Christopher S. Penn of Shift Communications – The AI Power Hour: Artificial Intelligence and Machine Learning

Christopher S. Penn is an authority on digital marketing and marketing technology. A recognized thought leader, author, and speaker, he has shaped three key fields in the marketing industry: Google Analytics adoption, data-driven marketing and PR, and email marketing. Known for his high-octane, here’s how to get it done approach, his expertise benefits companies such as Citrix Systems, McDonald’s, GoDaddy, McKesson, and many others. His latest work, Leading Innovation, teaches organizations how to implement and scale innovative practices to direct change.Leading Innovation, teaches organizations how to implement and scale innovative practices to direct change.

Christopher is a highly-sought keynote speaker thanks to his energetic, informative talks. In 2015, he delivered insightful, innovative talks on all aspects of marketing and analytics at over 30 events to critical acclaim.

He is a founding member of IBM’s Watson Analytics Predictioneers, co-founder of the groundbreaking PodCamp Conference, and co-host of the Marketing Over Coffee marketing podcast.

Christopher is a Google Analytics Certified Professional and a Google AdWords Certified Professional. He is the author of over two dozen marketing books including bestsellers such as Marketing White Belt: Basics for the Digital Marketer, Marketing Red Belt: Connecting With Your Creative Mind, and Marketing Blue Belt: From Data Zero to Marketing Hero. His new book, Leading Innovation, debuts in 2016.

Listen to the Podcast

In this special episode on artificial intelligence Christopher Penn talks about Machine Learning and:

  • The Four Elements of Artificial Intelligence
  • Automation’s Political and Social Impact
  • The Fifth Element of Sapience
  • Data Science Rising

Read the Transcript

John J. Wall: Hello and welcome to Stack and Flow. I’m John Wall. Today, our guest is a gentleman that I’ve known for quite some time, Mr. Christopher S. Penn. He’s my co-host on Marketing Over Coffee. Welcome to Stack and Flow, sir.

Christopher S. Penn: Thank you very much.

John: Sean Zinsmeister is also here with us. Actually, I’ve talked with Chris a number of times on AI, so we have some past experience with this. Sean will be taking the lead on a lot of the questions here, especially bringing to it from a predictive angle and some of the things that he’s got going on. Sean, how’s everything going?

Sean Zinsmeister: Yeah, it’s going good. I thought that this would be a great show to almost do an AI power hour, I’m calling, with Chris because I know that this is a topic that he and I chat about and have a lot of conversations about that I thought it would be interesting to have the audience be able to eavesdrop on the conversation. I think before we jump into the meat of it, it would be good to get some baselines which is, Chris, how are you defining artificial intelligence, AI, today? How do you take a crack at the definition? Let’s start with that.

Chris: There’s four buckets. Imagine four concentric circles or squares. The outermost layer is the algorithm. The algorithm is the foundation of everything and everything that we talk about and do in the space is algorithm based, from setting up … I’m going to set up a junk mail rule on my outlook so that emails from a certain vendor don’t come in. Algorithms are first.

The second inner circle is artificial intelligence. It is using machines to simulate cognitive intelligence processes that humans do. Anything that you say is part of human intelligence is part of that and keep in mind, if you think about kids, babies, they have sensory inputs then eventually they develop language and then they develop cognitive brain functions. AI encompasses the evolution of machines to be more human-like.

Inside of AI are two additional buckets. There is machine learning, which is the implementation of these algorithms in such a way with data so that you get a simulated, almost human-like result. There’s two subcategories of machine learning. The supervised learning where imagine teaching a toddler this is an apple and you hold an apple. This is an apple. You do it a bunch of times and then you put a bunch of fruit in front of the toddler and say, “Find the apple.” That’s supervised learning. Then there’s unsupervised learning. You say, “Here’s a bunch of fruit. Figure out what goes with what.” Round ones might go together or red ones might go together and machines learn how to categorize based on this unsupervised learning. That’s machine learning, which is a subset of AI.

