- Hello and welcome to today's HubSpot Academy Master Class. My name is Kevin Dunn and I'm the new in-bound professor for HubSpot Agency partner training. This master class is presented by HubSpot Academy, HubSpot's official learning resource and the world wide leader in in-bound marketing and sales education. HubSpot Academy offers free certification courses and free tools to not only grow your business but to grow your career, as well. Today, we are joined by Paul Roetzer . Paul is founder and CEO of PR 20/20, a Cleveland based content marketing agency and HubSpot's first agency partner. He is the author of the marketing performance blueprint and the marketing agency blueprint, creator of the marketing Artificial Intelligence Institute and marketing score, a regular contributor to leading marketing industry blogs and a frequent speaker on marketing talent, technology, strategy and performance, a graduate of Ohio University's EW Scripts School of Journalism, Paul has consulted for hundreds of organizations from startups to Fortune 500 companies and has been recognized by Smart Business as an innovator and business rising star. So, no shortage of--

- We could cut that in half.

- Plenty of accolades but Paul thank you so much for coming in today.

- Thanks for having me.

- Yeah, you bet.

- So maybe to get started, the best place, this would be a question for both you and anybody joining us on Facebook, as well. But I mean what is artificial intelligence? How would you best define it?

- Yeah, it's one of the core things. So when we started really exploring AI, trying to understand what exactly it is, how you can apply it, was the starting point for us. And the reality is, if you go to Google right now and search what is artificial intelligence, you're gonna find, take the first 10 experts, those 10 experts are all gonna have different definitions for what AI actually is. So my favorite in large part because of its simplicity is from Demis Hassabis. So Demis is the founder of a company called DeepMind which was bought by Google for about $600 million dollars in 2014. And so today, he is still the CEO of Google DeepMind. Recently, they were in the news for their Alpha Go Zero which we can talk a little bit more about later but Demis calls it the science of making machines smart. So in its most basic sense, machines have no knowledge out of the box. They don't think like humans. They can't tell a chair from a table. Like they have no knowledge. And you have to train these machines to learn, to think like humans, to do things in some cases at super human levels. So AI is this umbrella term that covers all the technologies and processes that make machines smart. The most common terms you would then hear under that umbrella would be machine learning and deep learning. But there's also things like natural language processing, natural language generation, image recognition, facial recognition. If anyone's lucky enough to have the iPhone 10 right now. You unlock that with facial recognition. That is a form of artificial intelligence. The phone, the hardware, the software doesn't recognize the human face. It's trained to do that using things called Neural NetS. So these are all types of AI that are taking machines, taking hardware, taking software and enabling them to be smart. So that's really what it is. The simpliest way is to think of AI as the umbrella term. Machine learning, deep learning, all these other terms fit under that. The basic takeaway from market is You don't have to have a deep knowledge of all of these tools. You just have to understand what it's capable of doing for you.

- Sure. And actually that leads into another interesting question. But I guess just to clarify. How would you distinguish machine learning from deep learning?

- Yeah, so machine learning is really the process of, well, deep learning is a type of machine learning.

- Gotcha.

- Basically. It's a more advanced version. So in machine learning, it's the machine, developing on its own. So a simple way to think about this is if you in a marketing use-case model, if you an e-book on your site to be downloaded, and someone comes there. You as the marketer set a set of rules. You write out the rules basically, that are a set of rules that say, what happens when someone downloads that e-book. So if they download the e-book, send these three emails. You as the marketer figured out which three emails to send. You figured out what to put in those emails. You picked the images, you picked the colors, you picked the subject line. When to send them. All of these are data-driven decisions and they should be, that right now a human largely makes.

- [Kevin] Sure yeah.

- In machine learning and in the human monitors performance and tries to alter what happens. Machine learning is the machine taking over a lot of those individualized tasks and learning on its own and actually altering the path forward based on what its figuring out at levels a human can't comprehend. And so now instead of like you saying send all emails at 2:00 p.m. on Thursday, the machine may very at an individual level when those emails go out. It may recommend different headlines 'cause this one performed better. So, that's machine learning. It's just the basic way that the machine uses kind of computer science to keep getting smarter. Deep learning starts introducing what are called Neural Networks which is basically trying to get the machine to think more like a human does. So when you look at, a great example of this, to understand how really unintelligent AI is at times, if you tell an AI that is a dog. You have to show the AI three million dogs versus three million cats before it figures out what a dog actually is and at a high-degree of probably can get it right a lot of times. If you show a baby, a 10-month old baby, this is a dog, a dark barks, that baby's gonna recognize the dog probably every time moving forward. So what's happening is, in the human brain, these Neural Networks, like layers of how you recognize things, that's how their trying to now train machines how to think, is like how these layers functions in our brain thus the deep learning part. So going much deeper because first you see there's fur then you see there's legs and there's a tail and it barks like all of these things our brain just does. And so there's lot of money for decades now has been poured into ways to try and understand how the human brain works and they've made a lot of progress. And now they're trying to give that knowledge to machines so Google DeepMind is one of the furthest along. IBM Watson is trying to replicate the human brain. Like there's lots of efforts. The US Government's been doing it since the 50's like. It's interesting.

- I think, what was the recent update to the iPhone right. I mean you could go to your photos and search dog and it'll recognize all the dog photos.

- Yeah and I just last week, the first time this happened to me and some people may have had this happen. If you're in your photos in your phone. It'll actually say, is this, this person. Like who is this.

- It'll recognize, yeah.

- Right. And what it's doing is it's having you tag somebody. So they would do for my daughter, so my daughter's five. So it said who is this. And I was like, well this is interesting. 'Cause I know what's happening now behind the scenes. So I said, well that's Eila. And then it pops up another dozen photos and it has check marks and all it says is this her? And there's photos of when she's two months old, when she's five years old.

- Really.

- And so what it's doing is it's using Neural Nets to recognize that person. Much like Facebook when you upload a photo and it says do you want to tag Kevin in the photo.

- You can autotag, yeah.

- Right.

