4 Types of Data Analytics to Improve Your Business

The 4 Types of Data Analytics:

  1. Business Intelligence Reporting (B.I) – The most common type of analytics, also known as data mining, and it is really all about understanding your environment. A few examples are: “I blasted this e-mail and this campaign is going well.” “I have a very small bounce rate, or delivery rate, open rate, click-through rate, conversion rate.” These metrics are normally displayed on dashboards not to dis-similar from your car. These metrics give you an idea of your marketing’s performance.
  2. Descriptive Analytics – Descriptive analytics are about describing who your customers are with real data. For example, “What is the median income? What kind of neighborhood is this? Or, what is their income level? What kind of car do they drive?” With descriptive analytics businesses understand who they are dealing with in their database. A perfect example of descriptive analytics are clusters, a “cluster” is a group of people who have similar attributes (they look-alike).
  3. Predictive Analytics – Predictive analytics, the word itself, is about the future: who are more likely to do something? But in the age of big data, it’s not just about future prediction. It also fills in the gap of incomplete data.
  4. Optimization Analytics – Optimization Analytics are about finding what combination of marketing efforts will yield the most revenue or profit. This is a multi-channel, omni-channel world. You have a limited marketing budget. How should you spend your money? That’s optimization.

Below is a lightly edited transcript of Episode 19 of the Inevitable Success Podcast.


Damian: We touched on this a little bit in our previous episode, but we felt: hey, let’s go deeper. And today we’re going to talk about the four main types of analytics and high level. They are: BI reporting, descriptive analytics, predictive analytics, optimization analytics. And by the end of this episode, you’ll have an understanding of what each is; an example, so you kind of relate to it; and then we’ll talk a little bit about solution-ing, as you call it, what to do.

Stephen: Exactly. Solution-ing is nothing but understanding what situation you are in and deciding what to do. But to do that, you have to know what each of these types of analytics do for you and what this is for, and…it’s like doctors knowing what treatment does to a patient. You’ve got to know what it is, actually. Even as a patient, you have to know what those treatments are. I’m not asking you to treat yourself. You’ve got to know what those things are.

Damian: So the first one: BI reporting—business intelligence reporting.

Stephen: So that’s the most common type, by the way. A lot of folks who say, “Oh, I’m in the world of analytics.” Data mining is really all about understanding your environment, what is going on in the world. And the easiest example that I can give you is that when you drive a car, you have to look at your dashboard. How much gas do I have left? How fast am I going? If I drive at this rate, how far can I go? And will I reach the next gas station? So ‘how long was I in the car?’ and all that kind of stuff. So that’s like what’s going on. So, in the marketing world, if you translate all this to the term that we always use, it’s like: Yeah, I blasted this e-mail and this campaign is going well. I have a very small bounce rate, or delivery rate, open rate, click-through rate, conversion rate. Or: how much money did I make on each click? Or: all these people who bought something, are they coming back? All these things are…and that’s, I guess, why we call them dashboards. It’s not that different from the dashboard on your car that tells you what’s going on.

Damian: You know, it’s so interesting. I’m going to be a bad host for a second and take us on a tangent but…I can do that. This is our show, right? So, you know, it’s funny: they talk about the future of dashboards in cars. And because of automation and, you know, all this, they think that they may go away, because you don’t need to have them anymore. I can almost see a scenario, like, if AI or predictive got strong enough, where it actually could do what they wanted to do.

Stephen: Sure.

Damian: You might almost not need that. But…who knows?

Stephen: Well, you’re getting into some philosophical discussion here.

Damian: Yeah, I have a bad habit of that.

Stephen: I love that, by the way. But let me ask you this. When you fly, you can just put on an air show in front of your screen. Instead of watching a movie, you can just turn that on. And that tells you how fast you’re going, what is the wind speed, what is the altitude; if you fly at that speed, when you’re going to reach your destination. Well, let me ask you this: you’re not flying the airplane, but there’s a comfort in knowing. That’s why you have your dashboard.

Damian: Yeah. You want to know where you’re going.

Stephen: Exactly. You’ve got to know what’s going on in your environment. So, yes, the machine will drive the car for you in the future, but you may want to know: okay, so I’m not driving, but when am I going to get to work?

Damian: Yeah.

