Does Your Purchase History Tell Companies All They Need to Know About You?

I’m going to give you superpowers so that you become rich by predicting what people are going to do next. I’ll give you two choices and you can only pick one First choice: I’m going to give you the superpower that whatever the person sees on the website or store you can see everything. In other words, it’s like walking into the store in a mall and you’re going to install the camera on their shoulders and you could see everything what they see. The second choice would be I’m going to just give you her past purchase history. Which one would one you pick? The answer is number #2. Find out why in Episode 23 of the Inevitable Success Podcast. Below is a lightly edited transcript of Episode 23 of the Inevitable Success Podcast.

Transcript:

Damian: We’re going to have a discussion with about what does what you buy say about who you are from the marketing perspective. So how much can we learn about somebody based off literally the products that they purchased? Stephen: Well the first thing that you have to do is to convert such product description into the descriptors of buyers and you may say that “Oh that’s crazy it’s the same thing”. No it’s not. Damian: Give me some examples. Stephen: A very simple example, and by the way, this is stemming from…let set up a little bit of why we value this. It’s not the product data that they we’re interested in. What we’re looking for is a product purchase data because if nobody bought it, who cares? We only look at the things that people bought. So premise number 1: we’re not doing this for better taxonomy, for inventory management system. The premise of this discussion is that we’re doing this to sell more things to people by finding the right propensity and affinity of products by the people. Example: We have a product, let’s make it a supermarket item as an example. You have food labels on every food item right? So for inventory management you only look at it as okay that’s sugar whereas this cookie is gluten free or peanut oil free or fat free. Or it could be organic. Well that’s another way to throw it in there. So we look at that as a description as a product but if you flip it to “Oh she’s only buying organic food” that tells you something. Right? She’s only buying sugar free or what they call “diet items”. Maybe she’s into weight loss. I don’t know. I don’t want to offend her by saying it explicitly but I can gently nudge some products like that if I make an offer of something. Just by looking at the same thing in a different angle okay? So instead of just looking at the food label you start describing buyers of such products. I’ll pick another example: baking soda & baking powder. Well how many use of those can you think of? Damian: First I was confused and then I started thinking about it. Stephen: Well you could use it to bake things right? That means that person who bought it is a baker right? But then a housewife could be using it because their refrigerator stinks. I would put an open box of baking soda in there and then what is she? She’s not a baker yet. Or you could be using it to brush your teeth for personal hygiene. Damian: You can use that to clean too. Stephen: So the same item can be used in multiple ways. And if you understand the purpose of it or the context of it then you can describe the person’s attitude better. So I’ll give you another example. Awhile back we were helping out all kinds of different merchants by putting together a lot of transaction data in one place. When you do that you have to categorize all these items into proper buckets because you’re talking hundreds of thousands of SKU descriptions from one merchant. At the time we had about fifteen hundred contributors of the data. Can you imagine how many SKUs we had? So the trick that we used was let’s not even bother with the products that didn’t sell a lot because we are, again, in the business of selling more things to people not to create perfect data taxonomy. So that made our work a little easier but you still have to categorize all of these things into the right bucket so they can be used for prediction business. I’ll give you example: You know the weather station that people sell? You hang on your wall, it’s really made of really nice oak and it tells you the temperature, barometer, humidity, and all that kind of stuff. You can buy this thing from an executive gift merchant or a catalog or you can buy the same thing from a nautical shop. The context is completely different. What you don’t want to do is ignore the context and then just decide that one product should be categorized into this one thing and we call it a nautical item. So for a person who bought this thing casually for their boss’ birthday are you going to create a category for him about how he’s a nautical guy and then start sending him all these ocean-related things? He will never respond to it. If the goal…and in very last episode we talked about goal-oriented mindset when you play with the data. It goes down to the simple categorization of a product because your goal is to sell more things not understand products perfectly. Even the same product, depending on what you’re trying to do, should be named differently. Now you may wonder why we get into all this discussion so let’s go back and review what we talked about last week. Last week we talked about what is the most powerful predictor of all or what’s the most powerful data. We talked about behavioral data, which is transaction data, being the most powerful