Building a Profitable Target
- There is no average customer. The average of your biggest spenders and your most frequent spenders does not exist. They are called your phantom targets. If you just go after the average based on those descriptors you’ll never hit anything.
- When targeting your highest-value customers don’t just guess what the cut-off is. Humans like round numbers and if you just make a guess of say $500 you could be alienating a large portion of your highest-value customers.
- Even if you spend a few $1000 to build a model it can still fail if you define the wrong target.
- The first question to ask when you are going to build models or segments is “Will this target the right people who will make me more money?”
Below is a lightly edited transcript of Episode 24 of the Inevitable Success Podcast.
Damian: We’re going to have a nice conversation about how to get better results in basically anything you do with marketing. Because we’re going to talk about a topic that is probably the most important thing to find your target. So I would start off with how do you define your target?
Stephen: Well not based on your gut feelings. I’ve seen disasters as a result of somebody’s gut feeling. There are a lot of examples actually. Let’s pick a innocuous example of a gift catalog or a gift merchant. They may sell really expensive gifts like those executive gifts that would cost more than a thousand dollars, sometimes two thousand dollars. And they always have the impression that that’s the only thing that they sell. So what happens there is that, “OK let’s go out to really rich people or people who can afford these things”. Now that’s your gut feeling if you look at the data though you may have a very different picture. You have of pockets of different types of customers in your base because first thing they were look at was what is your average customer value? Believe me it was less than three hundred dollars, they’re not selling a thousand dollar item all the time. So you may have multiple pockets of people in your base and you will never know.
Damian: And that’s just the phenomenon of averages versus median.
Stephen: Well it’s worse than that actually. By the way, I’m glad you brought that up because what that creates is this thing that you have two different pockets of customers. One being a big whale, spent a lot of money but they don’t come often. That’s one example. The other kind of a customers could be frequent customers that just love to visit. They love to browse and when they buy they buy small items but they buy a lot. Now the customer profile of these two targets could be vastly different so if you average them out and say “hey this is our average customer”…guess what? It doesn’t exist every year. The average of those two pockets of people do not exist. I called them phantom targets. You go after that based on even if you looked at the data if you just go after average of these things you’ll never hit anything.
Damian: Yeah that’s a great concept in general just you let that sink in. There is no average customer.
Stephen: There is no average customer. You have pockets of different types of people but no average. That’s why we call them segments actually.
Damian: Yeah. You know I had this conversation with Mike our founder a while back to where we said if you actually marketed to the average customer you’re literally alienating every customer that you have.
Stephen: That’s exactly what I’m saying. Yeah or your don’t hit anything. You’re hitting at the phantom target.
In other words, let’s imagine a shooting range. If you hang the target in a wrong place it doesn’t matter if you hit the bullseye. The target is in the wrong place. So targeting is that important and knowing your customer base so that you build intelligent targets is very important. And by the way, I’m not even talking about sophisticated situation where you’re going to adapt to the statistical modeling to go after all these things. Right. We could do that but the point is when people think about these things they always think about the methodology first. Are you going to do the segmentation or are you going to the real life regression modeling? Don’t bother.
I’m going to explain who they are to my list vendor and just get it that way. Well some are more effective than others. There’s no doubt that statistically built models are the most accurate thing you can do. But before you even think about methodology you’ve got to think about the targets first.
Damian: So how do you peel the layers back to figure out if you know the different types of targets within your average? How do you even like start?
Stephen: How do you start doing that? Well, it sounds like self promotion because we’re represented by BuyerGenomics. It’s on page of our dashboards.
I’m not saying that you have to use BG but it has all those reports in it. But the point is you have to look at some reports in other words, what are their distribution of the dollar banding? Not just average but each banding.
Damian: What do you mean dollar banding?
Stephen: Let’s just count how many people we have in say the 100 to 200 dollar range.
Damian: Do you mean realized value?
Stephen: Actual spend value. Look at the frequency of the good ol RFM. Well let’s take a look at it. Maybe you have inversely related frequency versus value. I don’t know that. So you’ve got to look at the data. So in other words, when you list all the transactions and customers in terms of dollar banding you may see a general curve like a bell curve. You may see that or it could be skewed towards more expensive or you may see like double peaks just like a camel’s back.