- It's using that facial recognition software to actually figure out who that person is. And then every photo from then on, you have now trained, you've done deep learning basically.

- Yeah, you felt it, yeah right.

- You've trained photos to recognize, in this case, my daughter, and every future photo I take will have, it'll know that that's her.

- Yeah, that makes a lot of sense. I want to go back to a second, you mentioned this a little bit. Well I guess, obviously, you've been putting out a lot of content and you've been speaking about like the inevitable impact AI and machine learning's gonna have in marketing right. And I think we talked about it for a second there, how all these decisions, when sending an email, right, maybe the tool can figure it out and optimize and kind of handle it on its own. But if you can expand on that, I guess. Like why should the marketer or any marketer care about AI and how is it gonna impact their role.

- Yeah, I mean this is something I've thought a lot about for, well it's 2011, is honestly when it started for me. When Watson went on Jeopardy. And to me that was kind of a pivotal moment because I didn't understand AI at the time. But I started from a marketing perspective thinking, wow, if you could apply that same technology. 'Cause the way Watson works, so people are familiar is like it doesn't actually know the answer. What it does is it takes a massive amount of data all kinds of information and then it looks at probabilities of an answer being correct. And if the probability hits a high enough confidence level it would answer the question. So I looked at it and said, okay so when people are trying to figure out how to spend their marketing budget, how to build strategies, we pull from this like limited knowledge and capabilities we have as humans. Take the best marketing strategist in the world and they still have a limited knowledge, they have their specializations, things that got them where they are in their career. And so I thought well, if you cold apply that same Watson type technology to marketing strategy, like what would happen?

- Right yeah.

- Wouldn't that be possible. And so then I read a book called Automate This by Christopher Steiner in 2012. And it talked about the disruption of Wall Street and how at the time, 60% of all trades made were done a 100% by machines with no human interaction. Talked about health care and logistics and transportation and all these places where AI was already disrupting. And there was a simple equation in the book of when to know when an industry would be disrupted. And it was the potential to disrupt plus the reward for disruption. So the way I looked at it in 2012, was the potential to disrupt the marketing is unequivocal, like it is everything is data driven or should be. It's all, so there's no doubt it would make sense in sales and marketing to apply it. The reward for disrupting marketing wasn't as great as health care and retail and e-commerce and these other areas. And there's only so many people who can do AI. There's a recent study that says about 10,000 AI researchers in the world.

- Wow, okay.

- And Facebook, Google, Apple, Amazon, admitted that they buy all of them. In some cases, paying millions of dollars for developers. So in marketing, I've looked at it now for six, seven years and said take every use case you just explained, like, email send optimization or recommending content on a site based on different factors, all these things that we currently do, that we term a lot as marketing automation. The great irony of automation is that it's mainly manual. So you have software that enables you to automate things at scale but the human, the marketer, still programs it all. It's still--

- It won't automate until the human builds it. Right.

- Yeah.

- So there's no real intelligence other than the human intelligence. Which is fine but it's limited. Like it's a finite ability to get better. So I've always looked at it and said, well intelligence algorithms have an infinite ability to get better. All it really requires is better data, more data, cleaner data. And increases in computational power. So the ability to computate at a higher level. Well in the last three to four years, you've seen massive improvements in computational power. And the ability to actually do things in the cloud that 10 years ago you couldn't even fathom doing. So, in marketing if you look at anything you do that's really time-intensive and especially if it is data-driven, a machine over time will do that better than a human, as soon as someone sees the reward for solving that to great enough to put resources to do it.

- [Kevin] Sure.

- So like HubSpot has started doing this with bots, with the acquisition of Kemvi which enables to start doing a little bit more on the sales side. All of those are recommending content is something I think that's moving forward. So all these different things that marketers currently do, you're gonna have AI, in very narrow use cases, be built to solve specific great.

- Actually let's pause there. I think as we can build off some of these questions. And I think you spoke to this we're talking a little bit about how like how it's gonna impact the marketer. But is there anything that you would note in regards to like digital advertising methods, this question comes from Karen.

- Absolutely, yeah. Programmatic ad-buying is something that's been going on for a while.

- Sure yeah.

- And that was a first entry. But actually a company to look at would be Albert. I think it's just albert.com. They use to be called Algorhythms. They just changed the name of the company. It's one of the more advanced AI systems on the market. And what they do is, it's pretty mind-boggling if you're new to the space and haven't really seen an application like this yet. What they do is, the human does the creatives. So you come up with, it's all for digital ads online. The headlines, the images, the calls-to-action, like you figure that out. And then you upload your variables, your variances of it. So like here's three headlines, here's five images, here's body copy.

- Sure, yep.

- And then you say I want to spend $5000 a day. Albert then in fully autonomous way, with no human interaction, starts running infinite variables of it.

- Sure, yeah.

- And it adapts which creative works best to which channel. So it may start running ads in Facebook, and Instagram and Google and LinkedIn and wherever. And on then it's own, running 24/7, it never has to stop, it starts figuring out the best creative and the best channel and it automatically reallocates the budgets to the proper channels.

- Yeah, that's wild.

- So in terms of what, digital advertising, it'll completely disrupt the space.

- Yeah.

- Now the challenge with Albert is they have a minimum ad spend of 20K a month.

- Got it.

- So it's not gonna solve anything for small, mid-size businesses. And I haven't seen any other true competing tool. Now it may be out there and if someone is aware of it I'd love to hear about it. 'Cause we're trying to find those kinds of solutions for people. But Albert is a great example in digital advertising space. The other thing would be like Pathmatics is a cool one. It's a company out of Santa Monica. And what they do is competitive intelligence. So they use different types of machine learning, different types of AI, to monitor what people are doing and what they'll do is they'll pull all the ads into your dash board and they'll show the creative your competitors are running. They'll tell you how much they spent on it. And they use different algorithms trying to figure out spend. They'll tell you how many impressions they had. And they'll alert you any time that the creative changes. And so you can actually, again, one of the interesting things we'll talk about kind of a theme throughout this is you won't often even know you're using AI. Much like in your life with photos or if you're using smart replies with Google, an app.

