategic anchor tenants who are buying the ads. We started to get a lot of liquidity and then the whole programmatic ad space came in. Does anybody here understand the ad ecosystem or am I speaking just really weird? Not much. Okay, cool. Anyway, so we started growing really quick and then a really interesting thing happened. Several years into building BlueKai, we were growing really fast. Ebay gives me a call and basically says, hey, we've decided we don't want our data in an open marketplace. We're just looking for tech to manage our own data. Could you license the tech? We're going to pull out. It was a really sad day for me, I thought because they were such a dominant supplier of data for us and we were the only source for that data. I'm like oh my God, this is going to kill my marketplace. So I made a decision at that moment to never be over concentrated on demand or supply by more than 10%. And that thinking allowed us to actually say great, let's just license our technology for people where we do no buying or selling of data. We just give them all the tech they need to run better data driven marketing. That forced us to open up that model we did and that thing just started taking off. Within a couple years, a dominant number of the top 30 brands started selecting us. This is people like Microsoft, it was some of the key banks. Hewlett Packard, a good part of the top 30 companies in the US selected us for this private pay us license fees to manage their own data. And once they were in there they realized they needed more data so they went and they bought it. So it build synergy to the other business and it took off. We finally hit profitability in year six and for some reason I really can't tell you why. Within a two week period, three CEOs of public companies booked time with me. Cryptic, not explaining why. I show up at lunch and dinner and they want to acquire us. It was just odd. Talked to the board, they said yeah, that sounds pretty good if there's three at the same time, let's at least listen to them. And a few weeks later they acquired us. So Oracle became the winning bidder. They acquired us for 420 million. This was back in beginning of 2014. So interesting thing happen. A lot of times you go into a company and you get acquired and they ask you, hey, are you the golfing CEO that you could just hang out and play golf and rest and vest they call it, or what type are you? I'm like no, no, no, I only have an on button. If you want me to play golf, let me go. You want me to do something, let me do something. So that was at the very beginning of the process. I met Larry Ellison the day after the deal closed and he said, look, I want you to build up this business into a really big business and I want you to think beyond advertising and I want you to tell me right now, if you could buy some companies, what would you buy? I'm like, you mean now? He said, yeah. I said, can I have until next Monday? So I came back with a map and we just started buying companies and we grew organically and through acquisition and we hit 500 million in revenue really fast. So it was just a really fun journey. It was just nice to be able to bounce ideas off of Larry Ellson and see his pocketbook versus the pocketbook I had access to. So that was the journey at Oracle. I stayed there two and a half years, continued to build the company. By the time I Left, I had four CEOs reporting to me on my executive team. That's kind of interesting dynamic to try to run a team with that. Many people are used to running their own show. Some magical reason it worked and we had fun and we built it. And I left. And one of the CEOs of the company that I acquired took over from there. All right, now I'm going to move over to Voicia. I left Voica, took a little bit time, a few months, and I recruited a gentleman who at the time was running the applied AI at Facebook. He reported to Yann Lecun. Some people here know Yann Lecun. Ness. Yeah, one of the. He won the Turing Award. So just a great person in deep learning. He reported him and build Deep Text, the system that understands all your posts in Facebook. And because of him, I was able to recruit a bunch of other people. I pulled people, a guy who was running moonshots for LinkedIn. I pulled people out of Microsoft, out of Apple, and that's expensive, by the way. And so got all these ML engineers. We went off to build what I'm about to talk about. And I realized I needed funding. And so I was lucky to have Salesforce Invest. Actually, right when I was starting the company, before we even launched, I got this email from Marc Benioff, who I didn't know, and I looked at it. I'm like, oh, this is a friend of mine pretending to be him. It can't be real. I looked at it and it was real. And I met him. And so Salesforce invested. And then after that, a whole bunch of other people. Google jumped in, Cisco jumped in, Microsoft jumped in. So let me tell you what we did. So. So first I'll ask you a question. Back in 2018, in the summer, there was a baby who said their first words, and it wasn't mom or dad. Anyone want to guess what it was? Alexa. Yeah, Alexa. Old news by now. And that's Joe Brady. And there's this new generation of People who have an expectation that they can go and have a conversation with the device. And this is just taking off. It's just really interesting how it's going to transform what you expect in, in dealing