predictor last week. And so let’s break down a little bit more. What is transaction data? Transaction data is very simple if you break it down. It’s about who bought what, for how much, when, and through what channel. That makes up the human being. Other behaviors are like, you know, all these clicks and views and all that. But I could still go back to the fact that if you really want to predict what somebody is going to buy, not just talk about things because talking about the thing….if you ask people “Do you like movies?” they all like movies well they all like music. But if you start asking questions like “How many times did you actually pay money to download music last month?”the answer is very different. Suddenly you don’t let music that much, right? “Oh I love movies.” “Well how many movies did you pay to watch last month? How many movies did you download? Do you have a paid subscription for Netflix?” The answer is very different. I mean we talk about attitudinal data last month as well as last week. If you ask people “Oh do you believe in green causes?” most people say “yeah sure why not?”. It doesn’t cost me anything. So let’s flip it. Did you pay money to change your lightbulbs or did you buy a solar panel? Did you install anything which is going to save energy? And the answer is very different. This is why of all the behavioral data and attitudinal data, the most powerful predictor is your dollars. How did you spend your money? In that transaction made of who, what, when, how much, and what channel is the most powerful predictor. But why are we talking about breaking down the pure product descriptions? Because collecting such a transaction data is not enough. You have to break it down. You have to distill it so that you convert what we call descriptors of buyers. Descriptors of a transaction get converted into descriptors of buyers and this is the first step. Damian: You know, it’s funny and there’s also probably some information in not just one product by itself but products purchased together. You know that thing that you made me think of when you use the shopping analogy was you know if you buy a dozen eggs, maybe you’re making breakfast maybe you’re making a cake too. If you buy other ingredients that are to make a cake then that might tell you something. Each individual product by itself can’t tell you that this person is a baker but that’s why I thought of that because you said baker. There are probably examples of that in most retailer’s data that two things purchased together maybe tells you more than just the descriptors of one product by itself. Stephen: That’s actually a good conversation because what you’re describing is what Amazon casually does as a “oh you bought this and you also want to do this” right? That’s based on product basket analysis. Another way to say it is what they call collaborative filtering: You bought this, here is related product. Of course, you have to look at other things but if you don’t have those things? What if you just bought baking powder and nothing else? I would look at the context of it too like from where and at what time and with others is makes sense. What was a total basket volume? I don’t know what else is in there but how much did she spend? Damian: I don’t know for sure but I could guess. It’s different for every business. I think the important thing I hear is to be curious to look for that. Stephen: Exactly, that’s what I’m trying to convey today is that it’s not about… And by the way I could write a whole book about product categorization and I’ve written a lot of articles about it. If you’re curious, I’m going to keep writing about these things in our own blog post anyway but today’s message is very simple. All these things that I’m talking about are about attitudinal change by the users of the data and don’t just be a hoarder of the data and hope to God that somebody shows up and fix all your problems. You’ve got to really think about these things while we’re consuming the data. All these other steps we have to do. I’m not saying that you are the one who does the data categorization and data modeling and building a database and actually building a model to predict who’s going to buy what. I’m not saying that we should all do all of these things even among the professionals, by the way, we divide this work among multiple different people. There are master manipulators of the data and there are business people like me who create a whole taxonomy of a different kind of a product category just for the prediction business and there are people who develop the software. There are people who build models, you know, statistically train people who build models and segments. It’s a combination of things and even when, by the way, in the future that machines will take over a lot of these things automatically. I dare say that you’ve got to give the machine the purpose. The machines are very good at categorizing these days but they don’t ask why I have to tell the difference between cat pictures and dog pictures do they?We made him do it. We are the guys who made them do it because you know why? I have a dog. I think I’ve said this in past episodes. Facebook knew that my dog picture was a dog. On his page. His age is displayed in dog years. Why? Probably because somebody thought that it’s funny. The machine will not know it’s funny or why we asked it to do it. It just did. Damian: They must have been pretty confident they didn’t put any humans with dog years. Stephen: Well the fur on his face is dead giveaway. What’s