Finally, at the core, the deepest level of this is what we call deep learning where you are chaining machine learning tools, techniques, principles together in long chains to have human-like or greater than human-like results. This is where we see examples like machines being able to do things better than people. Being able to read lips better than people. Being able to translate languages better than people because you’re taking all this machine learning things and just chaining them together to create an AI supercomputer of sorts per these different processes.

There is one layer deeper than that that doesn’t exist yet and that is general purpose AI. Right now, all artificial intelligence, all machine learning, all deep learning is purpose built. I want to recognize photos on Instagram. I want to do speech recognition. I want to do predictive analytics. General purpose AI is SkyNet, is the Terminator, is Data from Star Trek. It is an artificial intelligence that has no specific intended purpose. It is a general purpose sentient computer and that’s not here yet.

Sean: No, that’s right. I love that framework that you set up and then I want to break it down actually even further because there is the idea that I like challenging a little bit, which is how we define intelligence. I’m going to bring up the old John Searle Chinese Room experiment in which how much are the computers today really understanding … ? They can do as you said. They’re doing the character recognition and being able to put the translations together, but do they really understand what it is, do they actually understand the language base in terms of having that level of automation? How do you go about understanding today that strong and weak AI dichotomy that’s set up at the Searle Chinese room experiment?

Chris: There’s a couple of different ways you can look at this. Intelligence by itself just means the ability to acquire and apply knowledge and skills. I would argue that when you by that broadest definition we are there because, obviously, a chimpanzee has intelligence. It may not have as much as us, but it does. It has a level of intelligence. Where I think your definition is going is towards instead of intelligence it’s sapience, which is knowledge, experience, understanding, empathy, common sense, insight and that is different. That’s wisdom and, quite frankly, there are a number of humans that haven’t figured that out yet.

Bearing that in mind, we are not at the point where machines are sapience and it will be some time before we are there because, quite frankly, we don’t even know how we as human beings achieve sapience. We have some general ideas and as the complexity of our nervous systems or our neural networks in our brains sapience may arise from that level of complexity. We’re getting there because as you add more and more layers to deep learning, if you think the structure of the human brain, the human brain is a massive parallel computer that’s very slow. It is electrochemical and it is the complexity of those millions, and millions, and millions of interconnections that give rise to sapience.

At least that’s what you think is happening right now, whereas machines are serial processors. Your laptop has eight cores. It essentially has eight threads to be able to process stuff, whereas your brain has millions of them, but however, the machine, those eight cores can process many, many, many thousands of times faster than your wetware can in your head. As deep learning and massive parallel computing and all these technologies become more affordable and more scalable and things, we will get to a point, I believe, where sapience arises out of the complexity of those systems.

Sean: One of the things that it makes me think of as we start to … We’re building this human AI conversation just from what we’re talking about here. One of the things that I always think about is education and how that starts to change because you’re right, there are certain things that we have to accept that there are certain workflows that are going to be made redundant because the algorithms can just do them faster, better quality and you can set those up and repeat.

I wonder, if you look at it from an education standpoint, looking into the future, do you think this puts more … We actually talked about this with Jeff Marcoux a little bit on one of the past episodes. Does this put more value in things like humanities education where, essentially, you go into college to leave with more questions than when you went into? It’s that one of the things that we may see, perhaps it’s many years away as AI continues its relentless march. Do you have a feeling that there might be a renewed investment in that level of education?

Chris: I would argue that, and this is literally a PowerPoint slide in one of my talks on the topic, anything that is a template today, that you do with a template today, you do without a human tomorrow. That’s where AI will the greatest impact. Now, think about this in the education process itself. How much stuff is templated? How much stuff looks like the same thing you’re doing over, and over, and over again? You’re taking exams. You’re studying the thing.