with devices. And we had this thesis that there's going to be a voice first moment. Just like back in 2012, Zuckerberg had his mobile first moment saying, stop designing for the web, start first for the mobile and think of the web second. We thought that there was going to be a moment in the enterprise. This is what's happening in the consumer space, in the enterprise space. There will be a moment when you start designing interfaces for a voice conversation. And the expectation will be that the designers understand you, not you understand the interface. So you have to get into natural language and make it easy for people to interact. So that was really what we were building. We wanted to build the voice first world for a more productive work environment. We looked at meetings first, because if you look at meetings, it's like 20 to 30% of your time, and if you're an executive, it's like 70, 70% of your time. And billions of dollars are spent keeping people in meetings and you repeat the same things over and over. And what we wanted to do is just say we're going to make them productive and help you turn talk into action. You can walk into a meeting, you can make a few notes, say this is the action we agreed on, this is the decision we agreed on and it would be carried forward, shared with people and no longer be ephemeral. So. So I'll show it to you what it looked like. So I'll look into my system here. And we built an enterprise voice assistant called eva. EVA can basically read my calendar and just show up in a meeting if there's any sort of detail. If it's a dial in, or if it's a webex link or zoom link or teams meeting, it'll just show up, announce itself, show itself on the screen as this little animated character that everybody keeps confusing with Disney's Eva. That was the biggest question on our customer support line. Did you get permission from Disney? Disney's isn't Eva. I think it's Evo or something like that. In any case, so it just appears and it gives you closed captioning. So it's transcribing in real time what you're saying and it's trying to identify notes. Let me show you what happens after a meeting. So when a meeting's done, you get something that you can navigate, you see the concepts Right away. That was discussed in the meeting. You can go straight to it. Do you know what's so interesting to me is. So it immediately takes Lee to the point that anything was talked about or the multiple points it was talked about. It's in lockstep with video, so you can see what was on the screen at the time. It's diarized, meaning it's identified who the speaker is and it's tagged it by that speaker. I'll digress for a second. This is one of the most interesting design problems because in people's mind, they speak like Martin Luther King, like they're giving speeches. Actually, most normal meeting speech is more like Anthony Scaramucci. You know who that is? Yeah, it totally sucks. And when you see it written as a document, just as plain text, full screen, it looks terrible. It looks unprofessional. I've been on calls with, like, really good execs who are using the tool and they're like, I couldn't have said that. And you click to the audio and, yeah, they're repeating themselves, contradict themselves, bad grammar, the whole thing. So we had to figure out a way how to show this. And we took a playbook from Facebook, which is they realized that when people were doing comments, if they were able to put bubbles on it and make them think that this is casual, like text, people would do more of it and they wouldn't overthink things. And they got a lot more usage that way. So that's why we kind of designed this UI that's, you know, easy to navigate. The text is like split up into speakers and so on to make it easy. But let me show a few other things so you can navigate it. You see the speaker, you get speaker insights. This example, Amy and I at Cisco had a conversation with the Wall Street Journal. This was a journalist. It shows you speak time. So in this case, it was Angus at the Journal. He spoke the most. And I can tell you, if you're in PR and you're talking to a journalist, you want to get the journalist to talk about. You don't want to do all the talking. That's what a rookie does. So what we did is had a successful meeting with them. Then there's an edit share capability. We basically stepped back and said if people go to five hours of meetings, they don't want to read five hours of transcripts. So they're going to want. It's more of a summary and they're going to want those key items extracted and pushed to their workflow. So Here you can choose anything and basically push it to any of these areas. Trello, Salesforce, team teams, all that kind of stuff. You can also set it up so it automatically does this for you. So, for example, if you're in a meeting and you can say, okay, Eva, the action item here is I should send a bottle of champagne to everybody in the audience and charge my competitor. Thanks, Eva. What happens is that voice command gets captured. It creates an action item. It goes into my Trello board and actually puts that action item in there. Don't worry, I'm not going to send champagne to anybody. And you get these automated notes. You can also do things like translate to other languages so you speak and the audience can see it. Okay, Eva, switch translation to Spanish. Thanks, Eva. What's going to happen now is the