top of page
  • Writer's pictureDigital Marketing Hub

A Beginner's Guide to Artificial Intelligence and Machine Learning for Marketing People

Updated: Jun 25, 2021

Hi, I'm Jim sterne. And I'm here to give you an introduction to artificial intelligence for marketing. I've been in online marketing since 1994. I've written a slew of books about online marketing, advertising, customer service, etc. And my latest one is artificial intelligence for marketing. Now, I'm not here to teach you how to be a data scientist. I'm here to teach you how to be a better marketing professional. Artificial Intelligence is a new tool for us to use. And I'm here to explain what it is how useful it will be, and how you can up your game so that you will continue to be a valued marketing professional. Here's the outline. First of all, what is artificial intelligence? How does it work? What is it good at? And then what is it good for?

First of all, let's talk about what it is not.

Of getting a machine to recognise cats. We're going to give it labelled data. That means here's a photograph, and we're going to label all of the cats as cats. And it's going to look for what is similar? How are these things alike? Well, they all have these pointy things on the top of their head. They all have these almond shaped eyes, they're all sort of looking in this direction. And then it looks at this thing and goes, I don't know what that is. That's confusing to me. So you have to give it lots and lots and lots of cats in order for the machine to understand. And when I say a lot of data, let me just show you an example. This is how many data points it's necessary for CAPTCHA to understand that it's looking at the letter A.

So that is supervised, you know, the answers already unsupervised is when you don't know the answer. And you want the machine to figure out something for you. You want the machine to tell you something about the data that you didn't know before. So let's look at a bunch of my different customers, and tell me what buckets they belong in. These are the ones who are most likely to purchase. These are the ones with the highest lifetime value. These are the ones who are most likely to never buy from me again. Or let's go out and now that you know what my best customers look like, go out and find others out in the world who looked like that, that I should advertise to this is an labelled data really, really good for categorization.

The third category, the third type of machine learning is reinforcement. This is when you're not sure what the answer is, we want to put the right message in front of the right person at the right time. But there is no exact message to put in front of a person at exactly the right time. So we give the machine the authority, the capability of trying things out of experimenting. And if it gets the right reaction, somebody clicked or they engaged or they purchased, then the machine gets a reward, a mathematical reward. And so it tries to do that better. And it tries to optimise over time.

So we've got supervised, unsupervised, and reinforcement learning.

Okay, that's what it is. But how does it work?

So the amazing thing about this stuff is that it actually delivered on the promise of big data. Big Data was a great idea. But it's just too much for the human mind to hold too many permutations too many possibilities, too many numbers to keep track of. Fortunately, computers are very good at keeping track of numbers. So that's great. So data scientists created a bunch of algorithms. Now, it's not terribly important that you understand what all of these are. If somebody talks about decision trees or support vector machines, that's fine. Just go along with the conversation, they'll get around to it being meaningful soon. But just for the sake of interest, let's dive into what a neural network is and what deep learning means.

This is the simplest of possible neurons, we've got three inputs and a decision. The three inputs are, how much will it cost to go to the movies? How's the weather outside? And how much work will it be?

So how much does it cost to go to the movies? Well, if I want to go to a Hollywood premiere, I have to get on an aeroplane and rent a hotel room and rent a toxin. That's, that's expensive, that's too expensive. I'm not going to go doesn't matter about anything else. I'm just not going. But if I'm, if it's a normal price ticket, well, the next consideration is Oh, there's a blizzard outside, or it's 120 degrees outside. I'm not going anywhere. So it doesn't matter if it's free, I'm not going to go. Or Finally, the amount of effort I have to get a babysitter have to stop by and check up on my dad and make sure he's okay. Got to make sure we stopped at the grocery store on the way back. Oh, that's just too many things going on at once. That any one of these can override the others and be the go no go decision. Now, neural networks have hidden layers. So in this human example, the hidden layer is the last time we went to the movies, we saw science fiction action adventure. So this time, we're gonna have to see a romantic comedy. Hmm. Last time we went, I broke a tooth on an unpopped kernels popcorn, bad experience. The last time we went to the movies, my wife and I got in an argument about where we should Park and so that was not a pleasant experience. Hmm. Now, I'm not consciously thinking all of these things. They're just they're going on back here. So when she asks, Hey, honey, do you want to go to the movies? I'm going? Not really. And if she says why, then I've got to start sorting through all of my thought processes. That's what the machine is doing. It's hidden layer.

of computation. And it can have 1000s of layers. It can have many, many hidden layers, which are not logical. They're not reasonable. They're mathematical.