Stephen: You don’t know what you’re going to see until you see that data and everything that you say before that happens is your guess. In fact, I met an apparel company, pretty big actually at the time. And what’s amazing is that when I asked him I said OK so by the way when I meet with anybody the first question is that okay, why are you losing sleep? And that’s my first question because I want to formulate the solution based on their problems to solve first regardless of the long term goals.
He started talking about his customer profile in a really vivid fashion. That was great. But you know what that was? He was just visiting his own stores a lot and looked at people and guessed based on what they look like. Now there are things you can tell based on looking at people and there are things they cannot tell just by looking at them because how do you tell a person what is a person made of? We are all made up our past behavior. Now what is a behavior? It could be dollar that we spent, things that we bought, things that we browse, and frequency of those things that we do. And what do they look like? That’s demographics isn’t it? Oh they’re women or they look like mothers or they look rich or they look like they’re driving a nice car or whatever it is. When you peel off the real data you may or may not confirm your theory and more often than not, you get a very different picture. Another example is when I was really much younger but I was consulting this is not possible these days.
But it was a tobacco company, a big tobacco company and they were doing the similar things and they created a new brand for say, what they call “affluent women” and they want to understand who their customers are. So we ran a profile. We don’t guess these things, we run a profile based on their demographics first because that’s the data that’s mostly available.
And we found out that they are not as rich as their competition’s buyers and they are not all white collar, they’re mostly blue collar. And actually this customer was really upset. “That’s not our target” Well this is based on the real data sir. That’s number one. Number two their money is not good for you? This is your base. You got to chase these guys, you’ve got to mimic this pattern, you’ve got to find more these people not follow somebody else’s target.
Damian: It’s almost deep inside, what you do defines you more than what you look like.
Stephen: Well that’s why in the prediction business the most powerful predictor is your past behavior. We talked about camera shoulder last time right? Why is that? I would rather have everybody’s past purchase history to predict what they’re going to buy next. By the way, that’s if that’s the goal then I would rather have that kind of data not where I want to collect everybody’s every move, every birthday day. OK that’s fine. If you want to redesign your store plan or what product goes to where in the store then the movement that is very important. But if your goal is to sell more things to people that you can identify than your past behavior is the best predictor.
Damian: So you know we looked at our database. We found no hints that there could be different types of customers and they’re looking for peaks and you know, we can even include in the show notes some examples of pictures or what you’re talking about maybe from the reports screen. So now we have a hint that there’s probably different buyers here. What do you think we do next?
Stephen: Well they’re going to go deeper basically you have different pockets of people and, by the way, all these things that I’m talking about to find out what pockets you have could be called…let me just throw out interesting names that you may have heard. Segmentation that’s not the end goal. That’s the first step for us and personal profile that I talked about. Once you find multiple pockets of people now you’ve got to find out how different they are. The imaginary report lays on top you have all these columns representing different pockets that you imagine it to be. So it’s a hypothesis at this point.
Let me look at really valuable customers, frequent customers, new customers, dormant customers, tenured customers I’m just going on here. But let’s just say that all I really want is to compare how different people are in terms of all the variables that we have. Let’s pick two or three. Like let’s compare big spenders, frequent buyers ,and new customers because that’s the target that you want to go after. They’re in the same behavioral pattern and are different and the only way to look at it is to look at other data against those segments and see how different they are. Some of the findings will be interesting actually meaning that you could have a target within the target that some segments are so big that you need to divide this into four groups.
Damian: I would also have to imagine, because I’ve been in situations where that’s the case, it actually happens quite often happens all the time. I think that you have to figure out how different they are. Where that what you would do adds value and how big is it that you know it’s worth taking that extra step.
Stephen: That’s exactly why you’ve got to look at the size of the segment.
Damian: It’s almost like materiality on it and how big the difference would be in outcome.
Stephen: You know that’s exactly it. In fact, even when I face a problem, say errors in the data that I always have for the size of it because you know, he only has like point 2 percent difference. Who cares if it’s vastly different then? We’ve got to consider those things. So you’re right how different they are is very important. So that leads to the next one.