subtitles will switch to Spanish. As you can see, the last subtitle switched. When I stop talking, it'll move to another language. There you go. So it's now in Spanish. These capabilities are now getting embedded inside of WebEx. In the next version, you'll have closed captioning. You press and all of a sudden evil will take notes. And then we're going to start embedding the workflow items so that as you take these action items, push it through. We also have capability of doing automatically extracted action items, meaning you don't have to invoke its name. You can just say the agenda here is I'm going to demo this and then I'm going to open to questions. And because it knows that the word agenda is interesting, it's going to capture that and put it off. Or if I say the next step here is for me to send you the contract, it'll do that and so on. And so we had to blend a lot of design on explicit interactions with the AI and implicit interactions. This is one of the most interesting design points. Everybody wants to make it really easy to do the implicit stuff. Stuff just naturally this figure out and do it for me, those are really hard. One of the things we learned from Google, we actually talked to the team at Google about this is when they first started. What they did is they said, you're going to put in a keyword or two in the search engine the very beginning, and I'm going to show you 10 links and we think one of these 10 links will be useful. Look at the humility of the framing of the problem. It wasn't. You're going to say natural language and I'm going to show you exactly the one link you wanted Right. So it's an explicit interaction where you tell it, this is what I want and it answers to you. Look at Siri. When they launched, they started out with this kind of really big advertising with some of the best actors and actresses setting your expectation that you could have a full on conversation. And how did you feel about that? It kind of sucked. It didn't work as well and it took time for them to get better. Alexa came in with the opposite framing. They're like, hey, here's this thing, just tell it, turn on the lights or set a timer, play a song and it'll do it. And then over time it started growing skills. So look at the difference. So we tried to do the same thing. Blend. Hey, make it easy for people to explicitly say, okay, Eva, schedule the next meeting for Monday at 10am thanks Eva. Very explicit and kind of do some implicit things where it's just doing things in the background. This is just one of the really interesting problems when you're designing an AI product is dealing with the fact that people have unbounded expectations as consumers and you've got to bound it and over deliver on that bounding so that you don't disappoint them. So anyways, great journey. Then let me go back to kind of the journey of building Voica. So we started Voica in 2017, assembled the team, we launched in alpha in May, staying free. We did our funding right at the beginning as a seed round. Then we did our beta launch in November. And because we started having some stats on our beta launch that were really steep and up and to the right, we got our Series A right after that. That's when more strategics came in. I think Google and Microsoft, Workday and Cisco all came in that round. The seed round, it was just Salesforce. And then what we did is we recognized that a lot of people, the version we had would join your meetings automatically and then give you a web interface. There was no mobile interaction where you can just pull out your mobile phone and drop it on the table and talk in the context of there being no meeting. So we found this company, three people that were building a mobile app and they were using kind of off the shelf speech technology. We had invested in really good, accurate speech technology and so it was a good compliment to us. So we acquired the company and then what we did is we, a few months later we opened our paywall. It was a really scary moment for me because I had no idea if anyone was going to pay for it. They were really going to pay the Extra dollars. They're already paying Zoom and WebEx for the conference system. Are they going to pay an extra amount? We started converting right away. Yes. Question. You saved me, by the way. Thank you. I was wondering if you did anything or what steps you took at the early stage to prevent others from just copying your technology. Everybody was copying us right at the early stage. I'll repeat the question. Did we do anything at the early stage to prevent people from copying us? There were so many copycats. Why was I comfortable with that? I had an amazing tech team in a very high hard problem space and I knew that nailing the interaction with the customer was so damn hard. And I'll talk a little about that. Here's an interesting story. Before I started voicia, me and the CTO and my co founder who runs product were experimenting with this other idea that I refused to take funding for. We had built and an assistant that wasn't a voice assistant. It would read your calendar and your email and it learned very quickly to do things like hey, respond to this email. You forgot it. Hey, this person responds in 30 minutes. You take four days. Don't double cancel this meeting, you've already canceled before. Stuff like that. It's a really cool tool. Why didn't I start that company? Because