A Marketer’s Dilemma: Choosing What to Automate

This is our last episode of 2018, and we’re thinking ahead to next year. One of the big questions is what are we going to automate in 2019? This episode discusses the importance of planning and setting specific goals when it comes to automation. We’ll talk about both the benefits and the limitations of automation, and how you can combine automation with a human function to get good results. Below is a lightly edited transcript of Episode 34 of the Inevitable Success Podcast.

Transcript:

Damian: We’re thinking a lot about what we’re going to be doing in 2019, and one of the things that came up is what are we going to automate in 2019? Stephen, you had a very interesting response to that, which was…? Stephen: Do you know what you’re automating? You automate things that you know how to do already. Automation is not about creating something out of nothing. That means you need to have a goal first. We talked about that when we talked about the benefits of modeling in general. Modeling doesn’t give you any answers—machine learning, or AI, or human-made models—it doesn’t matter what it is. You have to know what you want out of it, and deciding that is a uniquely human function. Nobody is going to do it for you. Damian: Yea and one of the other things that I thought was interesting is if you try to automate things you really don’t know the answer to, who knows what the expected outcome is? What could you expect as the outcome? It would probably be bad. Stephen: That’s right. The whole of automation is really two-fold. One is to cut time so that we do things faster. Two is to do it with fewer people. Not good news for a lot of the workforce out there, and there’s a reason why a lot of publications talk about how many jobs will be gone if AI takes over. Some of them are total science fiction, but some of them are not unfounded. A lot of people will lose jobs. Let’s face it, though—we automate things to cut time and to cut human resources. That’s it. Damian: You know what’s interesting—to kind of flip that on its head a little bit—there are so many businesses out there that have one or two-person marketing teams, and their revenue doesn’t justify having more than that, regardless of whether they are amazing or not. I actually think that if a person really studied and became a student of how to be a good automator, that’s a massive opportunity, because then they could say, “I’m the guy (or the girl) who can do the job of ten people by myself, because I have this kind of technology.” Stephen: But let’s say that you’re a marketer, and you have a lot of jobs or things to do on your list. You have to break it down from the point of view of not just the things you don’t want to do, but also what is the most time consuming and what is repetitive. You automate those things first. Damian: That’s a great thing. I think that’s something you should write down. Stop and think about what you did in 2018 (or any year), and think of the things you did over and over again. Those are repetitive tasks, so can you be thinking, write down some rules, find some sort of technology, to outsource those tasks to a machine. Stephen: Now, if you decide that a task is right for automation, then what’s next? Damian: As I was saying that, I actually think that one of the next questions to ask would be what are all the things I was supposed to do, but didn’t do? Because I probably won’t do them again next year. Stephen: Well, if this is about writing a new procedure for something, then a machine is not going to do that for you; you have to do that yourself. Now, let’s say that you isolated a task or a bunch of tasks that you want to automate. Now you have to think like a machine, even if you’re not a coder. So, imagine you have thousands of offer codes, and it’s in really bad shape. You want to automate it, because you don’t want to go line by line and clean it up yourself. Well, there has to be some logical way to express that command, otherwise, no one can program it. Even if you use the pattern recognition module of machine learning, you still have to teach the machine what it’s supposed to clean up. You have to do it in a way that converts your thoughts into logical steps. I heard about a bunch of enthusiastic young mothers who now not only teach their kids foreign languages and math skills early on, but some of them also wanted to teach them how to code. It’s a very noble idea. Do you know how they teach code to four or five-year-old= kids? Damian: I’m very interested in this for two reasons, one because my first-born could be born at any minute—we’re shooting for the next couple hours, but it’s probably not going to happen—but the other reason, and I’ve said this a bunch of times, is that I think it’s an amazing thing to learn as a kid. So how do they teach it? Stephen: They take a task of say creating ramen noodles. You want a machine to open up a packet of instant ramen noodles, put it in boiling water, and start cooking at the right temperature. You know how to do it, because you’re a human being. You know when it’s done. But let’s say that machine has no idea what cooked ramen looks like. Write an instruction from step one all the way to however many steps it takes. That is the first lesson in coding. What would be the first task, do you think, if you were making a packet of ramen noodles? Let’s say that you have pots and pans and a packet of ramen. What would be the first step? Damian: So, I have all the ingredients? Oh man, this is a pop quiz. Stephen: Yea, you just have to do it. There’s no right or wrong answer. By the way, a machine should be able to do it. Damian: All right, I think it’d be something like, “Reach out with your dominant hand.” Stephen: Well, a machine doesn’t have a hand! Damian: Well, there you go. Stephen: Let’s just say such a machine exists. I would say that you have to talk about the measurement of water. How much water do you need? Let’s say 450 