INBOUND 2016: Paul Roetzer "Origins of the Marketing Intelligence Engine"

Thank you. So, a little background. As she mentioned, we were actually HubSpot's first partner agency back in late 07, early 08, I guess. So what we're going to talk about today is a little bit different than anything I've ever spoken about at Inbound. And I've been speaking at inbound since 08. Thank you. That'll help. But the context to understand why I'm talking about this is you need to realize our origin in the automation space. So we've basically been there since the beginning with HubSpot. We've seen the development of that product and of the company, and we've watched the development of the automation space overall. For me personally, I've always been fascinated with what comes next. I spend a lot of time and energy trying to optimize what we're currently doing. We don't use HubSpot as well as we could. There's plenty of plenty more we could be doing there. But I tend to be fascinated with the next leap. And so that's really what I'm going to focus on today is a practical approach to a topic that I think you're going to start seeing becoming a focus at pretty much every marketing conference next year. And for me, where it really accelerated is In April of 2015, I launched an internal initiative named Project Copyscale. And it was designed to answer one seemingly straightforward question. Can we use artificial intelligence to create content? More specifically, can we use machines to write blog posts? So, like many organizations and agencies, we were struggling to create content at scale while maintaining quality. I had just a month earlier gotten back from south by Southwest in Austin, where the managing editor of the Associated Press and the CEO of a company called Automated Insights were on stage talking about how the Associated Press had shifted earnings reports to 100% machine written. So they told the story of how every quarter the Associated press would do 300 earnings reports written 100% by humans. They were now doing 3100% written by machines. So I had also spent the better part of three years prior to that not talking publicly about it much. But I was trying to build software that would eventually use artificial intelligence to automate marketing strategy. So basic premise was you come in and you have $100,000 or half a million dollars or a million dollars, whatever your budget is, how do you figure out how to spend that money? And so I looked around and saw the complexity of marketing and all these channels that were emerging. And it wasn't just a few trade magazines and a trade show and a few Other channels, it was hundreds or thousands of channels where that money could go. And so I had seen what IBM had done with Watson on Jeopardy. In 2011. I wondered, well, couldn't we do the same thing in marketing? If you can teach a machine to solve natural language questions, can't we teach it to figure out things based on probabilities in marketing? I'm not a developer though, so I had no idea really how to do any of that. I just started kind of going down the path to figure out would it be possible. So because of that time, I'd spend it. I kind of knew theoretically where we could go with artificial intelligence, but I hadn't no idea what actually existed today and what we could use to transform our agency, the industry and business as a whole. So what we've Learned over the 18 months from April of 15 to today has transformed the way I look at what's possible now and in the near future. And that's what I'm gonna share with you today. So a simple exercise to kind of envision where this is going is think about how much time you and your team spend doing each of these things. Every I'm actually going to read through a few of them because it's helpful. So social media updates, figuring out keywords to optimize for planning blog post topics or editorial calendars, writing, optimizing, curating, personalizing, automating content, building workflows, writing ad copy, creating landing pages, predicting click rates, all of these things that we manually do today. Now in a lot of cases, we're actually using tools like HubSpot to automate this. But if you think about it, you're actually telling the machine what to do. You figure out what the workflow is going to be, you figure out the blog post topics and there are tools to help you. But as the marketer, you're still largely doing a lot of these things. Now imagine if you didn't have to, if all of those tasks we just showed were actually being done by a machine and you were just enhancing and curating and kind of reviewing and adding a human layer of intelligence over it. The crazy part about this is it's actually possible today. And there is a technology that exists to do that, and that is artificial intelligence. So now when you hear artificial intelligence, the natural thing people have is it's sci fi. It's like if you've seen how many people have seen Ex Machina or a movie like the Terminator Even in the 80s, that's something called artificial general intelligence. It's making A machine as smarter, smarter than a human. It doesn't exist. And theoretically it may never exist. We may never have machines that smart. That's what people think artificial intelligence is. It's not. This is Demis hospice. He's the founder