exactly what you're saying. I realized that I would not have a data advantage that Google and Microsoft own each half of the world's email and calendar data. And I as a startup would be competing to get more and more data and all they had to do is buy crappy competitor number 72 and they could still beat me because they'd have more data. So I didn't found that company. This company I founded because I realized that 99.99% of all voice in the enterprise is in the ether, not stored in a system of record. So nobody has a data advantage. Literally nobody. So I figured if we had the right algorithms and we created the right user experience, we could get ahead of competitors. And we had a lot of competitors. They didn't scare me. I was worried about. It's actually the same insight behind the folks who started Intuit. When they started Intuit, they said they're not competing with other creators of software for accounting. They're competing with a paper and pencil. So I was competing with people who just took manual notes, not a bunch of startups who were trying to solve a hard problem and didn't necessarily have an advantage. So great question. The other thing we discovered to your question was there is a new type of competitive Advantage I didn't think of before. Bluekai had a classic network model, meaning the more data suppliers I had, the more data buyers wanted it, which created liquidity, which made it easier for me to close the next data buyer. And that loop continued. That was a classic network model. This had a new type of network auto. I call it a compounding competitive advantage. Because if you could build AI that is constantly training on this data, it's different. Everybody who does AI says their model's concentrating. The hard part is to automate the entire loop of data collection to algorithm creation, to making sure that you haven't created more false positive positives or change the profile of how the system works and automatically deploying it every hour. If you could do that, you're competing with other AI companies who will ship new algorithms every quarter. So if you can learn people's language models. So you walk into a meeting and you say a new acronym, the system doesn't know it, and an hour later it knows it. It updates the language model. It gets better and better. It gets stickier for the consumer, and they stay with you, and you get more data. So that kind of compounding competitive advantage is really interesting in this new breed of company. All right, so where were we? We got the mobile company, we hit the paywall, People started paying. Our economics were terrible. We were advertising on Facebook to get people to get to know eva, and the conversion rates was just okay. And then for people to try the freemium, and then there was a conversion rate for people to try the. To actually upgrade to pay after 30 days. And if you said no, we downgraded you to not the full experience, but at least you kept it free. And so we spent the next six to nine months just optimizing every single aspect of that funnel. Our economics started to get better, our conversion started to get better, and we started closing a few enterprises. Then what happened is it's 2019, and Cisco comes to us and says, hey, we want to embed Eva in WebEx just as a partner deal. We're like, okay. They're like, we did an accuracy test, and you're really good. You're actually just a hair ahead of Google. How'd you get there? So we said, oh, this is how our system works. Essentially, it's a cheating system. Let me just say it right off. It's called an ensemble. And the way the ensemble worked is we built a bunch of our own engines that did speech, and then we plugged in a external engine. We tested IBM, Microsoft, Google. We selected Google and ensemble would take the answers out of all these engines in real time and assemble a better answer than any one of them. Because it was a deep learning layer that was trained how much it could trust them. It took the original audio, created its own model, and it beat everyone. It was actually the performance envelope of the best of them. It took all the best attributes and did better. So, yeah, of course we were the most accurate, a little bit more expensive because you have to compute that much more. So Cisco came to us. Great. And now that we know you're cheating, we're a little bit less impressed. So here's the challenge. Because we don't want to pay you and Google, because your cost now absorbs their cost. And we have the security and privacy concerns that now the data's out there. We want to control it. We want you to rip out Google and. And still beat Google. Talked to my CEO and was like, uh, oh, we sat down for a quarter and we just focused on that goal. And we did it. We actually created an engine that had no external parties and it still beat Google's engine. At which point they were incredibly impressed because we were a small team. Instantly came up with an offer to acquire us. So at that point, I had a really hard decision. I had taken around 20 million in funding the done a Series A, so didn't have a ton of dilution. Small team, 30 people full time. And I had this really interesting insight, and that was, I was modeling myself after Slack and Zoom. Now, Slack came into a world where there already were things like Hipchat. And so they were optimizing on an existing model. They didn't invent it, they optimized