In theory in higher education, you’re building complexity in your own neural network to understand things greater, but the reality is, you pull open any school textbook, you pull open an SAT exam, what is all that? That’s a template. From an education perspective, we’re not building people for the post-AI world. We are still stuck in an educational mindset of 20th century industrial factories. The fact that we have grades still in school and exams is manufacturing. We have a batch of products and have a QA test for that product. That’s not the world anymore because again, the machines can do all that faster, better, cheaper.

There’s an example recently in … I forget which magazine it was … A Chinese manufacturer with smartphone components had a factory of 60,000 workers. They replaced them all with robots. They now have 600 workers in that plan. Those 59,400 other people, no job. The factory is 283% more productive with a 93% lower defect rate. Why wouldn’t you do that everywhere possible? Now, for us as human and particularly for those of us our parents thinking about what does the future look like for our children, if we want to ensure that our children have any employment or future whatsoever, we need to equip them not with the humanities because again, education is templated, but with the ability to actually think non-linearly.

That means yes, you need to have an appreciation of art. You need to be able to write. You need to understand poetry, but you also need to be able to understand statistics. You have to be able to code to some degree or think any coding like fashion and be able to cross disciplines. That’s the heart of everything for the future. Multi-disciplinary, cross-discipline thinking so that the machines are not at a point yet and not until we get to general purpose AI where they can simply create from whole cloth. They can’t do that today. They’re getting closer and closer. We are at a point now, for example, in the marketing world where we have programmatic creative to go with programmatic advertising where the machines are assembling creative, but they are still drawing on the library of existing stuff. They’re not sitting down with a box of crayons and paper and just investing something from a whole cloth.

Sean: I think that when you look at this conversation, you have to step back and ask yourself, “Why is this happening? Why is over the last few years … ?” Maybe that’s over simplistic, but why now? Not from a technological standpoint because I think that that’s already been argued to death which is like we have better processing. We have the rise of cloud computing and, obviously, the access of data and some of the, what I would argue, democratization of data is certainly a part of it. Are there cultural nuances that have led us down this? Is this again the march of the industrial age moving to modern times? Why now, Chris?

Chris: I’m going to venture in a territory that is potentially sensitive here, so we can always edit this out later. A lot of it has to do with politics and society. The way in an industrial labor economy, there is a natural tendency towards income and equality. There’s a natural tendency for capital to flow upwards. Some of it still has to flow down because you have pay workers and once you pay those workers, they go out and they do things like buy houses, and buy kids, and eat, and stuff like that, but the smartest folks or the wealthiest folks have figured out is that automation, and AI, and robotics, and all these things reduce the amount of capital has to flow downward.

When you have a machine other than the maintenance of that machine and the amortization capital, all of the benefit of that machine creates capital that flows only upwards so you don’t have to flow any capital downwards. If I have a factory of 60,000 workers, I got to pay them. Even when I pay them really poorly, I still have to pay them. When they’re all gone, I’m getting profit for myself without a labor cost and that means that my profits are significantly higher.

From a socioeconomic big picture, the fact that we live in an era where there’s already a lot of income inequality, this trends are happening because the folks who have the income are saying, “I want more of it and I want to pay less of it down the tree,” because you want more income. That’s the rational outcome of a business. Whether or not that’s just, or equitable, or fair, or whatever, is a social conversation, but the reason that this is happening is because those folks who are the top want to stay at the top and want to amplify even more the investments they’ve made and AI is the way to do it.

Sean: We can’t shy away from that where people who start to look at their business and are running P&L sheets. I mean, you can look at all the marketing programs that you want. The most expensive piece on any P&L is going to be human capital expenditure and I think that you’re absolutely right. One of the things that I’m looking at is I don’t think that you can avoid the political discussion right now because what people are going to say is that innovation, yes, it breeds disruption. It also will generate new job creation.