And then the machine spits out an answer. And let's say that we're looking at supervised and it says, this is a cat. And we say, No, no, no, that's not a cat. And it's important for you to know that it's not a cat, and go back and figure out why you thought it was a cat and fix it. So the machine is going to work backwards, and is going to find which node overrode the whole decision and which node cause that one to change its mind and which previous one caused it to change its mind. And this is called back propagation. Now, this is as much math as I'm going to show you backpropagation says, Oh, I see why I thought that was a cat. I'm going to change the model, the mathematical model. So if I see something that looks like that, again, I'll know what's a skunk?

Now, here's where things get fun. Decision trees, random forests, support vector machines, neural networks, deep learning. What if we put them all together in an ensemble? What if we gave the machine the choice over which algorithms to use and how to use all of them? To figure out which one's best? Yeah, it's a little meta. But this is where data scientists are playing these days. It gets really interesting, really quickly. So that's what it is. And that's how it works. But what's a good at? Well, two things specifically, dimensionality, and cardinality. And if you're like me, those terms mean nothing. So allow me very briefly to explain. dimensionality is attributes per object. Okay, what's an object? It is a thing in your database? Let's call it a person. So you've got a bunch of people in your database, big data, lots of people great. But if you know a lot of things about them, that's many attributes, you know, their name, address, phone number. When was the last time they called when? What were the last 16 pages they looked at on your website? What's their phone number? Those are attributes about them. cardinality is options per attribute. So let's take phone number, that little that individuals phone number, how different is it from everybody else's? Completely? The number of options of phone numbers is as many people as there are, everybody has their own phone number. So what is their age? Well, somewhere between one and 120. What is their zip code? Well, there's 43,000 zip codes in the United States. What is their phone number? Well, there are 7 billion people, there are 7 billion phone numbers. So there are lots of dimensions, we know a lot of different things. And lots of options per attribute. And you put those together. And there are so many permutations, the human mind can't handle it, but the machine can. The machine is really good at high dimensionality and high cardinality. So here's a quick recap of what we covered. so far. supervised and unsupervised. Reinforcement Learning decision trees. Yeah, I skipped over support vector machines didn't have enough time. Neural Networks, putting them all together and ensemble. And then what is the good machine? Good. Okay, so how is it useful? Well, it's good for marketing. If we go all the way back to the definition of artificial intelligence, you remember, it included things like robots? Well, yes, robots. So this is a Japanese hotel that uses a robot to check you in, step up to the counter, the robot will check you in unless you go to the next counter over. And then this robot will check you in. And if you're really good, this robot will bring you your midnight snack. Robots are also being tested in stores. So the Lowe's, home improvement stores are using robots to greet people at the door and direct them to what they're looking for. When we get to computer vision, here's a company called gum gum that looks at social media to identify what's out there, and reports back to you on where your image shows up. Your brand shows up who's talking about you what kind of influence they have, whether you'd like to reach out and engage them with further promotions or to use their visuals and of course to track the competition. Now computer vision is also useful for augmented reality here to find your store or to identify pricing here to bring to 3d life, two dimensional picture or just to help you find recipes off the ketchup bottle. Warby Parker. is using it to help you identify what kind of glasses would look best on you. When it comes to natural language. Well, you talk to your phone, right? You talk to your Alexa or your Hey, Google device. These things are becoming more ubiquitous and are going to becoming a serious challenge to marketers. When you ask Amazon to send you more paper towels, which brand is it going to choose? And how do we as marketers get on that list? Now when we bring these together, it becomes a very compelling, natural language processing conversation, bots, voice recognition. I'm not sure if you're familiar with the staples easy button. But take a look at this video.