it. And they created a viral loop that was amazing and a really fun, simple interface. Zoom. Same thing. As a matter of fact, Eric came from WebEx, so he just improved on a model that existed, except he took a consumerized approach, making it really easy to be viral, to kind of grow, to grow. Zoom. So both of those really Good Enterprise B2B Consumerized viral models took existing understood concepts and optimized the funnel and the virality. We were creating something fundamentally new that people didn't understand. They didn't wake up in the morning and understand, what the heck is an enterprise assistant? Is it Alexa? Is it a transcriber? I don't know. And we'd expose them to them, like, oh, yeah, this is kind of cool. But it was a little missionary and it was early. And so our economics were getting better. Our growth was very fast, 700% growth. All that Looked good, but I didn't see the enterprises converting that into big deals. So I said, okay, I'm going to have to take probably another 50 million in funding in order to really prove out the conversion from individual user to enterprise. And I honestly gave it a 50% chance of succeeding. Meaning, I don't know, like it's too early to tell. So then I would have taken 70 million in funding on the odds of flipping heads or tails. Like I don't know about that. So when they came in with a good offer, we were like, okay, we'll plug it into WebEx, we'll be able to hit 300 million users. And it went from zero to two and a half years and it was a good acquisition with a good return from the investors. So anyway, I tell you that story because the first company, Bluekai, we went all the way to profitability, went six years, they acquired us for 420 million. This one was two and a half years. They acquired us for 125 million. But very low dilution on our part because we only had a series A. So sometimes you have to think through these things. Someone might look at me and say, you should have rolled the die. You could have built a billion dollar company. You didn't need the money personally, so just take the risk. And I kind of look at it as like investors are like, they're people I know personally. If I'm going to take your money, I have to believe in better than a 50% odd. And so that was my reasoning. Was I right? I don't know. Doesn't matter. Like I'm having fun. We're really psyched being part of Cisco and are now part of a, a much bigger set of users and continue to innovate. Okay, I'm going to switch and now talk about my rules of building a business. Consider anytime you want to take this in any other direction, or if you have any questions about Voica or the AI problem we're solving, go ahead and ask. First, if you're going to build a business, you have to start with a really big problem. Problem, you can't start with an area that is too small because there's no way that your original inception is going to be exactly right and require no change. So it might as well be a big problem space. So you give yourself the flexibility of moving and changing. So that's just rule number one. Solve interesting big problems and if it's not big enough, don't rush, step back and say, what's a bigger problem for me to solve? That's the first one. The second one is really focus on an awesome team. Go find people that are better than you and have the humility to believe they're better than you. This is the number one problem I find with execs is the ego that prevents them. Then people are better than them. So they will hire B players so that they can stand tall or they'll hire an A player and constrain the heck out of them, not give them the freedom to execute. So go hire world class people. Give you an example. When we built Voicia, like I told you, my co founder came out of Facebook. This guy is like really, really well known. I mean he just really understands the problem he's solving. My next co founder had run product teams before he had the product manager. He ran product management. He had 85 people reporting to him before he came in. Even though he was. He had no one reporting to him here. He had the experience set the guy who ran sales, ran enterprise sales for WebEx and then built the sales team at Bluejeans. So in the conferencing space. So everyone we recruited was really, really good. So do that and be humble about hiring just awesome people. It seems obvious, but it is one of the important things. The next rule, blue oceans. Anybody read blue ocean strategies? Cool. So blue ocean strategies are really simple. An example of it is Southwest Airlines came into an airline market you would think is crowded though the way long haul flights at the time were. Lots of competitors, deregulation coming in. It was like an ocean with red blood because the sharks are attacking everybody and it's just really hard. Competitive, competitive market with a race to the bottom just didn't look attractive. Southwest comes in and has this insight and says, I'm not competing with airlines. Anyone know who they thought they were competing with? Any takers? Buses. Yes, buses. Southwest competing buses, not airlines. They did short hauls and they created a new market space and they grew and they grew and they grew. And that's kind of what blue ocean strategy is about, is finding a category where you can kind of create a new form of consumption. And Blue Kai certainly was that. Because we didn't compete with ad players, we