- [Kevin] Sure yeah.

- Like you're using Netflix. You're using it all the time, you have no clue. That's what marketing is gonna be like. So you may look at Pathmatics and be like that's an interesting competitive intelligence. So that's kind of cool. Get a subscription. All the while not even knowing you're using AI now to monitor your competitors.

- Yeah, right. So I guess there's two things there that I think are incredibly interesting. So one, AI exists in our daily lives whether we are already starting to recognize it or not. We've already given some examples. But I'd be interested to see in the comments if anybody else has some other examples, right. We talked about the iPhone 10. We talked about Netflix. But I'm sure there's plenty of other instances out there.

- On Facebook, it's all over the place. It's how they figured out how what to show you on your newsfeed, how they figure out which images to show you.

- Who to tag.

- Yeah.

- Yeah.

- Amazon is a phenomenal example of everything they do to customize your experience on Amazon and--

- Yeah sure, what to buy next.

- Yeah, that's one of the trends you're gonna see is like, how you experience AI in your daily life and how convenient it makes everything. Google Maps, another example. Waze, like all of these tools, are everything in them in AI. And so your expectation as a consumer for that level of convenience starts to trickle into how you buy in your business life, as well. So if you're a B2B company and you're selling to another human, they go throughout their day with the convenience of all thee things that are done using AI even though they don't know it. And then they get to your site and there's no recommended content and you're getting emails at the same that everybody else sends the emails and--

- [Kevin] Right.

- Like your experience, there's a disconnect.

- [Kevin] Right.

- And I think that's the opportunity in marketing is to get up the level of where consumer expectations are for the experiences they have, dozens of times throughout the day.

- Something like timing, relevancy, personalization. And like all of that is just hyper-focused.

- Personalization is huge. I means it's really the holy grail of what we're trying to do with AI, is that one-to-one experience.

- Yeah. So, what is sounds like too and correct me if I'm wrong but like AI you know, incorporated in marketing so I mean, I'm thinking of like these time-intensive tasks that I have to do, right. Sounds like well A, if I'm a marketer, then yes I know I have to test. I'm gonna A/B test or A/B/C test. But it sounds like with the power of AI potentially, it's not only going to make those testing elements quicker but it's also gonna do so in like a way more advanced array. So like not only am I gonna test more efficiently, but it's gonna actually like present results way more advanced than I'd be able to do myself.

- Yeah, human, one of the first questions I was always get is like am I gonna lose my job. If your job is running A/B tests, yes you're gonna lose your job.

- Sure.

- I can't say that unequivocally for most career paths.

- Yeah.

- But if you're running A/B test of landing pages, like, you're not gonna be doing that in the future.

- So AI will be able to do A/B/C tests in the blink of an eye.

- So much better than a human.

- Sure.

- And like a really practical example I like to save for that one is let's say you have video on your site and you're trying to figure out what color to make the play button. Infinite number of choices of what color to make it.

- Yeah, of course.

- Who are you or I to figure out whether red or orange is gonna react better to people.

- Sure.

- At the end of the day, that color button should be different for every individual that comes to it based on what they're most likely to click on.

- As a unique individual.

- Right, so personalization to that level. You need a ton of data to figure that out. And it's not always gonna be perfect but you can use look a like matching and like--

- Yeah of course.

- To start weeding down but.

- So it sounds like we've been talking a lot too about how we're going to improve the tasks, the responsibilities of a marketer. Is AI bringing anything new to the table or like any new services or any new responsibilities that a marketer would have to take on in order to implement AI, if that makes sense?

- Yeah, so. Probably a good example of this would be so I started in 2011 thinking about AI a lot. But I didn't really go all into it until 2015. So at that point, I was at South by Southwest and there was a panel with the executive editor of the Associated Press and the CEO of a company called Automated Insights.

- [Kevin] Sure, yeah yeah.

- And so they told the story of how the Associated Press was doing 300 earnings reports a quarter with humans writing them. And using Automated Insights, natural language generation technology which is a type of AI they were now doing 3000 a quarter, a 100% written by machines.

- Yeah.

- And so I sat there and I was familiar with Automated Insights but I didn't really understand exactly what it did at that point. And I thought holy cow, like, can we automate creation of blog posts. Like how exactly does this NLG technology work.

- [Kevin] That's the approach to content that we were--

- Right.

- Yeah, sure.

- So we write a ton of content, it's basically what we do for clients. So I came back and we launched this Project Copy Scale is what we called it. And it was basically, how can we increase the volume and quality of content potentially using AI. And so what I learned over about a year or so process of beta testing different tools and exploring it is that there's this whole new career path potential that opened up. So what happens is, the NLG doesn't, you don't give it a data set and it just automatically tells the story which is kind of what I assumed. So I took a sample data set, dumped it into the NLG engine and I saw a blank page. And I was like, where's my content?

- Thought this was supposed--

- I thought I just generated, where is it?

- Yeah, yeah. And we were using the same technology to do Google Analytics narratives at the time and it just worked like magic. Like I flipped it on and hit a button and there was a seven page narrative of my Google Analytics. So I again assumed you put data in and narrative comes out. Not how it works at all. What has happened though is you now have the blending of data, science and journalism. So any organization can take their data, think about any stories they currently tell with data. It might be financial reports, monthly marketing reports, proprietary data of a software system, whatever it is. If there's data, that someone is writing a story about, you can train an NLG engine to write that story, at scale so you, a human could write it once as efficiently as the machine but then the machine can then instantaneously write it a thousand more times with any variable in the data set. So, what I realized is the answer was no, we can't use the machines to write blog posts by just putting the data in and out comes the story.

- [Kevin] Sure.

- But what we could do was take any data set and the best way to visualize is it has to be structured data in columns and rows. So again if like a really practical example to picture, if you're in an e-commerce store and you have 50 types of TV's you want to sell. Instead of a human having to write descriptions for each of those 50 televisions, you would take a spread sheet and you would have screen size, resolution, how many inputs, all these variables.