Stephen: So that’s why you need the dashboard. In business you have to know. I mean, decision makers? They have to know. You may not be the person who blasts the e-mail every day. You may not be that guy. But you want to know. You say, “Hmmm.” So that I can plan better. So I know for sure that, you know, it’s not a good thing for what we sell. Let’s not send these e-mails on a Saturday morning. So if you know that, even if you’re not the person who blasts the e-mail yourself, you can change your plan.

Damian: I couldn’t agree more. Even if AI does get to that point where it wouldn’t need it on a daily basis, you’re still going to have to manage the AI as if you would an employee at some level.

Stephen: That’s right. Machines are not that smart, by the way.

Damian: That’s why they’re always learning.

Stephen: That’s exactly right. It’s called reinforced learning, actually. The reason why it’s reinforced is because the machine is not that smart. They make a lot of mistakes. The good old story is that, you know…

Damian: Wait, what does…reinforced?

Stephen: Reinforced learning is simply a machine that will learn from its own mistakes, and then it gets better. Anyway, what it means is that, yeah, that answer is not optimal. So next time, let me do better. That’s what the machine does. The fun part is that humans still have a place in it, because you have to tell the machine, “You know what? That was wrong.”

Damian: Yup.

Stephen: Eventually, they don’t need you anymore, but you’ve got to set it up. And so—again I’m going off on a tangent here—but future analysts should be the people who can tell a machine what to do. And to do that, you’ve got to prescribe what kind of analytics you want to do.

Damian: And that’s, I think, where BI reporting comes in there.

Stephen: Exactly. It has everything to do with modeling, machine learning, AI, however you want to put it, but the fact that you have to have these options—even the machine, by the way, has to know. What kind of analytics do you want me to do? The machine will ask you that—or more like, we have to tell it that. So this is a really important distinction.

Damian: All right. I think we brought that all around, even with me taking you off the reservation…

Stephen: I love to go off on a tangent and come back to the…It’s fun that way.

Damian: Yeah. So the last movie that I watched when I was on a plane was Castaway, and it starts with the plane crash.

Stephen: You don’t want to see that on an airplane.

Damian: Yes. So that was like the opposite…My wife is like looking at it, and she’s like, “Why are you watching that?” I don’t even know why they have that as an option.

Stephen: Yeah, you don’t even…That’s a bad match…

Damian: So anyway, the next type of analytics we have here is descriptive analytics.

Stephen: That’s right. So let me go back to the whole evolution of BI reporting—and, by the way, when people say big data, a lot of analytics related to that is just BI reporting, what’s going on. Yes, sure, you can retrieve any transaction record that happened ten years ago in point five seconds. It’s in your display. It’s all part of the dashboard and retrieval, data retrieval. But if you go one step further, you want to know the surroundings of such events. Okay great, I’m getting all this traffic in a New York Madison Avenue store. Who’s coming? And I can guess, by the way. Oh, maybe they’re rich housewives or people with a mid-to-high income, and maybe people with kids, or people who are grandmothers. You can guess.

Descriptive analytics is about describing who they are with real data. Like: What is the median income? What kind of neighborhood is this? Or, what is their income level? What kind of car do they drive? So they can understand what you’re dealing with or who you’re dealing with. So in our product we have all these clusters, by the way. That’s a perfect example of descriptive analytics; you can even get the picture of these people.

Damian: Yeah. Tell us more about what it is you mean when you say “cluster”?

Stephen: A “cluster” is a bunch of people who look alike. Without any clear purpose, we just put people together in groups and, you know: “These people go together well…” and that type of thing. Typically, if you put the whole country’s full population—the United States—you find about fifty to seventy different or distinct clusters. And, by the way, there’s no solid number; it’s just analysts pick a number and say, “Oh, do I want seventy or fifty…?” You can just make up…

Damian: Right. How similar or dissimilar are these groups? How granular are you going to be?

Stephen: Exactly. So if you go seventy, that’s the maximum, I think. I have done hundred and eight once, but that’s a lot. And that’s why people have soft clusters or superclusters, so they can go up and down the scale. 

Damian: It’s kind of related to, like, a persona.

Stephen: Persona is a little different. In fact, we should have a separate conversation just about that.

Damian: I’m sure that would get a good response.

Stephen: I have a forty-five minute webinar just based on that topic. So we can talk about that later.