I think, unfortunately, the job loss, if you hold that in your left hand very high, is much greater than the job creation that’s going to happen through AI. It’s curious where I have to ask, do you think that education is the silver bullet there? Is that the key in helping to balance this out? Maybe it’s a retraining of worker’s program or something like that, but it has to be a rethinking of education in order to balance that out or, to throw something else at you, is Bill Gates on the right path where he’s thinking about these ideas of robot taxes? What do you think of that?

Chris: Robot taxes is an interesting idea for the folks who are currently in positions of power and exceptional wealth. They’re probably not going to want to do a whole lot of giving away more of that wealth particularly to the government since the government is not a really efficient consumer of revenue. Education is not the way out of this because you can retrain people, but as fast as it … Humans are slow learners. Compared to machines, humans are very, very slow learners. By the time you retrain someone to learn how to code, for example, you can automate the process of coding because coding is very repetitious.

You fundamentally have to look at one of a couple of outcomes. One, you have to look at something like universal basic income and imposing simply a sense of confiscation of wealth at that 1% level to pay for all of that and you can make the argument that we already have that to some degree because we already have a lot of structural unemployment, with social programs, essentially, are that to some degree, or the other which is so far out there that no one talks about it, no one will talk about and you probably shouldn’t talk about it, is we just have too many humans.

How do you reduce the number of humans by 99% then the robots providing the resources and things for a much smaller pool of humans who are balanced with the ecosystem around them because if you don’t need the consumer, other than for your new Rolls-Royce or whatever, then essentially, all those people who are out there are working against you, the one-percenter. They’re taking away that you could be using. It goes back to your point about the P&L sheet. If you have a million dollars in employee expenses and you have $7 million in revenue, obviously, you’re doing very well, but if you could then get rid of all the employees, you could keep that $1 million in salary just for yourself. Why wouldn’t you do that as a rational business actor? Forget being a decent human being. Just from a numbers perspective, why would you not do that?

Sean: No. You’re spot on and I’m looking at numbers where … Take the articles that have come out recently around Otto. For those who don’t know, that’s the self-driving trucks company that’s sold into Uber, which is exciting, but what they’re projecting is, what is the numbers, 2.2 to 3.1 million jobs stand to be threatened just by automated vehicles alone, but there’s maybe that ray there where I want to talk about human-assisted AI because I do think that there is some home in some of these because when you look at these interviews from the founder of Otto, he’s saying that there’s always going to be a human in the sidecar.

If you look at way most that are Waymos that are driving around, Mountain View Waymo is Google’s self-driving car program for those who don’t know, how much of this is going to be human-assisted? Obviously, there’s risk management there where you still need the human at this point to help guide and those are specialized jobs. How are you thinking about human-assisted AI and what do you think the longevity of that is?

Chris: Probably three to five years from whenever you start the process. Think about this: from an insurance perspective, once you fine-tuned the AI for driving a vehicle and you can prove that the machine is X times safer, faster, more efficient, obeys speed limit, et cetera, no insurance company will want to ensure a human driver. It will say it is too expensive. It’s like owning a horse. Who owns and rides horses? People who have money because it’s a curiosity rather than something that’s necessary. There will be a point, probably down the road, where if you want to to have a car that you drive yourself, you will have to pay a significant premium on your insurance, and your cost, and your licensing to do that because the automated vehicles are safer, better, and so on and so forth.

With humans, the only thing that you can really say about us is that it is our unpredictability to some degree that could provide the spark that the machines need until we have general purpose AI. That’s probably a good 20, 30, 40 years away, but we will get to that point at some point because as our computational power increases, as our ability to string and chain these algorithms together gets bigger, and bigger, and bigger, we will have a machine that achieves sapience.

Is there a role in the near term but in that timeframe? Yes. You will have the truck driver who basically sits there and watches Netflix while the truck does its driving. At a certain point, once you get even three or six months into the pilot, you’ll look at it and go, “Why am I paying this guy 30 bucks an hour or 40 bucks an hour or a dollar a mile or whatever when the truck clearly had no or very few human interventions? Even if it had 1/100 of a percent of accident rate, that’s still something anywhere 10 times safe than a human, you would accept the occasional accident by the machine as a cost of doing business in the way that you accept that cost out of humans today.

