You Ask, I Answer: Different Types of Machine Learning and AI?

In today's episode, Mark asks, could you share the different types of machine learning and AI that businesses like printers should be aware of and how they can apply it in their operations? Okay, so all machine learning and AI falls in three big buckets, three broad categories regression, classification, and generation. And arguably you could argue generation is actually a combination of regression and classification. But those are the big three. Um at Trust Insights, we use the acronym FOG to explain these categories.

Find, organize, and generate. Regression is when you have a series of numbers and you have an outcome, what mathematical relationships in those numbers do you see to the outcome? So if you have, say, marketing channels and you have sales or uh sales qualified leads, you would use regression to say, okay, what of all these variables, all this data that I have has the strongest mathematical relationship to the outcome that I care about? And in doing so, I can then say, well, these things are the most important and the most relevant, or at least the things we should test out. So if you did that and you were looking at uh Facebook and LinkedIn and TikTok, and you know, TikTok had the strongest correlation to the outcome you care about.

You could then say, okay, well, TikTok appears to have this relationship. Now let's build a TikTok marketing program or invest more in it. And do we see a an increase in the results that TikTok generates commensurate to the investment that we are making? So that's regression. And you again, you see this in attribution modeling, you see this in sales modeling, you see it in forecasting of almost ever any kind.

When you're doing those activities, you're using regression modeling. And certainly for a company like you know, a printer or a a um physical goods manufacturer, forecasting and predictive analytics are really important because your your content costs a lot of money to manufacture it. It costs you know real materials, physical, tangible materials. And so doing those predictive analytics can help you optimize your spend to focus on what's going to deliver results. Second category is classification.

This classification is exactly what it sounds like. You're organizing data. When you have a big pile of data like social media posts or the contents of your customer service, inbox uh transcripts from your call center, that's all what we call unstructured data. And it's a big pile of well in a lot of cases it's just a big pile of words. When you do classification, you're going to try and organize all that data.

You know the what are the highest performing social posts, what are the lowest performing social posts, what are the topics that they're about and things you're not trying to find an outcome. You're just trying to figure out what's in the box what kind of of sense can you make from your data. And classification, there's all sorts of things like you know K nearest neighbors and clusters and uh support vector machines and all these different ways to to classify data. But at the end of the day what you're trying to do is make sense of the data, trying to organize it with things like topic modeling, labeling and clustering just to just to clean things up to make it workable to make it manageable. In a lot of cases to classify things so you can reduce them from being huge data sets to just large data sets and then you can run other things like regression on the results of that so that you're not trying to boil the ocean all at once because some of the most complicated analyses, they have a they require a lot of horsepower.

They require a lot of compute power. And if you are trying to uh first use up an enormous amount of compute power just to do the the first round of feature engineering, you might never get done with your work. If you can classify it, you can reduce the dimensionality, the number of variables there are, um find patterns in the data, and then run the analysis on those patterns, which is a big time saver. The third category, of course, is generation, the making stuff. We're all familiar with what that does when it comes to again um printers or the physical world, direct mail and things like that.

Generation is a great set of tools that you can use to build creative, to build visuals, to uh revamp uh creative and content, uh do things like persona analysis and then persona based creative development with language models. You say, hey, here's my ideal customer, here's my sales copy, rewrite my sales copy in a way that will appeal to the the ideal customer based on their personality or based on their what you know about this industry and these these kinds of people. Those are those are really powerful tools. And I would say you need to have all three. You need to classify your data to organize it, you need to use regression to figure out what's working.

You need to use regression also if you want to train, you want to custom train uh some stuff along with classification, and then you generate, you make stuff with the data you have. So, really good question. There's a lot to dig in here too. There's a ton of math. Uh, but I think that's a good place to stop for now.

So thanks for asking. If you like this video, go ahead and hit that subscribe button.