of DeepMind, which got bought by Google for about $500 million in 2014. He was a child chess prodigy. He defines artificial intelligence as making machines smarter. It's called narrow intelligence. It's teaching machine to do one thing like IBM Watson. Teach it to win at Jeopardy. Taught it rules, engagement. You taught how to win. The key though is what it really does is it augments humans capabilities and knowledge. So you're going to hear terms and you'll probably start seeing them after this talk if you haven't already. Machine learning, deep learning, neural networks, natural language processing, these are all terms that are part of the AI umbrella. I'm not going to get into a bunch of definitions today, but again, just kind of familiarity. The other key concept to think about is an algorithm. So you again hear algorithms talked about software. You write algorithms every day, you may just not know it. It's just a set of instructions that tells a computer what to do. So when you go in and create a workflow in HubSpot, you're basically writing an algorithm. If someone downloads this ebook, send them these three emails. If they click on this link, send them these four emails. If they meet this lead score, then trigger this. You're telling a machine what to do. The difference with artificial intelligence is the machine tells itself what to do. It starts writing its own algorithms. That's machine learning. So that is like the core topic you'll hear talked about with AI. Machines start to get more intelligent, figure out new paths. I'm going to show you examples of this that are happening in other industries, but also already in the marketing space. So the best way to understand what is coming in marketing is to look at other industries that have already been disrupted by this. So in 1987, a guy named Thomas Pefferty hacked NASDAQ, literally took the feed coming from NASDAQ, taught a machine how to read what it was saying and make trades automated. NASDAQ thought that was a very bad thing because you needed a human or you needed someone to actually input on the keys what to do. So he then built a robot to read and tap on the keys to make trades. That started the downfall of Wall street in terms of being computerized. Today, over 60% of all trades are executed almost exclusively by machines with no human interaction. Again, I look at this and say deciding trades is probably more complex than marketing strategy. The variables that go into when to trade and what to trade is very complex. So for 30 plus years they've been doing this already in the financial institutions. Ups 55,000 routes a day. That's how many routes there are. The UPS does every day across the world. The Average driver makes 120 stops in a day. So what is the possible number of alternate routes that a driver could take? The answer is actually a quin sextillion which has 198 zeros. So if you think about the money that can be saved by ups if they make one less right turn every day, 55,000 times, that's the kind of complexity that's been being solved. UPS started something called Orion in 2003 to build a self learning algorithm to figure out how to optimize the routes drivers took every day. By next year they believe that they'll have all 55,000 routes running largely through Orion. Machine will be telling the driver what to do and where to go. Netflix. How many people have a Netflix account? So roughly 75% of what you watch on Netflix every day is recommended by some sort of algorithm. They're looking at your past behaviors. If you like this, you'll like this show. It's functions just like Amazon. What if you're buying it, it knows what you've previously bought. What people don't realize though is their shows are actually manufactured using these algorithms like House of Cards. They knew House of Cards was going to be a hit because they knew the behaviors of their watcher, the people that watch the viewers. And they were able to actually manufacture storylines and characters and plots and scenes based on a machine saying the probability of people watching this is this. There's a company called epagics that you actually take a movie script before it ever goes into production, you feed it into the algorithm and and it will tell you how much money that movie will likely make at the box office. It'll even recommend you to change characters and plot lines and settings and actors based on how much you can maximize in your profitability. So again, I look at this and say, well, how different is that than content? Like if I had an outline for an ebook or a white paper and I could feed it into an algorithm and say based on past behavior, this probability of this Getting more than 500 downloads is that so this is kind of where my mind has been for four years is just trying to understand what all this means. Another just examples in your everyday life. Any Tesla owners in the crowd. I'm jealous for one. If you are awesome. Have you used Autopilot? Yeah. So Autopilot is a feature that already exists within Tesla, which is kind of a partially self driving car. They announced about three or four weeks ago that every car currently in production and moving forward has the hardware to be 100% autonomous, meaning my 4 year old daughter may never need a driver's license because within 10, 12 years