Sean: There’s also some auxiliary benefits that I want to examine too that also intersect with the business world that I think AI is going to benefit and there’s been a couple of articles that are thinking about what does the next evolution of suburbia look like and what is happening with all the pain around transportation and some of the costs? I mean, on one hand, there’s certainly a political discussion around healthcare that is certainly one of the big causes for business. Probably not going to get AI to resolve that any time soon, but one of the things is that if you look at the Bay Area for instance, obviously, a very expensive place to live, as you actually move further away from some of the transportation hubs, those would be your Caltrain hub, your BART hub, et cetera.

Actually, if you run the numbers, for some businesses it almost becomes more affordable to put everybody in an Uber and shuffle them around versus being able to afford some of these office buildings and things like that. I think that there are really exciting transportation gains that businesses should be thinking about and actually creative models that they should be thinking about, which I think, to expand the landscape. I’m curious if any of that strikes a chord with you.

Chris: I would say it’s the opposite. To your point about healthcare, healthcare is extremely expensive. It’s something between 10 and 20% of that employee has base salary, right? The easiest way to reduce healthcare cost is to reduce the humans. You get rid of certainly the lowest level of humans in the organization, the folks whose jobs really are automatable. Hey, you’re an accounting person. Your job is to collate and run these spreadsheets. Guess what? We can automate that.

You will get to a point where when you think about it, the data center doesn’t have to be anywhere near the business. You may have a Silicon Valley startup, have three guys in an apartment or three girls in an apartment who are the management and the data center may not even be in this country. It may be in Northern Mexico. It may be Southern Canada and all of your workforce, the AI, is another location entirely where, I think, companies will find ways to get creative. It will be where do you put that workforce? There will be technological challenges like machine latency and network latency and stuff. Certainly, we already deal with that to some degree today.

There will also then be tax questions, business questions. Hey, my “workforce”, going back to the robot tax idea, the US wants to put a robot tax on. Okay, well I’m going to move my robot workforce to South Africa or wherever there is tax benefit to doing so. Because you have no concerns about human rights violations at that point, business will move to wherever the maximization of profit goes.

Sean: To take the conversation in a slightly different direction but also keeping it focused on business in the workplace, recent Wall Street Journal article really talked about AI’s place in the workplace and I think one of the concerns was at what point does AI become too invasive? In particular, they were talking about a number of solutions that business are early adopters of this, talking about building models for employee retention. Essentially, turn models for employees, if you will. Tracking systems and things like that. Where do we think that that threshold is going to be where things like okay, it does become a bit too Big Brother, a bit too invasive? Maybe we’re far away, but what thoughts do you have there?

Chris: It’s a margin question. There are some things, for example, like if you have an office space, yes, it is theoretically possible to construct and build a fully autonomous robot to vacuum the office. The ROI on that is almost certainly negative other than to say that you have this thing, but the margin on it would be terrible because it’s not a revenue driver, so you will still need a human for that thing. In terms of the invasion of AI into the workplace, it’ll be everywhere that has high margin that you can increase margins.

For example, there was a recent thing about how one of the major big brands out there has a new algorithm that is helping people do accounting audits and they showed this example. Your average accounting audits go millions of dollars. A couple of months, hundreds and hundreds of consultants come in and sift through piles of paper. There’s one company did this whole exact same audit. They were able to find 99% of the financial anomalies that the accountants did, found 30% more anomalies that the accountants missed and instead of taking it three months, it was done in 11 minutes.

Now, there’s margin to be had there because you’re no longer spending the human capital and time of that, and you can make your accounting more efficient, and get your books into shape faster. Any activity where you can reduce your expenses on humans or you can increase your margin is going to be up for grabs for AI because for the business owner, if you want to maximize profit, that’s how you’re going to do it.