- [Kevin] All the, yeah.

- So each column is a variable and then each row is the brand of television. You would write one template. Maybe it's 200 words long. It's a 50-inch screen TV with x, y, z, da da da da da. Price point is this. And then the NLG would actually be able to take the next 49 rows and take that same thing. And in your narrative, your template, you would put synonyms, so you would do branching logic. Like if it's this size, then--

- You can put this, this way.

- Put this sentence in.

- So you're as the human, training it but it is the blend of like, envisioning a story driven by data and then being able to tell the story. It's really two career paths that have sort of converged. And then you can write 50 descriptions in less than a second.

- [Kevin] Right.

- So we've been looking at way to apply that to different organizations that have a ton of data. And so in the case of like creating new things back to your original question. A long way to get back to the question.

- [Kevin] Sure, no yeah, that's good.

- What it did was instead of just writing stories and just analyzing data, it created the ability to tell stories at scale using a form of AI technology.

- [Kevin] Sure.

- Now most NLG on the market today, doesn't get smarter on its own so there's no, back to some different variations, there's no machine learning involved. Like, it doesn't then look at behavior on the site and how people engage when--

- [Kevin] How descriptions impact, yeah, purchase.

- It doesn't alter the description and make it better.

- [Kevin] Sure, sure.

- But that's where NLG can go is like it can blend machine learning over it. And start to actually improve its own content. And that's back to the Google example. If you go into your Google app and you go to respond to something like say someone sends you a meeting. Like you want to meet this day, this day, this day.

- Yeah.

- It'll recommend three answers to you. And you can just pick one and click it. That's using AI, that's using natural language processing to understand the words and what was asked.

- What's the desired action of this and kind of leading you to the answer.

- Exactly and then the NLG is producing words that you can use to respond.

- To string the sentence together, sure.

- So Google has a team of 35 engineers led by a guy named Ray Kurzweil who is one of like the forefathers of like modern day AI thinking and so he's leading a team working on that piece of the product.

- Wow.

- And so if you think about it, it's like why would he be at Google writing three word responses. The answer is that's not what he's doing.

- Yeah.

- They're learning so they're trying to figure out the human language, how to generate it.

- Yeah.

- And so I think in the next three to five years, you're gonna see a lot of right now, humans are uniquely capable of writing narrative. Machines are uniquely telling data-driven stories at scale. I think in the next three to five years, those lines are gonna start blurring.

- Really?

- Machines will start being able to create narratives. it's a really really hard problem. But if Google has a couple dozen engineers--

- The brightest in it, yeah right.

- Like they can probably make, so start to imagine like instead of just smart replies, what if you're in Google Doc and you're staring to type a paragraph in Google Doc and it starts recommending how to complete that--

- What the next phrase should be or yeah, wow. Let's pause there for a second. I think Stevie brings up a really great question. This actually may be a good transition. But is there a trade publication you would recommend to stay up on new artificial intelligence solutions, technologies. I mean, any place that you would recommend to learn more even you know.

- Trade pub, not necessarily. There's a ton of great online sites for it though. So, like Futurism is a blog I read. The Singularity is another one which actually Rieker is involved with that. The Future of Life Institute. O'Reily has a great AI newsletter. So there's a lot of sources and I think so Mike Kaput is the director of our AI Institute. I think he's probably involved in the chat so he can throw up some of the other resources we look at.

- Oh perfect.

- So people can kind of check out in the thread there. But yeah, there's probably about a dozen or so that I follow regularly to stay informed. 'Cause again, when I started in this, I'm not an AI guy. I don't have a PhD in computer science. Like, I'm not a developer. So mine was just an obsession to gain knowledge and understanding. And so I did the things you would do. I went and found the books to read. And I've probably read, I don't know, 10 or 12 books on AI and from the basic stuff starting with like automate this to the stuff that'll blow your mind. Like, Pentagon's brain and Life 3.0 and our final invention and like. So you just start to grasp that and then I built a private Twitter list of dozens of AI experts and so when something happens in AI and I think it's important, I'll go check that feed and say like, do they think it's important? Are they talking about it also? And then you start to kind of see through that. And then with the AI Institute we've done a ton of interviews. And that gives us the chance to learn through those experts as well.

- Some of the influencers and experts in there, yeah. That's great. One question that I have that we haven't really spoken about. Where do like chatbots live amongst all of this, right? So I mean obviously, HubSpot has GrowthBot and I'd be interested to see like what your thoughts on them and kind of like where they play in this landscape.

- Yeah, chatbots are huge. Obviously, Facebook made a big play there.

- Right, with Messenger.

- Probably a year and a half. Two years ago. So it's an obvious entry point for a lot of organizations. And for a lot of marketers, it'll be one of those tools that this looks really neat. You integrate it into your site. Don't even necessarily know or care that it's AI but all of a sudden things are being done more efficiently. The reality is most chatbots today are still largely human coded and driven. Like the branching logic is largely--

- [Kevin] It's still a manually tasks to set up and implement, yeah.

- It won't be that way for long. But in most cases, the bots aren't that intelligent I would say.

- [Kevin] Sure.

- But they can drive massive efficiencies using, they're using types of AI. Like they're using natural language generation to like pump out text that you would read or a voice that you would hear.

- [Kevin] Yeah.

- They're using natural language processing to understand your words and your questions. So it's not that it's not AI, it's just not machine learning.

- We're still in the early stages so to speak.

- But we're at that surface level AI I would say.

- Sure. Now I know you brought up a really terrific example of okay, a retailer whose trying to sell TV's and how we can get all their descriptions up and running. I mean do you have any like real life examples or any companies like regardless of size, small or big, that seem to be using like AI from a marketing perspective, like really well. Are there any examples out there?

- I don't know that people at Cosabella. I haven't talked with them.

- Sure.

- But in like four or five of the AI companies we've featured, they come up as a case study. Like, whomever is at Cosabella, whoever is the CMO, like, I would look at what they're doing.

- Sure, yeah.