created data marketplaces that were new. So I find those really interesting. The polar bears. Everybody looks at polar bears playing with nice little cute polar bears, think they're cute. What's interesting about polar bears is they eat their children. Yeah, ugly analogy. I know. So that's what it feels like when you're running a company and you run into a business model. That is so disruptive to you. You have to ask yourself, maybe I need to kill my own model before my competitor does. So that's what happened with us when we were at Blue Kai. Ebay called and said, hey, I want to use software, not your marketplace. And then I saw another competitor have the same model and I'm like, they're going to eat my lunch. If they give their software and it gets adopted to everybody, people, people will use my marketplace less. Why don't I invest and I do that. When we did that, the board was like, dude, that's expensive. Really? You're starting another business side to business within 12 months. I remember they came to our last board meeting with literally a crate of champagne to hand out to her. The funny thing is I don't drink. But they didn't know that. And so I looked happy. And so Polar Bears is all about cannibalizing your own market if you have to and not getting over attached to the way you've done your business. The next thing I would say about building a business is that I think there's two models of finding product market fit, surfing and sculpting. So let me describe them and you can tell me what you favor. So sculpting is basically the artist looks at a block of material and sees in their mind's eye the beauty that they want to create. And then they go in there and they creatively etch out what that masterpiece looks like. So that's one way to think about a startup. A surfer doesn't create. They don't create the waves. They identify the waves and the artistry is in how they ride it. But they're not fooled into thinking that they're wave creators. So those are two different models. Any preference? What do you guys think a good startup is? Yeah, I guess I live to witness. That's my opinion. It doesn't have to be right. When you think about the way Apple created kind of the ipod and stuff like that, it feels a little bit more like the sculpting model, but maybe not because they saw out there a need. They saw that it was underserved, that there were poor interfaces. So I don't have to be 100% right, but it's just my way of saying do more to read the way the environment is moving, the way ecosystem is shifting, and don't drink your own Kool Aid and thinking that you're the generation or you're generator of that trend and the rest of the idiots out there just have to get on board. Again, arrogant entrepreneurs will somehow think that. That's right. Last thing I already said this was have a deep understanding of the buyer ecosystem. So the reason we were able to do Bluekai and walk into the CEO of Expedia and land it sounds like a ridiculous deal back in 2007 to go to him and say the CEO of Expedia was going to approve a deal where he hands his entire real time data asset to a startup. Same thing with ebay. He sounds insane. And the reason we were able to pull that off is we knew how these folks think. We gave them a pitch that was so, so simple. We went up to CEO of Expedia, Dara, who now runs Uber, basically went up to him and I said, you have a small advertising business. He said, yeah. I said, do you believe in the next few years you're going to be the number one advertising company? He's like, no, not really. Number two, no. I said, how about this? Why don't we make everybody in the ad ecosystem compete for travel dollars by bidding on your data so every ad that gets served gives you money. It's like, great. How much money would that make? We had a conversation back and forth a little while we closed the deal and then we close a whole bunch after them. You can't do that if you're not from the ecosystem. Understand their psychology, have the connections and have the credibility. So that when they're asking, who's this dude asking for my data? Kick him out of my office, does background checks. He knows who you are. His CFO already knows you. So you have to be from the ecosystem. Unless you're doing B2C again, my idea. Anybody want to challenge that? Because I don't have to be. Right. I think you could be knocked from the ecosystem if your partner had a team with you. You said your team must be better than you, more experienced than you, and they are from the ecosystem and they are building and paving the way for the product. 100% right. I agree with you. It doesn't have to be you. Your team has to have someone at the top level it can't be buried in, but someone who's influencing your strategy, who's going to meet with other people in the ecosystem, who's recognized and who understands them. I agree. Definitely doesn't have to be the CEO. It helps if it's a CEO, but if they're a prominent member of your team, that's awesome too. Totally agree. Cool. So those are my rules for building the business. In terms of team building, I already covered the first one, which is be better than you. Hire people that are better than you. And I mean this meaning the technologist should be a better technologist, the salesperson should be a better salesperson, the marketer should be a better marketer. That's what you should be striving for. And if you