Sean: Could you also make the argument too that the consumer in some of this is going to win? If we look at taking it away from the business environment to more of a B2C question where things are processed faster at greater quality, one example I’m thinking of to bring it back to this Otto self-driving trucks, you could make the argument that the shipment of goods is going to be faster because there’s no breaks to be taken, that food can be delivered in higher quantities and quality across in transportation of good. Isn’t there a lot of benefits for the consumers to be excited about with all of the disruptions happening?

Chris: In the short term, I absolutely think about this. For example, for those folks who may not be familiar with the geography of the United States, within the continental United States, there’s not a single location that you cannot get to within two days of steady drive. That means no breaks, no sleep, no meals, et cetera. We can’t do that. Robots, no problem. You think about a company like Amazon, their prime shipping, not always necessary available in all places and things, but now if you have autonomous vehicles that can get goods anywhere within the United States within two days, 100% of the US has prime shipping suddenly. You want 21-foot aluminum bleachers? Guess what? They will appear on a robot truck.

To your point about reducing costs, the price of goods will get cheaper, and cheaper, and cheaper because you have machines that can make them faster and better with fewer defect rates. You will have more selection, more quality. You’ll have customization options that previously were not possible because instead of having to retool machine and retrain a worker, you simply change the algorithm and now the machine can make the new thing. Mass customization has always been a market topic for a while, but there’s no reason why now you can’t have true mass customization because you could have a 3D printer. Guess what? If the AI can interpret your request, they will just turn it into the new thing. In the short-term, there will be significant benefit to the consumer. Everyone will get things that are better quality, cheaper and faster. The long-term question is, though, with what money will they buy those things?

Sean: On that note, looking at the vendors themselves that are driving this technological disruption, it seems that like when you go on Twitter to see that if you haven’t gotten your .AI domain name, you better go out there and quick because it seems that they’re getting bought and then launched in the next week. I wonder, what’s the danger of the AI stuff becoming a lot of hype and what advice do you have for people that are actually looking at some of these technologies to leverage and try to help their business whether it is to reduce cost or punch above their weight as it were? How should people be looking at this stuff to cut through the noise? It feels like there’s a lot of junk out there.

Chris: There’s a lot of junk and the thing I’ve been telling people is to start to learn the stuff yourself even if you fail miserably. You install our studio to try and learn supervised learning, you know, code everywhere, but at least by getting out there, trying these things, watching some of the channels on YouTube like Siraj Raval’s channel on YouTube about using TensorFlow and he does some hilarious stuff. I strongly recommend his channel. There’s a piece on how to get an AI to compose music and it’s wonderful. It’s a wonderfully disastrously wonderful thing.

The more understanding you have of the field, the more you can call vendors on their BS. This is something that I do at trade shows because I’m a jerk. I will walk up to a booth and say, “Our thing is AI-powered. We are an AI-based content marketing.” “Okay, what kind?” “What do you mean what kind?” “What kind of AIs? Is it supervised learning? Is it unsupervised learning? Is it deep learning? Are you using TensorFlow? Are you using SystemML? What is it?” Inevitably, the sales guy is like, “I don’t know which AI. That’s what they told me to say.”

That’s what you can do to extract more out of the conversation, to ask people, okay, what’s the mechanics of this? How does it work? Prove to me that you’re not just using the buzzword because it’s hot. Prove to me that you actually know what you’re talking about. Connect me to a software engineer. Connect me to a sales engineer. If you pull back the layers of the onion, sometimes you’re going to find yup, this company definitely knows what they’re talking about. They are using supervised learning. They’re using naïve bayesian in here, but they’re using just clustering over here and that makes total sense.