- So they had a quote in a publication in like October of 2016, they had moved Albert as their media buying tool and said like, we'll never go back to human buying media. Like it would never make sense.

- Sure.

- So they seem to be doing some really interesting stuff. We did a case study on Delta Faucet that was using a company called One Spot to do recommendations of content but also then intelligently targeting people off-site. So it's an Austin-based company, One Spot is an Austin-based company that's doing it. Coca Cola, there was a quote from one of their heads of global marketing that said that they see the future of story telling being done by AI which is a pretty provocative thing to say. Because most people believe today that qualitative narratives can't be told by machines.

- They're uniquely human, yeah.

- Yeah. And I generally agree with that. My opinion changes as times move on a little bit. But they've said in the article like we're experimenting with ways to use AI to generate social media, to pick songs, to do all kinds of things in their advertising.

- Sure.

- And if you're a company that has an ad budget, I have no, a billion, I don't know Coca Cola spends.

- Yeah, right, yeah.

- Throwing a hundred million at something experimental. You know, why not.

- Sure why not.

- And again, a lot of the challenge or the opportunities in AI is more of someone just committing the time and talent and money to do it.

- Sure.

- It's not that it can't be done. There are some like going back to my original thing of how do we build marketing strategy with AI.

- [Kevin] Yeah.

- I'm still not sure how that happens. It can be done and I've mapped it out how it could happen. But I think we're still at least five to 10 years away from someone actually being able to do that. Again, if you put enough time and money, maybe you could.

- Sure.

- But most of these more narrow use cases, really is just a matter of someone having the right people to work on the problem. Most of them aren't that hard for people who can build using machine learning.

- [Kevin] Yeah.

- It's just they have other things they can build, that they can make more money building. So why build that?

- Sure, yeah.

- Actually, there's pause there. Here, this is actually a good question so from Carla. What are some of the limits you see in AI? Like currently?

- That's a great question.

- Or even potentially.

- Yeah and so I referred to, on the AI Institute site, the Marketing AI Institute site, we've done 32 spotlights. So we've interviewed these 32 AI powered VC-funded companies. Combined, they have 586 million in funding. Seven of them don't publicly share their funding so there's actually probably much greater than that.

- [Kevin] Sure yeah.

- We're tracking about 511 of them that combine have 2.9 billion in funding. So there's just like all this money pouring into this space. And so one of the things I wanted to know was this exact question. So when we do this spotlights, we ask the same eight questions of every person. It's usually the founder or like lead engineer or something like that. And one of them is what are the limitations of AI. The biggest limitation is, in my opinion, is probably that general intelligence doesn't exist. And by that what I mean and you and I talked a little bit about this beforehand is what you see in the movies, like these intelligent robots that can talk and have--

- [Kevin] Sentient like.

- Yeah, sentient beings.

- Yeah right.

- That have consciousness and they can have all these senses. Likes it doesn't exist. There's no AI platform. There's no like one company you can go by that just has all these AI tools baked in. So, the machines have to be trained or train themselves to do very specific things. And so the largest limitation is you can only build it to do a narrow task. So pick like email, subject line writing. Or send time optimization. Or recommending content on a site, like. You have to build it. And often times you have to have a lot of data and it has be to be really clean, good data. So limitations are you can only build narrow. Data is often times critical. So it usually limits it to larger organizations that have a lot of data.

- [Kevin] Sure yeah, right.

- And then you just need the time and resources to build it. And then you don't just build it and forget about it. It usually requires a massive amount of human training and ongoing support.

- [Kevin] Yeah, right.

- So like you'll have, a lot of people will hear AI and think wow I can just magically fix this thing not realizing that there may end up being quite a lot of effort going into training in it.

- [Kevin] That's a great point.

- Yeah, so that's a really good question.

- Yeah, I guess, so if I'm a marketer right now, right. I mean obviously, we've touched a little bit about like education and where like certain like things to digest from content and like learn but I mean how do I started? Maybe I don't have a budget like Coca Cola, right. So like how would you recommend staying ahead of AI?

- From a marketers perspective, I think first you just have to commit to learning, to not being overwhelmed by the whole idea of it. To realizing that you don't have to go as deep as I've gone to like really learn and like wake up in the middle of the night thinking about this stuff. You just have to know what's possible. The potential of AI. And educate yourself on that basic information. When you're ready to actually try it, pick a single use case. So again, if you're a HubSpot customer, there's a number of tools like Seventh Sense I think does email send time optimization. They're integrated in HubSpot. Predictive scoring is built into HubSpot. The new content strategy tool has, there's some machine learning mixed in, that helps you actually build it. Like pick those so you can start to experience it and realize what it does and how it works. And start getting more comfortable with it. and then take your, take a slice of your marketing department. Say a month or a year of data, and look at which things you spend the most time and money doing that are really time intensive and that maybe are really inefficient or they have a bunch of data involved. And say, could I intelligently automate this? Is there a tool to intelligently automate this task? So for us, it was Google Analytics reports. We would spend seven, eight hours a month on each client producing a narrative of what happened on their site the night before. We integrated natural language generation, had the templates created, and we can instantaneously write 20 client reports.

- [Kevin] Right.

- So.

- So really it's just identify okay, what are the most time intensive tasks or tasks that I rely heavily on. Like the review and like analysis of data. And then look to bring in some automation around that.

- Yeah like right now, so I run an agency in addition to the AI thing. And I have a CO report I do every month that has probably 40 different metrics in it that come from three different sources. And so I would normally go through the rows and I would like write the narrative each month or each quarter. I'm just gonna train the machine to write it. So I'm gonna write it with different variations. Here's a monthly, here's a quarterly. Here's an annual report. And then I don't ever have to write them again. I just go in and edit them and add a couple of insights.

- So just to confirm right, so kind of as we've talked about the process it's one, making sure you have clean data and organized data. But then it also requires that template. Like this is how this is usually constructed. The machine can then go ahead and build that out at scale.

- Right, for story telling, yes.