cars on the road will be largely autonomous and every manufacturer of cars is trying to do the same thing. If you use Siri and ask a question in Siri or a really cool one, if you haven't noticed this yet in the new iOS for the iPhone, go in and search for an image of something, a tree. I did Legos. Like my kids have Legos. Search for dog. It will pull up every photo you have. I have 26,000 on my phone. It automatically recognizes images of dogs or legos in all 26,000 photos and surfaces it for you. It's using something called Deep learning to do that. It's actually recognizing images with you never having to tag anything. Alexa works the same way. Natural language processing. It's understanding what you're asking it and figuring out what you want. A really cool thing that I'm a assuming Dharmesh is going to talk about today is Growthbot. So HubSpot's doing some really cool things through Dharmesh's kind of personal initiative in this space where he's built a tool that's basically a marketing assistant. How many people in here have tested Growthbot Cool, like Start? You see the potential, right? Like it's. So what he did is Dharmesh. And again, I'm not. He told this story at Partner Day. So I'm not telling you things you haven't heard, but he saw this opportunity, like to build a bot that could become a virtual assistant for marketers and answer questions like, what are my top five blog posts the last 12 months? What software do my competitors use? What keywords am I optimized around? What keywords are my competitor optimized around? And to spit out immediate responses to that. So he's been hacking away himself building this tool. And that's an awesome entry point for HubSpot. Like they're playing in this space. We just don't often know it's happening behind the scenes. But again, I'm assuming that they're going to talk quite a bit about that at some point today. So why marketing? Why now if all these other industries were already happening? We know back to 30 years. I tell you, the government has been working on AI since the 1950s. Like, this is not a new concept, but we weren't ready yet. There was a couple of pieces. One is data. Artificial intelligence needs massive amounts of data to be really, really smart. When Watson won at Jeopardy, they fed it the entire Internet like it consumed everything. Wikipedia page, any knowledge it could consume, it took in to figure out responses. So data was one. Processing speed is another. To do this well, you have to have computers that are really, really fast. So Even in the 80s, computers weren't that fast yet. Now they're very fast. The other thing, and the most important thing in marketing is the reward for disruption. So this is, if you want to, if this is intriguing to you, I would say start with this book. So this is what led me down this path. It's called Automate this by Christopher Steiner. In the book, he talks about how do you know when an industry is ripe to be disrupted? And it's this simple equation, the potential to disrupt. There's no debating. Marketing has the complexity that makes it worth disrupting. Like there is a potential there to make it more efficient, to make us more productive. The reward, though, is the key. There's more money on Wall Street. There's more money in healthcare, there's more money in logistics and transportation. Those industries came first because there's only so many people that actually know how to do engineering for AI. Machine learning engineers. Why would they build marketing widgets if they could take down Wall Street? So you needed kind of this curve of adoption to where there was enough people and enough reward to spend their time in marketing. So one of the key ways to figure out when a market is going to be disrupted is to start following the money. So this is a really simple look at the automation space, which we'll just lump HubSpot into for the sake of this argument. So starting in 2012, you had exact target IPO for 161 million. Oracle buys Eloqua for 871 million. Salesforce buys Exact Target, who had bought Pardot for 2.5 billion. IBM buys Salesforce. These are all marketing automation plays. And then currently HubSpot's market cap as of this morning is 1.3, $1.935 billion. So just in the 10 years HubSpot has existed, this whole automation, automation space has emerged and massive amounts of money have flowed into it that led to all these innovations. The challenge with automation as it exists today, though, is as we started this talk about, it's largely still manual Automation takes what you tell it to do, and it makes it wildly efficient. It makes us more productive, but you're still telling it what to do. The other problem is IBM tells us that 90% of all the data in the world was created in the last two years. So if you think about as a marketer, all the data you have access to that could inform what you do next, where you put your marketing dollars, where you spend your resources, what staff you hire. All of these are sources of data for us. But the problem is, like, how do you figure out what to do with that? Which analytics report do you actually look at to drive your next action? So as humans, we have this finite ability to actually figure out what to do, to process all this data, to optimize our performance, to create content at scale. But