mL. Then when does it boil? Up to what point? You have to teach the machine when the boiling point is, so that means you need the module that measures temperature, or some observation like there are a bunch of bubbles coming off the water. Now it’s boiling. What do you put in? Open the packet. How do I do that? Cut it from the top, from the bottom, sideways, or look for some indication of a cutting line? Damian: Wow, we somehow turned this into a cooking show, I don’t know how we did it. Stephen: Yea, because everybody can relate to this. The point is, let’s say that your task is to clean dirty data—I was just looking at some data this morning, and do you know how many variations of Facebook I saw there? Facebook is a good source, right? So, can you imagine all the variations I saw? You could have Facebook with a lowercase, it could be www.facebook.com, it could be m.facebook.com, or it could simply be FB, etc. The point is, now you’re not doing it, the machine is. Now you have to think like we did with the packet of ramen. Where do you start? Is that FB a combination, or do I give an example of what FB could be, or let the machine just do some deep learning? If it’s learning, you have to tell them FB is right and FBC is wrong, because the machine is not going to know. Damian: That rule could be used in a lot of places, because most often, when I’ve encountered the need to write something like that, it’s been for reporting. I actually think that’s a great thing to automate if you can, because typically that’s something you know, and it’s repetitive, and sometimes it’s something you should be doing but aren’t doing, and sometimes it’s something you really shouldn’t be spending so much time doing, but you are. Stephen: That’s right. You also have to think like a machine does. What if none of the rules that you give to a machine capture every error that there is, and yet you even have to think about the fact that if you do all these things in a very exhaustive way, and you still have some things to clean up, call operations, or whatever. But if you just don’t say it, the machine is not going to do it for you. You have to say it explicitly. Damian: Yea, I definitely think there’s a place for troubleshooting and Q&A (quality assurance) on anything you go to automate. It’s actually a good point. When you’ve automated something that is not a last step, you need to look at that. Stephen: That’s absolutely right. There’s a meme floating around—let me paraphrase it, because I don’t remember exactly. Do you know the difference between machine learning and deep learning and AI? Very simply, AI will correct their own errors, whereas machine learning still needs human beings to correct them. It’s a very simplified version of the explanation, but we use these words interchangeably anyway, and probably they’re not wrong. For a layman, who cares, as long as they’re not the one doing it. But the point is, you are the one with the goal for what the machine is supposed to do. Is it about finding some patterns that are useful for a future sale? Is it about building an actual model to predict who is going to be the most valuable customer? Or is it about sorting things so you don’t have to sort them to find the ten most valuable leads? What is the goal? It’s not about math or whatever. You have to have the goal, that’s number one. And number two is to determine if this repetitive and automatable? Let’s say that finding the best lead is the goal. Let’s say the machine produced a lot of high scores. From the machine’s point of view, with the data it has, that’s all it could do. Now you have to break the tie—how do you do that? In that case, there is human intervention. A person ultimately makes the decision based on the results the machine produces. So that is the point: you don’t have to be the coder, but you still have to think in terms of a coder and think about what the machine needs to break things down. Think logically in terms of how you are going to instruct the machine, and do you have all the ingredients to do full automation. That would be my advice to marketers who want to go to the next level with AI to make their businesses faster and better. When it comes to what the machine is actually supposed to do, you have to think about it for awhile. Damian: I agree. In closing, I would strongly encourage, if not challenge, the listener to write down two or three things that you plan to intelligently automate in the coming year. Then think about what you are going to do with all that time you just opened up, because that is where the ROI is. Stephen: Right, well, there’s no shortage of work I hope. Damian: Well, if you just saw automate what you did last year, and then don’t do anything with that freed up time, you’re going to get a flat ROI. You’ll have more time, maybe your golf game will get a little bit better. Stephen: Hopefully you’ll think a little more about marketing, but you know, you’re right. There’s a joke among coders that the laziest coders write the best macros for automated modules because they don’t want to do it again. So, some laziness is a good motivation for automation, yes, but with that extra time, hopefully, you come up with a wonderful idea of how to sell better in terms of new ideas and new products, not just repeating things that you’re doing all the time. Damian: Right, well, on that, Happy New Year! Stephen: Happy New Year! This is the year of the pig, I believe. It sounds prosperous, so happy and prosperous new year to everybody. Damian: If you enjoyed today’s episode, we ask you to please leave a rating and write a review. Or, better yet, share with another marketer. Be sure to subscribe to the podcast for new episodes. Also, check out the show description for complete show notes and links to all resources covered in today’s episode. If you’d like to speak to someone about any topics covered in today’s episode, please visit buyergenomics.com and start a chat with the team today.