really think you're just better than them all, then wow. And there are a lot of people like that. Egos are plentiful in this world. So empower and hold people accountable. This one seems obvious. It's where I think everybody falls apart here is in the tactics of it. Everybody thinks they're holding the team accountable, but then you go speak to the team and they don't have an agreed vision on what exactly are the small number of goals? Three to five this quarter, so that they understand everybody in the business, from the CEO to the very last person in that company, how they contribute to those goals. So I evolved this over time. But generally what I do before the beginning of every quarter, even in big business, I still do this. I start out with before the quarterly plan is developed. I start out about a month in advance with my own vision of what I think needs to change, how the ecosystems changing, what the goal should be. I then start week number one, go to my exec team and say, this is what I think. You have one week, go back to your teams and you tell me what you think it should be. So I come week number two, I just do a readout. They tell me what the bottoms up vision is for, what changes we need to make, what's important, what's not. Week number three, we do all the planning so that we end it with an off site and collectively decide what the goals are. Week number four, we come up with a goals poster that basically says these are the four goals or three or five with precise measurements that we can use. And then every week that goal poster is put up in front of my executive team and we basically say, where are you here? What help do you need? And we keep the focus on that. And I get, is there a reason that you don't get your team's vision before sharing yours with them? Do you think sharing yours first kind of biases their thinking on what the listeners look like? It definitely does. The question was, is there a reason why that I don't hear them first before I come up with mine? It's a really good question. I thought a lot about that and you want to have. I tend to be someone who cares a lot about the vision and finding that a Unifying vision through the company energizes everybody. And if it doesn't come from the top and you have more of a democratic vision that's coming from everybody, it's easy to get off track. And when I'm talking about vision, I'm talking about real fundamentals. So. So by me spitting it out, I'm able to set some guardrails so people don't feel like they're completely being rejected. So if somebody comes in and says, hey, I want us to do sock puppets, it's a really hard conversation. Well, it's an easy conversation to say no. But you don't want to outright just be cutting out the ideas that are problematic. So instead what you're doing is you're saying, here are the bounds and it's a wide bound. Your execution boundary may only be like this, but at least by setting it here, you're cutting out the rest of the area where you're just pretty sure we don't need to go there. You make a good point that that might not be the right answer. But if you're doing this on a quarterly basis with an execution mindset that you've got to actually get this stuff done in those 90 days, you kind of want those bounds to be tight. The conversations you want more open ended are a little bit longer term, like that Horizon two, Horizon three ideas where, hey, where do you want to take this? Those could be completely out and there you want kind of a little more inbound. Now recognize that week two is all inbound. So I don't really constrain people to say, oh, you said something I didn't say. Most of the good stuff pushes the edges, but you make a good point. So in any case, the last thing is that you take your team and now they take those goals and they take it one level deeper, all the way down and execute. And then you stand up in front of the company and you give anyone permission to say, if I come as a CEO with real excitement about a new idea that falls outside of these goals, you have permission to tell me if you want me to execute that. Tell me which of your goals you're going to drop. And that happens all the time. I will come in and say, hey, customer says this, can we get it out? And my product guy who knows me really well is like, here it is. Which one drops? And you have a lot more discipline in your execution as opposed to people getting excited about what the CEO says. Then they drop what they should be doing because they think the CEO liked it. You get all this randomness. So empower your team to get stuff done. Now, once you've agreed on these goals, you get the hell out of their way. All your job is to make sure they have what they need to succeed. And you don't script them on how they work because you've already agreed on what. Let them figure out the how and if they need to change in the next, what, one quarter out, balanced team. This one's an area where people fall apart, where they'll get very engineering mindset, but not someone who really understands how to take it, go to market, or someone with a very good go to market mindset doesn't understand the product side. You need that balance. I've been amazed at how when you speak to a set of really good algorithm engineers, they will look down at the salespeople as if they're stupid. I'm just going to say it like that, like literally they'll talk about it as if they're