You Ask, I Answer: Challenges Preventing AI in Marketing?

In today's episode, Heidi asks, what challenges keep you from examining and or using AI for your current marketing? Well, I we don't really have that problem because we do use AI for a lot of our current marketing. But let's take this from the perspective of, say, a client. Um, there are a bunch of different challenges that can keep a company from using artificial intelligence. Uh, first, and probably most fundamental is bad data, right?

So if the data that a company is working with is bad, if it's in bad shape, uh, if it's, you know, uh in all sorts of wild and crazy wacky formats, if it's in hard to access systems, it becomes very difficult to use that data for uh predictive purposes or even just for classification purposes to figure out what data do we have. And this becomes really relevant when you're doing something like, for example, attribution analysis. If you have missing data from your attribution analysis and you're doing uh really big model, something using maybe like Markov chains or uh certain types of decay models, uh, or even just multiple regression models, and you've got missing data, data that is important, but you don't know that it's missing, uh, you can build an attribution model that will not be correct, right? It will be uh something will be off, and you may or may not know that it's off. So uh in in cooking terms, if you were to think about this in cooking terms, imagine baking a recipe, a baking a cake, and you leave out an ingredient and it seems like it's okay, but in fact it's not okay, right?

So maybe you're making chocolate cake and you leave out the cocoa and you have something at the end that's edible, right? And it's it tastes like cake, it just doesn't taste like chocolate cake. And if you're doing something like unsupervised learning where you don't know what you uh what's in the box, you may think, oh yeah, um this is vanilla cake. And in fact, it's supposed to be chocolate cake, but you don't know that you're missing the cocoa. And so that's an example where bad data, in this case, missing data can have a substantial impact on the model.

The second thing that causes issues, uh, and sometimes very substantial issues, is thinking about artificial intelligence as a solution. Artificial intelligence is a set of tools, right? Think about imagine if we as business folks, we talked about AI the same way we talked about spreadsheets, right? Um we go around saying, well, should we use a spreadsheet for this? Maybe this is a spreadsheet problem.

Let's let's uh let's try using spreadsheets for this. And you get how silly that sounds, right? If you're dealing with something like, say uh you know, public relations stuff, like writing a better media pitch, a spreadsheet's probably not going to help you do better writing, right? It may help you categorize, say, uh the prospects that you're pitching, but it and it's unlikely a spreadsheet's going to help you write a better pitch. Um, a word processor would be the better choice.

And so one of the things that happens with artificial intelligence is that people think that it is um a solution when it really is just a tool, right? It's if you're in the kitchen and you've got you know a blender and a food processor and a toaster and stuff like that. Do you say, well, what can I use my toaster for today? No, I mean, you probably don't think uh appliance first when you're cooking, right? You think about objective first.

I want bacon and eggs, I want a pizza, I want sushi, I want you know something along those lines. And then you reverse engineer based on what you want. Do you have the ability to make that dish, right? If you don't have rice and you don't have a rice cooker or some means of cooking rice, you're not having sushi, right? If you don't have a blender, you're probably not having a smoothie.

I mean, you could, but it's gotta be a lot of work. Um if we think of AI as essentially a fancy appliance, then suddenly it is less about uh using the technology, like I've got to use this convection oven. No, no, no. You've got to make a dish that you want to eat. And then maybe AI is the right choice, maybe it's not.

Generally speaking, artificial intelligence is really good at problems that have a lot of complexity and a lot of data and a lot of data. So if you are dealing with a problem that doesn't have a lot of data, AI may not be the right choice for it, right? AI may be the wrong choice for that problem. In fact, uh there are certain problems where AI makes things more complicated, right? Um, where it's just not the right fit.

It's like trying to use uh a blender to make an omelet. I mean, you can, but it's not going to taste very good. Um, you're much better off using a frying pan. So those would be the the major challenges where I think people run into trouble. When companies are hesitant to adopt AI, it's because they don't understand the the technology itself, right?

It's like getting a kitchen appliance and you don't know what it does. You're probably not going to use it for your big dinner party, right? You're probably going to take some time and say, okay, let's let's see about maybe uh using something we know. Uh and so uh if we want to encourage more adoption of AI, we've got to simplify people's understanding of what it does, right? If you take apart your blender, there's gonna be all sorts of stuff there controllers, chips, solenoids, uh, you know, depending on how fancy your blender is.

Do you need to know how an electromagnetic motor works? No, you just need to know what the blender does and what it's good at and what it's not good at, right? The inner workings really aren't as big a deal. AI is very similar, right? You don't need to know how uh a neural network works.

You need to know is it the right appliance for the job? And to do that, you've got to have problems that are well suited for using AI. So those would be my my challenges that I think people struggle with uh with artificial intelligence. The rest of it really is just math. It's just math and data.

So if you can grasp the strategic uses and the conceptual uses, the implementation is relatively straightforward. Not easy, but straightforward. It's not overly complicated once for for most marketing problems. So really good question.