- For story telling, yeah. I mean have you gotten any feedback or I guess like I'm interested in learning more about like the overall impact on marketing performance. Like has there been like a higher sentiment in the quality of reports and some of the story telling. Like pieces that you put out so if you could speak to that--

- Yeah, we've told those, we've done a few case studies on the Marketing AI Institute site. And told the stories through the tech companies that are doing it. And they all claim, yes

- [Kevin] It's more effective, right.

- It's hard to argue the efficiency side. Like if you get it right, certainly that can do it. Again, email send time optimization is such a simple example so I'll stick with that one again. But if you, like let's say you send a weekly email newsletter to 10,000 people and right now, you send it on Thursday's at 2:00 p.m.

- [Kevin] Yep.

- Those 10,000 people may be across like five different time zones. A portion of them may actually do most of their email reading after 10:00 p.m. on Saturday nights because kids went to bed and they finally catch up on stuff. A human is never going to learn that. You would never dedicate enough resources to continually segment that database and keep personalizing down to that individual level. And their preferences change and it's like how are you gonna know?

- [Kevin] Sure, yeah.

- So it's just the perfect example. Like there's no debating that if you have a machine that is always getting smarter and always improving it, it's going to have a positive effect like--

- The fact the people consider all of those variables just, yeah right. It's something that a human honestly can't do at scale.

- Right.

- Right, yeah. It looks like we have a couple of questions. Let's get in here quick. When do you thinK AI will be a part of marketing strategies for SME's? With not as many resources as the top. That's a great question from Andrea.

- All right so the, yeah there's a couple levels of this. So the first is, it's gonna be really hard because they usually lack the data necessary for the machine to get smarter. However, if someone like HubSpot were to anonymize 37,000 customers, data sets, and be able to recommend strategy or content topics based on a much larger data set.

- [Kevin] Be able to reference the data set, yeah.

- Or like a good example would be, social media send, like the strategies around when you publish and how you publish.

- [Kevin] Sure.

- Rather you using it based on a study that happened at a fixed period in time and then, you share that same recommendation with everybody that downloads the report or attends the webinar.

- [Kevin] Yeah.

- If instead each portal is alawys learning and optimizing send time and what to put in that, which hashtag to use, what colors to use, what images to use, and always learning every time you post, it figures new things out.

- [Kevin] Sure.

- Those are the kinds of tools, that SME's could start seeing much sooner. Figuring out what to do with a $100,000 budget, no time in the near future, it's not gonna happen.

- So you basically have to lean on organizations, maybe it's within like a tool like HubSpot right. So analysis of data at a much larger scale and then being shared on the smaller scale for the individual. Companies or users of the tool.

- And I think that's a good, going back to your question about to get started, the other thing I always recommend to people is talk to your existing marketing technology stack. Like go to your automation, your CRM tool. If you're with HubSpot, great. Find out what else you could be doing. What are the more intelligent, so if you're not using predictive scoring yet, is that a fit for you. If you have enough data to look at, to start using predictive scoring. As new beta tools start coming out. To be able to start experimenting with those. It's probably going to be the best play. Because again, there aren't that many people in the world capable of building these solutions. Most of them get acquired for insane evaluation. Like I was taking a machine learning class through Coursera last year, the University of Washington I think. The professor's company got bought by Apple in the midst of that program, it's like a six week thing. For 200 million dollars, I think they had six employees. So it's like, they're so uniquely capable of building things, that one of these massive companies is gonna come and pay a whole bunch of money to get, so even--

- [Kevin] It's wildly desired.

- Right, so even if you start going to find like 10 or 15 tools that are all AI powered, if they're actually any good, there's a really good chance in the next two years, they're gonna get absorbed into one of the big platforms.

- [Kevin] Interesting.

- Yeah, so it's gonna become a platform--

- That means it's available, right.

- Right.

- So I mean is it a safe bet or is it like just a confirmed right. So if I'm a SME on a smaller scale or like a smaller team, I'm looking to leverage AI. Well one, I should review my current technology stack and see what AI levers are already available to pull.

- Right.

- Take advantage of those.

- Yep.

- Yeah.

- Yeah, for sure. And then, again, there are a lot of tools and so the 32 spotlights we've done, part of that is to try and like help people figure it out. Like if this is something you're trying to solve, here's a company that does it. Is there, you know, crunch based funding. Here's there G2 crowd rating. Like, do the work yourself.

- Sure.

- We're not recommending this tool. It's just we're trying to put the spotlight on ones that seem legit and have like a user base.

- Again it's like to keep pulse on where AI's going and where machine learning is going.

- Right.

- Yeah.

- 'Cause you can't go into like G2 crowds, they show me AI powered marketing tools. Like, I don't think they have that category. Maybe they should.

- I'm not sure.

- But yeah, I mean you still go in looking for email marketing or automation or CRM and you don't necessarily know which ones are intelligently automated and which ones aren't.

- Sure.

- 'Cause a lot of these companies are struggling with how to explain AI and how to position what they're doing as an organization because it's like the sexy thing to slap on your name right now.

- Yeah, right.

- But it doesn't necessarily mean marketers know what it means or really value it. 'Cause they may not know. So there's a ton of education going on. Kind of like the early days of inbound marketing. Well like what is inbound marketing, why should I care? There's a bit of that plus AI's just so hot right now you almost start tuning it out a little bit.

- Right.

- That's why I take the approach of like it's just gonna be so present in your life as a marketer, you're probably not gonna know.

- And it's just gonna feel like the new normal versus like this whole new thing that you're gonna have to learn how to handle.

- Yeah like if marketing software isn't intelligently automated in the next three to five years, they won't be in business.

- Sure, right.

- Like it's every marketing software company is going to have AI.

- Right.

- Baked into it.

- So as a marketer, you're not looking for the AI tool or the AI company, it's like okay, let me find the company that differentiates itself or the tool that differentiates itself by incorporating AI. Like what's going to be the leading email tool, well it's the one that incorporates AI for performing.

- You still have business goals and business challenges and that's what you're trying to solve.

- Go for it, yeah.

- You just may be able to find tools that do it more intelligently and more efficiently that drive performance at a higher rate.