The cluster is really lookalikes. It’s like, who are likely people who look alike among each other? And then, what are the major characteristics of these people? So we can describe them by their occupation, place of living, home ownership, income, all that kind of stuff. All those are—either clusters, or what we call “customer profile”—they are descriptive analytics. Who are we dealing with here? And those things are very useful when you write copy, for example, so you can even see the profile…

Damian: Right, for your message or offer…

Stephen: Exactly. “Oh yeah, I’m dealing with mostly housewives with kids. So this is what we’re going to offer. This is what I’m going to say. These are the kinds of pictures that we’re going to put in the e-mail.” All that messaging is stemming from descriptive analytics.

Damian: Right. It can inspire hypothesis and creative problem solving. Things to test.

Stephen: Exactly. In fact, I’m glad that you mentioned creative. It becomes a common language among multiple players in the marketing world. For example, in marketing, of course there’s a sponsor, there’s a merchant, right? There’s a marketing agency who do the creative work and write copies and whatnot. And there are data people, like ourselves, who have to really distill the data and group them into these things, right? It becomes a common language. When we describe a cluster of people and say, “Oh, these are an up-and-coming young population.” We just name it, right?

The way we build it and the way we use it could be different depending on who you are, but it becomes a common language that we can all share. You will immediately understand “Oh yeah, that group is what we’re talking about when we talk about…” So when you do targeting, you pick that group; when you write a copy, imagine those people. So it becomes a common language among different players in the market.

Damian: So the next one we have here, predictive analytics—this one is very, I would say, you can see it; it’s being used a lot more by everybody. I don’t think everybody sticks to the rigid definition of what this is.

Stephen: No, they don’t.

Damian: It’s kind of all over the place. I know it’s a hot button for you, but why don’t you talk a little bit about what predictive analytics is.

Stephen: Predictive analytics, the word itself, is about the future: who are more likely to do something? But in the age of big data, it’s not just about future prediction. It also fills in the gap of incomplete data. For example, let’s say that you are promoting some HBO special called, say, Game of Thrones. It’s a popular show, and you want to feature this thing when you push your cable service. I’m just making an example—but, by the way, it’s not completely fictional.

Damian: I’ve watched the entire series; I binge-watched it in, like, about a month. And now I know that it’s like a year and a half ‘til the next one comes out.

Stephen: You have to watch it again before the next one comes out.

Damian: Exactly.

Stephen: Okay. Now if you know that somebody is into Game of Thrones, that’s easy. Just treat them that way. Show more pictures from Game of Thrones and you’re done with it. But the thing about the world is that you don’t know everything about everybody. And you know tidbits of different information about people. You may not know if she really likes Game of Thrones, but you know where she lives; what kind of cable package she has. Is she a parent? What’s her education level? What has she ordered in the past on pay-per-view? Or, is she a subscriber of a streaming service or not? What kind of device does she use? All those things are the things that you know. Based on all the things that you know, it can fill in the blank of what you don’t know. Like, I don’t know for sure, but I think she’s very likely to be a Game of Thrones fan. That’s the power of predictive analytics. It completes the picture when you don’t really know.

Now, the future sentence would be: how many units of this computer device will I sell next quarter? Now, that’s a real prediction. Just like, “Will it rain this afternoon?” And just like the weather report…I think we talked about this last time…

Damian: Yes, we did a little bit.

Stephen: But let me just go back to it, so that we remind them of what we talked about last time. When I say 70% chance of showers this afternoon, a lot of variables are in it. Now, the weather forecaster who said that, does he really know? No. But the probability, if it’s high, is worth knowing. That’s what predictive analytics is. So the form of predictive analytics—and you may hear a lot of different words about it. For example, “response model”: okay, who’s likely to respond, right? “Affinity model”: who has that product affinity? Who was an early adopter? Who’s, you know, more likely to travel internationally for business or leisure? All those things are affinities, right? Who’s more likely to buy a fashion brand versus a generic brand? That type of affinity model, those are part of predictive analytics.

And of course, trend prediction: who’s more likely to stop buying from us? Now, if you predict that, that’s value right there, because you prevent turn by proactively reaching out to them before they actually stop buying from you. And those have a real monetary value attached to it. And I talked about the forecasting; that’s quintessential predictive analytics. I don’t know it for everybody—and if you know that, that’s what I call customer value modeling. But even on a product level or a channel level: how much will I be generating next quarter for this product line? How many clicks will I generate? How many call volumes will I have? How many actual units will I sell? And how much money I’m going to bring in. With that kind of information, you can set up the entire prediction for supply channel management, channel management, budgeting; all those things come into play. And so a big manufacturer spends a sick amount of money for some prediction like that, because every chance that you have, you want to predict the right number; not over-prediction or under-prediction, because you want to maintain just the right amount of inventory. And that kind of a power is already in play for most of the companies.