stupid. And then you look at the salespeople and they'll look at the engineers like they're such geeks. They don't understand business. They don't understand, like you have like all. And you don't want any one of those ways of thinking to dominate. You want all of it to come in a play field where everybody can bring their best and you want to foster some sort of respect between these people. Like you don't need your salesperson to code and you don't need your coder to sell, so on and so forth. So you want that balance and you want that balance also in thinking like extroverts, introverts, male, female culture, any diversity you bring in the team, the team will be a better team. Next one. Culture matters, but there's no culture in my opinion. That's really like the right culture. Like you got to figure out the culture for you and you need to figure it out early and roll it out. At Voic, it was really simple. We cared about gsd. Get stuff done. That was like a dominant thing because in a startup you need to move fast, come up to ideas, test it and so on. Second one we cared about was being data informed. And so not the highest paid executive in the room makes the decision. The data informs it. It doesn't drive the decision because sometimes you need intuition too. So that was one we wanted. Team players. I had learned early on when I went to mit, later came to Stanford, this mindset of the brilliant person is the one you want on your team. And that's all that Matters and my first 10 years of my career was that mindset. I just want to surround myself with brilliant people. The problem is sometimes brilliant people are assholes. Not more than non brilliant people. It's just some of them are and they're very disruptive because they will prevent good work from the people around them. So you want both. You want people who are really competent, but people who are what I call multipliers, that when you put them in a group, everyone around them works better. And so that was like one of our cultural things. I won't go through the rest of our culture, but that's what worked for my team. If you're developing something, be thoughtful about the culture because otherwise you'll have a random culture which is just the amalgamation of the character of the founders. That's not very thoughtful and you'll have to figure it out later. And you don't want to do that. So culture matters. And the last one is over communication of goals. You heard me say it before. I put the same goal poster every week in front of everybody. People then at the end of the quarter get to look at me say, how did we do against these? I can't BS them. Right. I've attended. I can't tell you how big corporations I've seen where general managers will get up and they'll show this thing about how well the business is doing. And you have no idea what they promised before. So they're just cherry picking what metrics make them look good. A huge number of executives play this game. Just show something up and to the right. Why are we looking at jujubes going up to the left? Does that matter? What did you promise last week? Right. So over communicating the goals gives you that sense of accountability. All right, let me stop and ask open for questions. So you work on this really interesting AI technology. It sounds like focus is really important for you, but like how do you stay so focused when there are so many ways? How do you not get exciting problems? Yeah, the question is how do you stay focused when you have this exciting AI technology applying in a lot of areas? It's a really important question that you have to believe that the area you're focusing on has enough fruit, right. That has enough room. So you want to do this more kind of depth first. Once you've chosen the right area, go deep, deep, deep to make sure you nail it. Then you can nail the next adjacency. The problem is if you're wrong about the area going deep in, you would have foregone the ability to explore a breadth first approach to find that one area that's going to just nail it. So you have to balance those two a little bit. We stayed our whole time here, but we had a plan that mirrored our cache that basically said if we haven't nailed it by this time and we're worried about cash, we'll start exploring. So we can make sure to give ourselves an opportunity to find another area. But we got lucky. Our area just had enough growth. Good question. Last question. The voicemail how do you ensure that the set comprehensions. Accents? Yeah, really good question. So speech speech models have two components. The question was how do we handle access? Short version. So speech models have two aspects to them. The first one is an acoustic model and the second one is a language model model. So the acoustic model will listen to your speech and turn it into phonemes, at least in English. And then the language model takes phonemes and statistical language models like how much words co occur and it basically says it's likely this word. And so the acoustic model is where you need to be able to customize for accents. It's one of the huge problems in speech. And what we did is we would map individual voices into an n dimensional space according to acoustic characteristics. And then what we do is we'd find this speaker's nearest set of people in that space and bias the answers the engines to that. That's how you take care of essentially acc. Sam.