It's just a blue pen, right? sounds simple, but it's surprisingly complex. What brand? What type? What quantity? Now magnify that across the chaos of all this stuff you order for all the people in your office? What if managing this was as easy as saying, hey, please tell me what you need. Blue pens. Now it is. That's the magic of the staples easy system. using simple spoken words anymore post it notes. The easy button lets you order everything your team needs to be productive, fed, and caffeinated French Vanilla coffee and letter size copy paper. Got it. That was easy. It's an extra set of hands, helping you save countless hours, manage costs, and to maintain control of ordering supplies through boxes of Sharpies. The staples easy button brings the power of on demand to businesses on any device exactly when something is needed. Drivers markers order with voice through the easy button or through the app border with a text with an email with a slack bot with a photo. Or even with Facebook Messenger. It's easy to track shipments, I need to check on my order and get in touch with customer service. The brain behind the easy system is IBM Watson. Watson's cognitive intelligence turns what you say into data, and then turns that data into answers you need. The staples easy button learns over time. With every interaction, it becomes more personalised and more intuitive. Please tell me what you need. So when you tell the easy button, you want blue pens, it knows the exact blue pens you want. Got it? That was easy. That's the staples easy system.

I think you'll agree that's a pretty compelling story, and a glimmer of what kinds of opportunities we as marketing people have to work with in the future. But what about today? How do you bring this on board? How do you bring this into your company? what what what's the first thing you need to do? Step one, you are responsible as a marketing person to know marketing, and to know the industry that your marketing for what questions you're trying to ask what problems you're trying to solve what business imperatives you're trying to address, the machine will only do what it's told you need to know what to tell it. Number two, know your data. Where is it collected? How is it collected? How is it stored? How is it cleaned? How is it integrated with other data? how might it go wrong? What How might you bring two datasets together that address different realms but somehow need to find a way to mesh together? That's a human problem. So where do you begin? You start with tasks that are repetitive, and sort of boring, and have a lot of data available. Whether it's sorting sales leads, or coming up with a response to social media questions, or creating a chatbot. These are the things that there are lots of iterations and an opportunity for a machine learning system to help you. The next question is should you build this stuff or buy it and I am strongly on the buy side building, it means you have to hire data scientists, you're going to compete with everybody in their brother who is out there building this stuff, to find the PhDs who can figure out algorithms. But at the same time, large companies like Google and IBM and are adding machine learning tools into their data systems. If you're a Google Analytics user or an Adobe analytics user, those companies are building machine learning systems into their marketing platforms. But there's also a lot of venture capital money being spent at startups. So startups are coming up with ways of taking what data you have applying their data science and giving you insights and answers. You have to decide what problem you want the machine to solve. And then you have to decide what data it should Consider, if you give it too little data doesn't have enough to work with. If you give it too much, it won't be very confident in its answers. So there is a correct amount, or a correct variety of data to be considered. And finally become proficient at the smell test. Now, this is something that's uniquely human, the machine is going to take the question you asked, take the data that it's been given and come up with an answer. And if you change the data, it can change its answer, that's fine. But it can't tell you whether the answer is meaningful. There are some things a human can look at and just go, nope, that is not gonna work. I can tell. So where does the human come into this? What advantage Do you have over the machine, the machine is crackerjack at large amounts of data. It can correlate quickly. It's very accurate. It's low cost, it doesn't get tired. It doesn't take vacations, but it ain't human. It doesn't understand compassion, or empathy, or insight. It doesn't understand ambiguity, it cannot take two conflicting concepts and make sense out of them. And that's what the human can do. So here's your homework assignment. Your job will be to stay tuned with what kinds of tools are available, and what they can do. They're changing rapidly. Stay tuned. Number two, train your bots. This is a very interesting area. You're already using spellcheck, right? Well, there's a system that Google has that will guess the rest of your sentence, not just the spelling mistakes, but suggest entire sentences for you to just hit return, and make it faster for you to write up a response to an email.

If those kinds of tools are available to everybody, then your special value comes in training, your version of that tool to understand you better train your bots, because eventually you're gonna walk into a company, and you're gonna have an interview, and they're gonna say, What experience do you have? What tools do you use? Well, I'm very good at Word and PowerPoint and Excel. And here are a half a dozen bots that I have trained to help me do market research to help me do analysis to help me write reports to help me identify good graphics to use. And those are part of your skills. You're bringing that trained bot into the picture, and that makes you a better catch in the future. My name is Jim stern. I run the marketing evolution experience conference. I'm co founder of the digital analytics Association. And I hope you now have a strong introduction to artificial intelligence for marketing. Thanks for listening.

| Intro to Artificial Intelligence for Marketing by Jim Sterne licensed under a creative commons (reuse allowed) license. Based on a work at |

Recent Posts

See All
bottom of page