Damian: Yeah, I agree that it will be transformative.

Stephen: This is—by the way, there’s a book called…oh, I forget the title. Anyways, it’s basically how you win business, how you win in business decision-making, by using analytics. It’s really the norm these days. So it breaks my heart when people say, “I tried a model, but it didn’t work for me, and so I don’t believe in models.” That’s like saying that you went to a doctor once and it happened to be a really bad doctor and you just lost faith in the medical industry. Please don’t do that.

Because—and maybe a lot of things could go wrong. We talked about that last time: why don’t people adapt to analytics more rigidly? It’s because they tried once and it didn’t work out very well, or maybe they met somebody who charged too much money, so the model worked, but it didn’t pay for itself. It could be many different reasons, right? But—as the technology grew and the database got bigger and more things are automated these days…and let me tell you, what is machine learning really? Do you really care if the machine builds a model or a human being builds a model? The result doesn’t care. But what matters is that it’s cheaper—because human beings are touching less and less and less. You just teach the machine and let it go. And it’s faster, because there’s no human intervention. Machines can work twenty-four hours, so that means you can react faster.

Damian: It could potentially be more effective, if you have better data…you can crunch…

Stephen: Of course. Exactly. More immediate data. It’s not just better data, it’s just immediate, that…

Damian: Right, right. Also more complete. You know, you can maybe take in more variables.

Stephen: Yes, more variables. In fact, one of the benefits—if you ask an analyst, one of the benefits that they talk about is they are reaching into data types that they never imagined touching because it’s automatic. Hey! Let’s just let it all loose. However, what I’m trying to make a point about is: all these are in the realm of predictive analytics. So whether you start with a one-off model or just buy a premade model from off the shelf— and that’s okay—or just go full-blown machine running, the idea is do you know what you want to predict? Now that never changes. If you know that, then what you get out of all this predictive analytics is much better, because it fits your goal.

So I think that the first thing that you have to do—and all this, by the way, is true for all the different types of analytics as well. You’ve got to define the goal first. So that when you say, “Oh this model didn’t work…” You know how many times I audited the whole modeling process and the model itself was the only thing that was right mathematically? But maybe the input data was all off. Maybe the target was defined in a really strange way.

There are many examples, but I’ll give you another example I’ve seen recently. There is a gift catalog that we deal with—I cannot divulge who I’m talking about, but let’s say that they have a very big dichotomy in their customer base. There are people who buy, like, $2,000 gift items, but you look at the customer value and about $300 is the average value of a customer or transaction. If you go after the best of the best of the best, you’re going to miss out on the majority of the customers, because the model is predicting the wrong thing. And the average between the really, really big-ticket item buyers and the regular $300 item buyers—there’s no such thing as average between those things. If you target that average, you’re shooting at a phantom target. And people blame models if that doesn’t work. But the real question is, did you ask the right question?

Damian: Right. There is no average customer.

Stephen: There’s no such thing. So knowing all this, by the way, knowing that there are multiple pockets of customers in your base, is descriptive analytics. And knowing what to target, and going after each of those things, is predictive analytics. Now, I’m just giving different names to it, but people have to understand that there are different kinds of methods to get to the bottom of all these answers. This is not just one-size-fits-all, and you have to go through these steps to get to the right question and the right answer.

Damian: Which is a good segue to optimization analytics.

Stephen: Optimization. That’s a cool one, actually. If you talk to marketing agency people, when they say analytics, that’s all they mean, by the way, because that’s what they do. Think about it this way: this is a multi-channel, omni-channel world. You have a limited marketing budget. How should you spend your money? That’s optimization. What combination of things will yield the most revenue or profit? Because, you know, it costs a lot more money to run a TV ad; there’s a cost element to it. So, making sure that you… say that you have a $10 million marketing budget at your company. You’ve got to do these things right, because by increasing, say, all your budget on one channel may not yield, because maybe you hit the plateau already and you don’t know about it. And so— you have different names for all these things, by the way. If you look at it from the budgeting perspective, it gets called marketing mix modeling. If you really want to go backwards and say, “You know what? Let me just understand what worked, what channel yielded the most return,” that’s attribution.