- Guys, we still have plenty of time left, if there's any other questions. I mean they've been terrific questions thus far.

- Yeah, it's awesome.

- So please keep them coming. Just one question, probably more on the lighthearted side but what aspect of AI machine learning, deep learning, what is most exciting to you?

- I guess part of my discovery process has been what is the future of the agency. So I look at what we do on a daily basis, and again, we're primarily B to B. As you said, we were HubSpot's first partner, so we're big in automation, and CRM and lead gen. Nurturing and qualification, and I don't know how many of those things humans are going to be doing in three to five years. So the thing that's exciting to me is there is this whole new world that is going to be artificially intelligent and we're, for the most part, it's going to enhance our knowledge and capabilities, not necessarily replace them.

- [Kevin] Yeah, right.

- But I also think a lot of what we do manually won't be done manually in the near future. And so the chance to try and figure that out and kind of re-define the path of where we go and where our clients go, that's cool. Like it's not trying to figure out how to just publish through the noise and get people to come to the site and subscribe. Like, the same stuff we've all been doing for 10 years and I think there's a better way to do it and I don't know exactly what it is yet but it's a lot of fun to try and figure it out.

- So if all these tasks are soon to be automated where is the marketer's time gonna go?

- Probably places we're not aware of yet.

- Sure.

- So, funny side story I did a mackerel level AI talk a few months back in New York and the first question I got was, are we gonna lose our jobs? And I said listen--

- [Kevin] It's a frightening prospect.

- Yeah, it's actually a legitimate question. There's different opinions. There's a study by Bank of America that says there will be a nine trillion dollar impact on the knowledge work industry within the next decade. That's a lot. Sales force says that we'll create 800,000 net new jobs in the service industry alone. Like, nobody actually has a clue. But this guys says, "Am I gonna lose my job?" And I look at this, and it's a tough question, but what I can tell you is there's a study in 2014, University of Oxford did the computerization of industries, two professors. And they looked at 702 professions and the likelihood, the probability that they would be computerized completely. And I said, all I can tell you is you don't want to be a telemarketer. And he goes, his face drops. He's like, "I'm a telemarketer." and I thought he was messing with me. And so I just kinda looked around the room, and I was like nah really, and he goes, no I'm serious like, are you saying I'm going to be out of a job? What am I gonna do? 'Cause he was like later in his career, and I was like oh my god, I was not expecting this. And so then I went into this explanation, like take any great technological advance, and there are people a lot smarter than me who believe the impact of AI will dwarf the creation of the internet itself. Like what's going to happen to the economy, to, forget just marketing, think about society, the economy and, once everything is intelligent and automatic, what do we do? And so anytime there's been a disruptive force and a technological advance, careers and jobs are created that you can't even fathom because the tech doesn't exist yet. So rewind 10 years ago, the iPhone doesn't exist. When I started animation in 2005, Facebook wasn't available yet, outside of college campuses, Twitter didn't exist yet. Blogs were the thing the geeks did, like no one else was blogging. The iPhone wasn't around. So if I just look at, actually ironically, today's the 12 year anniversary of my agency.

- [Kevin] Happy anniversary. Yeah, congratulations.

- And so I think back 12 years ago, what we were doing, and I didn't have any of these career paths, like the things we do for clients today, didn't exist 12 years ago.

- [Kevin] Sure, yeah.

- And so I look at it and think, it's just going to open doors, like yes it'll replace things that honestly humans probably hate doing anyway. Unless--

- Way to look at it.

- Yeah, like most people don't wanna sit there and crunch Google Analytics data and trying and figure out what it's saying. Most people don't wanna take spread sheets and write 50 product descriptions. So it's going to intelligently automate that, it's probably automate a bunch of things we can't think about yet. But it's also probably gonna create all kinds of really fascinating opportunities that don't exist and it's probably going to give you all kinds of knowledge you never had. Like you're going to be able to perform at a high level. So the way I look at it for marketers right now is, it can be your competitive advantage, within your career. For your brand but also you as an individual. The more you grasp it and what it's capable of doing, and the more you start finding ways to do things more efficiently in driving performance, yes you may replace some of the tasks you probably hate doing anyway, but you're going to be a better marketer. And so it's going to give you super powers.

- Yeah. It's like repositioning for the individual.

- Yeah.

- I could be able to say I know where this is going, what the power, what it can help me do.

- And you can put your head in the sand and pretend it's not gonna disrupt your career and just go about your business. I think they'd probably like embrace it because a lot of people are gonna be afraid of it. And the more you're willing to just kind of open arms, like alright, explain it to me, let me learn more, let me look at some tools. You're gonna set yourself apart and most companies are gonna. You think they're looking for people who know marketing automation today, wait three to five years when they're trying to find people who can intelligently automate stuff.

- Right, right exactly. So it's interesting right. So I mean, you mentioned some of the things the agency does now, you could have never even fathomed because some technologies didn't exist yet. Mobile app development. Social media sites--

- All on Palm Treos in 2005.

- Who knows what the, A, the service softwares are gonna be, but two, like what roles and what responsibilities look like in a marketing department. That's really interesting. Let's transition back. Is there a way over to our comment section here, is there a way to remove the need for tons of data to make use of AI? If my goal is to personalize, why do I need tons?

- 'Cause the machine needs to learn. So the data sets are what it uses to better personalize.

- Sure.

- So it can try and shortcut it with lookalike analysis and things. So if Kevin shows up on the site, Kevin came from these channels, we know this demographic data about Kevin. He probably will like things like person will have a year of data on, again, no human is really wanting to go there and say figure all this out and draws these correlations. The machine can do it instantaneously and it can do it at scale to like 100,000 visitors. So that's why data is needed. You want the personalized data. And so you need interactions overtime. The more of those you have, the more data you can capture. Now there are ways again to take anonymized data sets and try and apply it to those small businesses, and then have a ton of data. It's tricky.

- Sounds kind of like Watson at the very beginning. Right, when we were talking at the top of the hour. Is the volume there to help just improve our confidence level on the decision?