algorithms, that set of instructions, where they actually start learning themselves is limitless what it can do. So they have the infinite ability to process the data, deliver predictions, recommendations, content. So like, lead scoring is kind of like a predictive thing in HubSpot now where you put information, but you're teaching it, like, if it meets these criteria, give it this lead score. That sort of stuff is like moving into the prediction space. But largely, marketing is still human powered with some automation mixed in. But the future is actually a lot closer than we probably think. So I mentioned earlier the Associated Press So they're writing 10 times more earnings report reports using a technology called natural language generation. So again, it's just kind of one of those AI terms. But what it means is the machine is actually creating narratives. So this is an example of an earnings report by the Associated Press. This is from Q2, but you can see up top, the byline is by Associated Press and it reads exactly like a human would have written it. Except at the bottom it says, this story was generated by automated insights using Zack investment data. What they did is they took a feed of data and they taught a machine how to tell the story. And it can tell this thing at scale infinitely. It can do thousands of stories in the time it took you to open the Word document. So this is eight months ago. I actually thought the machine just did this. So go back to that question of can we create content at scale? This technology is what made me think, instead of having to figure out how to staff a bunch of writers, maybe we could actually automate it through machines. So I started looking into natural language generation and we had this use case. We have this online assessment tool where we have about 3,000 people who've gone through and rated their Marketing. So I thought, well, let's try and automate this. So instead of just visualizations, after people go through and rate themselves across these 130 factors, we could actually give them findings and narrative based on their report. So this is an example. Like, you know, if you go in and do the marketing technology, it may spit out and say, marketing tech is critical and these are the areas you rated poorly. And we would recommend focusing here. So it's trying to recommend activities. Right now we have to manually do this. And if we go into a single report, it takes probably an hour to pull like 10 to 15 key findings. They have to write it and edit it. And it's a process. It's cool and it could be really valuable, but it's time consuming, so we don't do everything we could with it. So I looked into Automated Insights that the company that does the Associated Press, and I figured, well, let's just marry it up. Let's connect this data source to theirs and boom. So lucky for us, Automated Insights actually released a product, a SaaS product that for 250amonth you can subscribe. Now, put your data in and it spits out a narrative in theory. So we get automated insights. I upload data and you see, insert data, add synonym, add branch. I'm like, where's my narrative? Like, there's no words here. It's a blank screen. So what I then learned, natural language generation is actually a human writing the original narratives. You still have to envision that story. But then like when you're creating branching logic in HubSpot, you actually tell it if it's greater than 80, it's strong or it's fantastic, or you create all these variations. And then so the green is like the data points which automatically get inserted. The purple are synonyms, and the blue underline is branching logic. So I was like, okay, so there's actually work to be done here. But that. So one, it was like, oh, not as exciting as I thought. Then the other part was like, but who knows to do this? Like, there are, I don't know a single marketer that even knows what natural language generation is. I know a lot of companies that have massive amounts of data and are right now manually telling the stories of that data. So it started kind of opening my eyes to the possibilities for marketers to integrate this kind of stuff with your inbound marketing campaigns. Because if you do like state of the industry reports or if you have proprietary data, like you're a software company of all this user data that you just don't know what to do with or internal reports. We've been using automated insights for a year and a half to do our Google Analytics reports so that they actually built a template for, so we didn't have to create it. But in real time we can just hit a button and it'll take any portal we're connected to in Google Analytics and turn out a seven page report of everything happening on the site with recommendations of what to do. So that's like a really practical application. And there's another company called Narrative Science that has a free product to do the same thing. So you can just connect it to your Google Analytics and it'll send you like weekly summaries of what's happening on your site, your AdWords, Facebook campaigns, things like that. So for us again in that use case we're building, if I wanted to do 50 briefs a month based on marketing score assessments, it would take about 50 hours, about an hour of brief. In this version