There are many types of attribution, but mostly if you talk about “attribution,” that covers every channel, even the untraceable, non one-to-one channel. It’s based on modeling; it’s called attribution modeling. Those are, by the way, similar things, but the equation is backwards. But those are all part of the optimization. And other names are budget optimization and similar things; if you go beyond channel optimization, it’s budget optimization. So those things are done in the name of analytics, if you will. But depending on who you talk to…if you talk to a one-to-one marketer and you say “analytics,” they think, “Oh yeah, you’re talking about predictive modeling.” Where if you talk to the big data guys, they think that it’s reporting and BI trace and all that. If you talk to agency guys, then they’re thinking about optimization.

Damian: There’s an episode that we did with Gary Beck on the marketing mix, and there’s some pretty interesting things there. Actually, I ran into this…literally, I think this morning, I was looking at one account and we had focused a lot on new customer acquisition. And this was using, let’s say, like, a paid medium. And, you know, we were getting target acquisition at the price that we wanted. And it was a good relative investment versus some other channel. Right? But one of the things that we found was, because we started to ramp up new customer acquisition so much, new customers did something that existing customers didn’t do as frequently. And that thing was leave reviews. So, for example, the level of review capture that we got went through the roof, because out of the mix of customers who were being brought in, so many of them were new customers, and they exhibit that behavior of leaving reviews. And that review ramp-up actually made the brand start ranking way more in organic search. So there was actually this interesting phenomenon that happened, where we were acquiring new customers at a good price, but then it actually tipped the scales for organic earned search in Google and Bing and stuff like that, to start ranking where it actually—there was a much higher net benefit across…you know, almost like an integrated search channel there. It was really interesting. So there’s always sometimes these, like, synergy events that happen when you look at, you know, like, optimization analytics.

Stephen: Actually that’s a good segue. Whether you deal with attribution or regression or back-end analytics of any kind of campaign, one thing that I must point out is that all these things are directional. They’re not baseball scores. So we need the attribution, for example. Oh, by the way, I have seen political organizations where, if they’re standing from an old catalog world, they give all the [tie scores?] to catalog first. If you go to an online company, they give all the credit to online or lost touch, and then they don’t care how many catalogs they send previously to create that lead.

Damian: Oh, I’ve literally heard clients say, “Everybody’s telling me that we’re up, but we’re not.”

Stephen: Yeah, right? So it’s also…there’s a political element. But having said that, we are not living in some isolated world. We are not. A lot of things are working together like that. So when you take an economics class when you’re in college—I don’t know about you, but this bothered me a lot…

Damian: I know where you’re going with this…

Stephen: They say, “All things being equal…”

Damian: I knew you were going to say that! Yeah.

Stephen: “If the demand goes up…”, and blah blah blah. But I say, you know what? When does anything stay equal?

Damian: I’m guilty of actually using that phrase a lot. But—you’re right.

Stephen: Oh, you have to isolate the factor. But I understand. That’s what modelers do: we have to isolate the dominating factors; we have to explain things, right? But again, people have to understand that all the results of reporting analytics—solution, recommendations—are directional. Nothing is carved in stone.

Therefore, one thing that I want to leave behind is analytics is part of science. What is science? You make a hypothesis. You try different things. You prove or disprove and make it, as a theory, go on. This is a scientific approach. We’re dealing with numbers and figures using real data. So therefore, when you make a hypothesis, we’ll set up a simple AB test. The first big starting point of AB testing is knowing what to test. That’s science right there. Now, when you go through all these things and the different analytics, and you make a conclusion and you want to reinforce good behavior…Hey, you know, that’s all science, but just mind yourself, please. This is part of the science, not some magic; and it’s not some baseball score that this one thing or the other is definitely higher than the other all the time. You’ve got to try them out.

Damian: Thanks so much, Stephen. See you next week.

Stephen: Yes, let’s do that. This is fun.

Damian: And I think we’re bringing a guitar next week, right?

Stephen: Yes, let’s do that.

Damian: Okay. [music] If you enjoyed today’s episode, we ask that you please leave a rating and write a review. Or better yet, share it 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 BG team today.