Everybody goes after say high-value right. OK I want to go to high value target. Well how high is high enough for you? By the way, please don’t say things like a thousand dollars because that’s your guess. Human beings always pick round numbers. How frequent is a frequent flyer? And if you ask that question to the United Airline executive versus just general website answers are vastly different because United Airlines will look at only the flyers and then determine who really flies a lot. If you compare that to the general population and who flies a lot. Hey you know what, three plus per year is a lot but if you get only the base of United Airlines frequent flyers, you’ve got to fly at least ten times a year to be a frequent flyer. So how high is high enough? That’s not your guess.
You get to do some analysis just to go after.,..even you if you want to build a model you have to do this anyway. How high is high by the way, in modeling, you could set it up as a continuous target. But that’s a technical term here. Even as a request of this targeting whether you build a model or not you’ve got to understand how high is high enough. Good rule of thumb would be, OK let’s just take the top 10 to 20 percent or something like that. But even such a rudimentary selection you need to line people up in the descending order of the value and draw the line. So please don’t go to what we call go-to answer: “Oh yeah, spending more than five hundred dollars per transaction” because by saying those things you may go after a fraction of your ideal target. That universe could be smaller with less than 2 percent of your universe. What you’re going to talk to your 2 percent of your base and stop there?
Damian: I mean I’ve even seen it that as a problem outside of customer data. When you’re doing different types of targets I mean we always kind of I guess like rag on household income as as a target. That’s a good one. This is one where if you say I want households that make over 100,000 dollars a year you are talking about such a different household.
Stephen: You know in, let’s say a rural area than you are in Manhattan. Then you have to get into what is a real spending power or buying power? Yeah that’s a very different question.
Damian: You know a family that makes eighty thousand dollars a household in the Midwest could be actually pretty well-to-do but in Manhattan they’re actually on government support.
Stephen: That’s exactly what it is all about spending & buying power. But let’s go beyond that actually. Why did you pick income? Because it’s familiar. Right. By the way that’s number one reason if you really want to target really drop the preconceived notion and trusted that and you get to get a lot of data.
And by the way that’s what we do. We don’t make decisions for our customers. We make customers have all the data in your own hand to make these decisions.
Damian: So if you’re doing modeling this is pretty obvious I think, the process that you go through right. You know the truth becomes apparent like the math sets you free from some of these potential pitfalls. If you’re not modeling what are some of the things that you can kind of help make sure that you’re being more robust in your approach?
Stephen: Actually I could argue that whether you built a model or not the process is the same. If you set up a wrong target and build a fancy model and you may have spent a few thousand bucks to build a model like that and I know a lot of cases where the models failed miserably. Why? Because you defined the wrong target. When you ask how did you define the target the client says “well…you know people who spend more than a billion dollars a year”. Well what if that’s 1 percent of your base, that’s your target? What if you have another pocket that could be more responsive? Have you ever thought about that by the way? In this case you have two models anyway. This discussion whether you build a model or not you have to have this discussion of what is the target? The first step is always segmentation, you look at the data and see who sticks together. The second step is how different are they. You’ve got to have that conversation.
Now if you have enough numbers, by the way, modeling is not that decision of “oh I want to be fast and do models”. No you don’t. That’s like saying “well I’ve got to have a surgery because I heard that that’s the most advanced thing you can have”. Well what if you don’t need to have a surgery? You’ve got to think like that. But if you do need the modeling and here’s why you need the model: your ideal target in your base is so small they have to mimic those guys to go after somebody who look like them. That’s a business goal not some fancy modeling talk. We build models not just because you want to go for a model but you need to have a model. When you have a small pocket of people to mimic those guys to be successful then you have to mimic those guys. Modeling is nothing but mathematical term for mimicking. It’s really not a difficult thing. In fact, I could argue that let’s say that in this data abundant world, I could think of more than a thousand variables that describe each household and individual.