using kind of this machine assisted approach, we could cut it down to about 15 minutes per and that's largely editing time because it'll actually spit it out instantaneously and that's about 12 and a half hours. So we could save 37 and a half hours and create 50 times more content than we're currently creating. That's an AI technology that exists and is available to you today from multiple vendors that you can marry with HubSpot to start telling better data driven stories. And again the key is you actually need humans that know how to do this. So you need people who are writers and that can figure out what they the stories are going to be and then tell them within there the benefits more accurate. There's zero human error. So if you think about the Associated Press earnings reports, unless the data is wrong, there is no chance of having an error within the content because it's just pulling right from a spreadsheet. More content published, more cost efficient, creates more engagement, more opportunities. So where is this going? Like what's kind of next for it? So this is again that follow the money idea. Private Investment in the AI sector has grown from 1.7 billion to almost 15 billion in 2014. The AI based analytics alone could grow to 70 billion by 2020. There are already, and this is what we've kind of learned in the last eight to 10 months. Dozens of tools, companies, products being built that are powered by artificial intelligence that can do these things, predict, plan, create content, optimize the content, promote, measure, analyze, personalize, and it can all in some well in some cases, it can actually be integrated with HubSpot to do it. And that's the real key. And one of the things I'll touch on later is you need to look at your current tech stack, the things that are core to what you do and find ways to complement it. There is no AI platform you're just going to move everything to and you're just going to leave here and everything's going to be solved. You can start using pieces of this to enhance what you're doing with your inbound marketing program and potentially make leaps forward in terms of efficiency and performance. Again, the follow the money concept, almost 280 million just to these 10 companies. These are all AI powered companies that do some variation of analytics or content creation or prediction. What we found is most of the time tools that are being built that you can actually integrate with a HubSpot or you can use to augment. Inbound marketing are primarily built for enterprises, so they're still largely customizing the solutions. So it seems like, oh, this sounds awesome, I'll just plug this in and we'll go. They're still spending a lot of time customizing for people, so you often need the bigger budgets. Automated Insights is one of the first I've seen to create kind of a cloud product that's very affordable, that can make it approachable for people. By the way, this deck is going to be online afterwards too, so if there's things you want out of it, it'll be on my Twitter account. I think Inbound will publish it. A big. I wouldn't say it's a watershed moment, but an important moment in the marketing industry happened in September. Salesforce introduced something called Einstein. Salesforce has bought nine AI companies in the last two years, spent over $2 billion doing it. And what they've done is they've taken all these technologies and created an AI layer that now sits throughout the Salesforce product. And it's emerging. But if you have HubSpot plus Salesforce, you could actually start integrating some of the technologies baked into their product to do some more fascinating things with both products. And this is kind of, you know, a quote from Benioff like, this is the future. This is where we're going. We have to accept this. And they've gone kind of all in on this play. If you ever wondered when you upload a photo to Facebook, how it knows who else is in the picture with you, that's deep learning. It's facial recognition. It actually takes like a couple dozen points on someone's face and marries it up to what they have in their database. So they use deep learning kind of a. It's not a big secret, but for people who haven't followed it, it's a bit of a secret. Facebook is an artificial intelligence company that happens to have a social network to gather a ton of data on you. Like that's what they are. If you looked at the earnings report call, I think from Q2. Zuckerberg said he was kind of stepping a bit back from a CEO day to day role to focus his energy on three things. One of them was artificial intelligence. They in fact have an entire research lab at Facebook built to do exactly that. So they say, we're committed to advancing field of machine intelligence, developing technologies to give people better ways to communicate, seek to understand intelligence and make intelligent machines. So earlier this year, when they made the big splash around bots, that was kind of coming out of this. Google is back in 1999 or 2000, Sergey actually said, we're an AI company. Same thing. Everything you do on Google, they learn from. They have more data than probably anybody in the world and they're using it to do all kinds of interesting things with their technology. This affects you as a marketer because earlier this year.