- Yeah.

- More or less what it is.

- Yeah, it's the probabilities of the being correct. So like I mentioned earlier, so this is a really cool example. If you wanna again have your mind blown, do a Google search for AlphaGo Zero. Three weeks ago Google announced in the Nature Journal, that they had created a system called AlphaGo Zero. So about a year and a half, two years ago, you may have heard about AlphaGo, and the deep mind team at Google had built and AI system that beat the world go champion. Lee Sedol I think was his name.

- Yeah.

- Go is like a 3,000 year old game, insanely complex, I think at any given time there's about 13 million possible moves. So far more complex than chess.

- Sure yeah.

- Most people didn't think AI would be able to win at a game of go for at least another decade. They thought the technology wasn't there yet. So Google did this. And they did it about two years ago and it was kind of a Watershed moment within AI world.

- Sure.

- Fast forward to three weeks ago, they built something called AlphaGo Zero, and what AlphaGo Zero did was all they did was gave it the instructions of the game of go and in 72 hours, it learned, by playing millions and millions of games against itself, how to play go at a level that looked alien to go champions. It then beat its predecessor AlphaGo, a hundred games to zero.

- [Kevin] Jeez.

- In 72 hours it learned this.

- [Kevin] Yeah.

- And then they shut it off, so they don't even know how smart it could have gotten. So what Google is actually trying to do, so it wasn't the first, I don't wanna overstate this, but one of the first publicly known instances where something from learned from scratch on its own by playing itself to win that game that in as complex as it is and then they literally did interviews and the go champions are watching the games playing these alien moves. It's making moves nine steps in advance. Like a go champion looked like why in the world would it do that? Nine turns later it's like oh my god.

- [Kevin] It did nine turns later?

- It was predicting nine turns out what it should do now. And they're just like, this is absurd.

- [Kevin] So it's like the growth is machines have the ability to learn and then we can kind of, we give them some base knowledge to then learn from. But then the advanced layer on top of that, which is AlphaGo Zero, is okay, have them start from scratch, learn themselves and that's what we're seeing now.

- And again there's, there's technology, you get into the surface level stuff, you read what a bunch of marketing ad companies are doing and you're like, wow that's really advanced. The stuff you'll see publicly is nowhere near what is actually happening behind the scenes that may not find it's way to the marketing industry for a decade.

- I mean, timing wise, it has to trickle all the way down now.

- This isn't new stuff. It's literally 30, 1997 a guy named Thomas Pefferty hacked NASDAQ, and by hack meaning, took the feed coming from NASDAQ and taught a machine how to make the trades. Then got in trouble because they said that the trades need to be entered in, like the machine, so he built a computer, a robot that did computer vision to read the screen and then punch it in.

- Actually punch it in.

- With like a machine punching it in, 30 years ago. Like today, 70% of all trades are made by machines. And I look at it and I say that is far more complex, far more variables then how to spend $100,000 or a million dollars across a hundred channels. Like, yes, it's really hard for humans, but there are AI tools built to do things far more complex in other industries.

- [Kevin] Sure.

- And the promise of things like an AlphaGo Zero that probably won't be commercialized for quite a while.

- For a long time, yeah.

- But at some point it's all gonna kind of. So my thing is like, just embrace it. Like, it is crazy and if you read the wrong books, you can't unlearn the stuff you'll then know and your mind can really go down some dystopian paths, but in the near term, focus on the opportunities from others.

- Well, in the theme of embracing it, here's a question from Caroline, so I mean, I know we've named off a few people, but let's just recap, who are the top people to follow in AI?

- Yeah, that's a really good question. So.

- What does, probably just a clear list look like?

- Some of the people that come to my mind, like Mark Cuban is a fascinating one, people know him, he owns the Dallas Mavericks, he made a lot of money selling Broadcast.com to Yahoo! in the early 2000s. Shark Tank, I think he only invests ad companies now. Like he is a huge advocate of AI, a huge believer that's gonna change everything. So he does a lot of interesting stuff, invests in a lot of interesting companies. Demis Hassabis is like from my perspective, one of the true brilliant minds of today Elon Musk, there's this fascinating dispute I guess or differing of opinions between like a Mark Zuckerberg who's massively invested in AI, like Facebook is huge into AI and Elon Musk believes we may be going down a really scary path with AI. So there was actually a thing where Mark made a comment about, he thought Elon maybe overstating the concerns around AI.

- [Kevin] Really?

- Elon Musk, I remember it was a Tweet, I think it was a tweet of an interview. He said I've spoken with Mark about AI, don't know that he knows what he's talking about.

- [Kevin] Really? Really, oh really?

- Yeah so.

- Controversy.

- Yeah, Elon Musk started an organization called OpenAI, I think Stephen Hawking, maybe Bill Gates, some other guys are invested in that, and they're trying to, the challenge moving forward, like the big picture is you gotta program ethics into it and morals and like it can go down a really crazy path.

- Guard rails maybe. Yeah, right.

- So if you want to like, the crazy stuff, the really big mackerel level AI then you're looking at Elon Musk and Zuckerberg and Stephen Hawking and Bill Gates and the Future of Life Institute and O'Reilly and like some of the ones I mentioned earlier.

- Sure yeah.

- If you're looking at marketing it's kinda hard to find, there aren't a ton of people in the marketing space. I mean, HubSpot's doing some stuff now with AI, they're starting to talk about it more. Some of the bigger platforms are starting to talk about it more. I know Dreamforce is this week. Salesforce has bought like nine companies for a couple billion dollars.

- [Kevin] Sure yeah.

- So Bennyhoff's big on AI, believes it's the future of everything. So yeah it's more at this point, it's the bigger picture AI experts and professors that are talking about the most interesting things. And then you have to connect the dots.

- You had to opportunity to plug your own Twitter there and you held back.

- Yeah.

- Yeah and Paul. Guys I think we're right up against the time. So I think we'll call it here, but Paul thank you so much.

- No, thank you.

- I appreciate it.

- It was fun.

- I hope you guys enjoyed and we will see you again soon.