Other companies are going to say, “Well yeah, we have this AI thing which means that we just make one API call to Watson and that’s just to get a sentiment number and really there’s no actual algorithms that work. This loops back to where we started at the beginning. At the end of the day, all of this stuff is algorithms. If people can explain the algorithms that they’re using, they don’t have to give away all their trade secrets, but they at least need to be able to say, “Yes, we are using this type of supervised learning. Yes, we are using that type of unsupervised learning. No, we are not doing deep learning.” Then you can assess the solution and make a good choice for your business. If you don’t have that vocabulary yourself, build it. Learn it. There’s a number of different resources out there that you can use to get started in asking these questions and then try it out for yourself.

One thing that you may find is that from the major vendors, the Googles, the Microsofts, the IBMs, the Amazons of the world, there may be things that you may want to build internally, particularly if you have development capabilities, and then you understand it perfectly. You are in charge of it and you have control over it. I’m working on this one thing right now with predictive analytics where I can figure out with reasonable accuracy what the next 365 days of data look like based on a data input and it’s not perfect by any means, but it’s good enough most of the time that I feel confident saying to somebody, “Yes, this is legitimate predictive analytics. Here’s the outcome. Here’s what you do with it,” and I can explain the underlying mechanisms. If someone did that to me and say, “What kind of predictive analytics are you doing?” “Autoregressive integrated moving averages.” “Okay. You know what you’re talking about.”

Sean: Yup. There’s actually a wonderful resource and it’s a book that I actually highly recommend, especially for folks in business, called Data Science for Business. It’s by Foster Provost and Tom Fawcett. The reason that I love this is it really goes into thinking about problem-solving. Just like with marketing automation, it’s not necessarily key for all sales and marketing go-to-market professionals to be able to sit in an eloquent chair and be able to click around do that. I mean, that’s great. There will be technical operatives who can help with that.

However, you should have a good understanding from a strategic layer what’s possible and I think some of the things that you’re talking about, Chris, which is identifying that indeed, we have an AI problem, recession modeling is the greatest way to do this because we actually want a number outcome to understand how many people are using this service, for instance, or do we have a classification problem where it’s like we just need to know if this fits or it doesn’t, or will this one group be influenced or won’t they.

These are the types of basic concepts that I think are going to be really, really clear. Now, I think that they can really get in down and dirty with the data like you suggested and I think that that’s actually the best way to learn is to get hands on, but just being able to connect problem to solution and operate a strategic layer, I think that sales and marketing professionals will be well ahead of their peers if they start there, if that makes sense.

Chris: I think it makes total sense and I would say also that it’s going to a restaurant. You don’t have to be a master chef to know whether your steak is cooked correctly or not. You just have to know what the output looks like and, frankly, you as the customer may not even care. Did they use a broiler or did they use a charcoal grill? No. You know what the steak is supposed to look like. If it comes out well done and you ordered medium rare, they messed up. It’s some of the similar things for AI, and marketing automation, and predictive analytics, and all this stuff. At the end of the day, you need to know what your business goal is and be able to say, “Yup, that’s what I ordered.”

Sean: We reached the end of the show here, but there’s one other question that I wanted to explore with you, which is there is this idea, it was coined … I believe it’s something that was coined by Google. AI first versus AI later. I think that this is an important concept for people to start to think about as they start to investigate these types of technologies, but it doesn’t seem to me that AI is just something you sprinkle on later. There is such a thing that certain products that were born from AI. What do you think of that separation and is that something that people should also start to gain an understanding of?

Chris: Individual purpose build AI, absolutely something that you can add in later because it’s essentially supplanting a specific process. As long as you’ve got the process clearly defined, and you’ve got a model of governance, and you’ve got a model of output and measurement around it, that’s fine, but if you are considering pivoting more than a single purpose part of your organization, then yes, you have to do it first because it is a different way of doing business. It is a way of doing business that says, “We are from the beginning not going to invest in human capital.”

John: Chris, how about for the next year now? What’s on your radar as far as tools that you’re looking at for businesses and for people to introduce into their stack? What’s on your radar as the most exciting thing out there?