interesting is that the whole ingenious idea that somebody thought about it. I think that’s the human capital. Damian: This is actually an exercise physically done and it adds value almost every single time I do it when I’m looking at coming up with ideas to know the customer better in the effort to market more effectively. So what I’ll do is I’ll grab like one specific day. Maybe it’s like 100 transactions that day. I’ll literally look inside the database at every single product that that person bought and what they bought together and try to see if I can synthesize commonalities or see if I can understand what that person is about. Stephen: You could guess and I don’t think that that is… Damian: It’s not the same thing as like machine learning. Stephen: What you are describing the process the machine goes through. What you do is very valuable. Sometimes you have to look at it, right? But you don’t have time to look at 100,000 transactions every day right now. Or it could be millions of transactions. If you’re right if you know the process you can repeat it. If you keep repeating it, if you know methodically how to repeat the process and get the end result that you want who cares if the machine does it or a human being does it? We use a machine because it’s much cheaper. If you put hundred thousand people into the auditorium or some football field and say that “you know if you feel that way bring up the red card” and you just count those manually, if that’s cheaper you’ll do it that way. It’s not for now so use machine learning. So again the purpose. Let’s go back to that purpose. It’s very important. Remember we talked about predictive superpower? I think the whole notion of having data just for the sake of data and without any plan to further refine it, including the way we are going to categorize these things… …well what can I say? I talked about prediction and past behavior as a predictor. I say that for a reason because I’ve seen, first of all, so many models that are very functioning and powerful and it’s based on both variables. So let me ask you this. This is a camera on your shoulder question. OK. Let’s say God Almighty comes to you personally. Yes I’m going to give you superpowers so that you become rich by predicting what people are going to do next. I’ll give you two choices and you can only pick one First choice: I’m going to give you the superpower that whatever the person sees on the website or store you can see everything. In other words, it’s like walking into the store in a mall and you’re going to install the camera on their shoulders and you could see everything what they see. The second choice would be I’m going to just give you her past purchase history. Which one would one you pick? Damian: I’m guessing from the intonation I should take…number two Stephen: I’ve been picking number two and had no regrets. Ok fine, I put a camera on her shoulder and she looks at shoes. Maybe she just love shoes. Will she buy it? I don’t know. Maybe she’s killing time. I just read an article by some old friend of mine on how to survive in the Age of Amazon and all that. What he wrote is kind of interesting. People still get entertainment value out shopping. It’s kind of interesting, isn’t it? If the act is purely shopping, of course I always buy shoes from Zappos because I can find my perfect size, width, and color within five minutes. If I don’t like it there’s no risk, I’ll just return and get a new one. Fantastic, but why do we go to the mall? Because I just feel lonely? I don’t know. Maybe we just like the act of having to talk to other human beings.   Damian: Retail therapy. Stephen: Is that what it is? The point is, camera on your shoulder is great, if you want to maintain the whole movement data and then design the best mall layout. Yeah that’s what you need. Going back to the goal-oriented data analytics, that’s the kind of data that you need and that’s why. If you’re building a plan for what’s the next best product for next season you cannot categorize products the way we talk about it by buyers. No, you really have to look at the product as the product if the prediction is about the product. Conversely, if you’re in the business of going back to predicting what people are going to do next you’ve got to look at their past history. Now the joke is this though, it only works when the person stays relatively stable but luckily for us most people are predictable, including myself. Golfers will buy golf stuff. Fashionistas will continue to be interested in fashion. Bargain seekers will continue to look for bargains. I’m a bargain seeker in some categories. One day I may…by the way I stopped playing golf so there is all this golf data that I left behind in the past. If they’re still sending me golf offers they won’t work but that’s okay. This whole analytics thing is not about being right every single time. It’s about increasing the probability of selling something to somebody so that you don’t waste all these communication opportunities. That’s what this is about. So therefore, again, a goal-oriented mentality is the first thing they all have to go through. Damian: Love it. So you know I’m kind of ending this if you are what you eat… Stephen: I don’t know, are you what you buy? In you way, yes. I’m trying to downsize my house and I looked at all the things that I’ve accumulated over the years and I’m like “oh my god…this defines me”. Maybe I want to throw it all out and reboot. I don’t know. But yeah we are. We are what we buy. Yes. Damian: All right. Till next time.