Our base alone provides about 400 variables to predict a person based on our product and channel data. That’s a lot. So if I ask a human being “OK so now mimic your target” well good luck because that’s like us giving a box of crayons with a hundred and twenty eight colors to a three year old and guess how many colors that they’re going to use. They use about two or three colors. Now you wonder, why didn’t I just tell some crayons from Applebee’s or somewhere? I mean you’ve got to really think that way. I’ve met a lot of smart people in my career. No human can think in more than two or three dimensions at the same time and actually consider correlation between those things. Mathematics does it automatically with 10-20 variables and after the consideration of like a few hundred variables at a time. People talk about machine learning all the time but I cannot laugh at the notion that they’re waiting for the machine learning but they are not believing statistical techniques that distinguish learning is it’s the far most advanced form of modeling, seriously. But I could argue that even the machine learning will fail if you set that machine going after the wrong target it will fail and people blame all this model didn’t work. There are countless examples where I showed up as a consultant and looked through the whole process.
Even some mighty big banks by the way and the conclusion nine times out of ten is there’s nothing wrong with a model. Something went wrong before modeling meaning you set up a wrong target or there was multiple targets and you targeted the average for it or there was a continuous target but you set up the hard line in a really strange place and you go after the some really small pocket of people.
Damian: Just define continuous target again.
Stephen: For example, when you say frequent flyer, that’s a number. How many times do you fly? Is two plus long enough? Three plus? Spending level is another example, dollars. Oh OK well our average spending level is three hundred dollars so let’s just be a little greedy in this pick a 500 dollar target. It may or may not work because you really didn’t look at the whole distribution of things you don’t know if it’s a bell curve or skewed curve or whatever. You don’t know that. You’ve got to rely on numbers. So whether you’re a big or small, if you build a model or not, this definition of things is very very important.
Another case in point, by the way, when you have a really large target and you have a very small sample to mimic we all go through the sales process right? Let’s say you’re selling something that requires sales qualification. Likely to respond to the first outbound mailing could be very different from among those people who responded to the email who would actually buy. That could be very different. In fact, even if you use the same set of variables to go after these things you realize that it’s a simple concept. In the beginning, there are under 130 million households in this country and 300 million people so finding somebody who is just like your best buyer, and people call it finding a needle in a haystack. Finding a needle in a haystack it’s very easy. All you have to do is bring a big magnet and look for metals. What I’m talking about finding ideal customers out of 130 million houses or even if it looks like regional target of 10 million. Finding the best guys you need to talk to three to four hundred thousand is as difficult as finding a needle in a needlestack. It’s that difficult. Please don’t think the do this on your own. One day it will be simpler as simple as let’s define the target and let the machine go to work and do it.
But let’s not even think that machine will pick the target for you because it doesn’t understand your business goal yet. It doesn’t know your pain points yet. Is this model for your prospecting to get new customers? Is it to sell more things? Is it to find out who’s going to be valuable and treat them differently? Or is it going to be who’s going to be an outrider….who’s going to quit your service? All those predictions are a business decision. “Oh this is my business issue, this is the model I need even if you don’t need somebody to think like that anyway.”
For example, attrition. You don’t need a model every time, you might need some kind of a simple scoring algorithm. We have it in our BG. It’s not that simple actually but it’s a the scoring algorithm. What do you think that is? That’s a model. So you’re using this thing all the time but as a user what you have to think about is “okay what am I going after and how do I express that what I’m going after in a mathematical term?”. Do this instead of some ambiguous high-dollar customer. You know what, let’s define how high is high. That’s what I mean by continuous target. If you cannot, then there are other techniques to deal with that but let’s have that discussion with a real analyst. If you don’t have that luxury, at least take a look at the distribution report which is available in our report.
Damian: I would say that everybody listening has that luxury because we can have them reach out to you and have a quick conversation you.
Stephen: And let me just confess by the way.
Damian: I wasn’t we’re going to take confessions today.
Stephen: This whole booth feels like confession. Although my last confession, don’t ask me when. But anyways, long enough so that I don’t remember.
The truth is that’s how we designed the BuyerGenomics reports. I can tell how many pockets of customers you have to consider in less than three pages. I never really ask anybody to run a special report and this report is real to you. Yes we’ll have to think about it, look at the report, and all that but I would like to have that journey with our customers. Let’s talk about it’s not just about who has a fancy report and all that. You know what? We didn’t create fancy reports so that it looks fancy okay? Do the things that you make more money targeting the right people. This is the first discussion whether you build models or segments or do the manual targeting. That’s all secondary.
Damian:This was a great conversation. To leave it off there till next week.