Chris: That’s a big question. What I think is really most interesting is how much all the major tech companies are throwing to open source because I think, they’re realizing they can’t do this by themselves and, in a lot of cases, they are better off collaborating than they are competing. At end of the day, the impact is not the technology. The impact is how you deploy the technology.

I personally am looking too, on a regular basis, Facebook, Microsoft, IBM, Google, Amazon. We as the individuals are getting the crumbs from their labs in some cases, but that’s what I’m working with right now is playing with all the different tools that they release, trying to make operational some of used cases that they have, and figuring out how I can take an individual task, or item, or things and automate as much of it as possible.

That’s what I would advise anyone to do. What can I do, even if it’s a small step, even if it’s something as simple as, “Hey, I need to run some predictive analytics on my website traffic,” start there and then as you become more comfortable with the tools, start building them in, you will find that the used cases will snowball faster, and faster, and faster until that’s the majority of your time. For folks who are interested in deployment stuff, you need someone who has either R or Python skills. You need someone who understands cloud computing really well and you need someone who understands statistics because the vast majority of machine learning is based on math.

John: Okay, great. Sean, how about for you? What have you got on your radar over the next year? What’s at the front of the table for you for exciting stuff?

Sean: I think the thing that I’m thinking about the most is efficiency metrics. If we’d look at the work that we’re doing here at Infer, it really all starts for us with the sales and marketing sides of the house and those business units because there is an incredible amount of inefficiencies that we’ve built. I think those inefficiencies are actually going to grow, which opens up an opportunity. I’m probably actually one of the few that is barish on the rise of the SDR and the human spam canons that we see.

I think in theory, it seems like it’s a great idea. In practice, I see it getting abused more and more and we seem to be going down the same or we’re headed for the same cycle that marketing automation went and it’s an easier way for us to spam people. I think that as people start to think about where they’re applying predictive analytics and AI to their business, they need to be thinking about sometimes it’s not even the addition, it’s where you can reduce. It’s using it. It’s such an important auditing tool to understand how inefficient your business is, whether that’s locating marketing channels and reducing marketing waste or it’s understanding that building an army of telemarketers is probably not the most efficient way that you can grow their business. You can actually do a lot with a lot less.

I think those are the types of questions that if you’re trying to be the hero to your CFO, how you can start to utilize this type of technology. I’m big at looking at efficiency metrics right now and stressing that especially with whether your VP of marketing, demand gen, or sales leaders, or operations leader, I think that there are so much efficiency gains not even for what we can actually boost and supercharge with AI, but just to clean up what we’ve been doing today. I think that we’re going to see some problems grow a lot worse. Like I said, fairly barish on where the SDR model will be in five years, but I think that there is also a very exciting way that efficiency metrics drives quality and a better customer experience. I think that that’s something that I’m really looking for that equation to play out. That’s where my thoughts lie ahead.

John: Great. Chris, for people to keep up with what you’ve got going on in AI and automation, what’s the best way for people to see what you’ve got going on?

Chris: You can do a bunch of different things. You can read my blog,, and sign up for the newsletter. You can follow me on social media, just search for me, and you can subscribe to the Marketing Over Coffee podcast at

John: That sounds great. Sean, how about for you, if people want more info on predictive and Infer?

Sean: You can definitely head over to We actually are relaunching the blog, which is pretty exciting. We have a lot of new material over there that people can check out if they’re interested in AI, and predictive, and things like that. Other ways, just find me on Twitter, @SZinsmeister, or grab me on LinkedIn, or just Google Sean Zinsmeister. It’s pretty easy to find me there.

John: All right, that will do it for us for this week here. You can find out more over at and feel free to leave us a review over on iTunes, but until next time, we’ll you see in the stacks.

John Wall

John Wall

John J. Wall speaks, writes and practices at the intersection of marketing, sales, and technology. He is the producer of Marketing Over Coffee, a weekly audio program that discusses marketing and technology with his co-host Christopher S. Penn, and has been featured on iTunes.

Stay Up to Date