DITCH RULE-BASED SEGMENTATION
Even with great advancements in technology for data management, analytics, and campaign tools, holistic personalization is not achieved in everyday marketing campaigns.
Rudimentary segmentation techniques are widely used in marketing, however, segments built with a handful of variables based on users’ imagination are not enough for proper personalization. In a data-rich environment, one must employ more advanced techniques to harness the full potential of data.
Segmentation is one of the most misused words in marketing today, and further, it is used interchangeably with another popular term “persona”. These are two very different types of modeling for vastly different purposes.
In the session, Stephen covers:
- What are the key benefits of segmentation based on statistical techniques?
- Why model at all? What are the key benefits of modeling (over simple rule sets)?
- Data ecosystem around modeling practices. Many efforts fail due to inadequate datasets and improper targeting.
- Key differences between segments and personas.
- How to implement segments and personas in marketing campaigns
Today we’ll talk about how we use the data for one-to-one marketing more effectively. I’m not saying you ditch your good old segmentation tool today. All I’m saying today is that it can be a lot better. We have abundant data out there and there are many ways to improve your performance using all kinds of techniques.
Unfortunately, a lot of people are stuck with a good old rule based segmentation, which is based on, at most, four or five variables. How do we get over it and how to go to the next level and create a what we call “Ultimate Personalization”? That is the hot topic these days but you’re not going to get there by using some rudimentary segmentation so we’re going to talk about how we jump over to that. We’ll also talk about some benefits of actual statistical modeling how to go about it.
Let’s talk about why marketers use segmentation anyway. Is it because your predecessors have done it that way or because you have the segmentation rules laying around? Are you using it as a communication tool? When you say “Yes this is the golden years target those are the people that are on target to sell really high end furniture.” Then when you say those segment name you can talk to anybody in your team or even outside your company when you talk to the agency guys.
We all speak the same language and this is really good for planning. Talk to those folks with you retirement age rich folks would buy expensive furniture or is it about targeting? Ok well I am sending some targeted messages to certain people because the thing about one-to-one marketing is that you should know who you should be contacting anyway by using it for targeting. Now is that the best way? We’ll talk about that.
Messaging is a key reason why we all use segmentation. In other words, when I say “golden years” you all can picture what the segment looks like. When you write copy for Post Office people you can imagine that you’re talking to some old people over 55 with a certain income living in a certain area. You know how to talk to these people so it’s really good for messaging.
We overuse the word “personalization” all over the place but using segmentation is better than not using anything at all. But is that the best option? We will talk about that. One thing I want to make sure that we all understand at the end of this session is segments are not precise. I hear those two terms being used interchangeably but those two are very different things they are different techniques and for different purposes. We’re going to talk about that as well.
The bottom line is that before we touch a single piece of data, we should really define what we are trying to do first. I may list only four reasons why we do anything but let’s just define what we want to do first. Then we can find the best tool sets and make the best usage of data instead of going back to the rule-based segmentation out of habit.
Let’s start from the top: What is one-to-one marketing? When I say these things people will say “Oh I’m not in a direct marketing space” or “I’m not just using email so why do you call it one-to-marketing?”, but like it or not, everything is one-to-one marketing these days. We’ve been talking about how even TV screens are a one-to-one marketing channel. The reason why it’s not happening is not because of the technology, but a coordination of efforts. In the future, even the billboard on the street can be a one-to-one marketing channel.
What we have to remember entering into this world, the good news is that if you have been using the one-to-one channel traditional like e-mail or snail mail or catalog or whatever you do, that discipline still lives on. What is that discipline? When you’re really into this one-to-one marketing it’s a really simple thing. It’s really about two things. One is do you really want to talk to that person and we call it a target? That’s the question isn’t it? Everybody’s being bombarded with all these messages and if you think about your own email inbox right now, most of the email that’s sitting on your personal inbox are what we call “I don’t know highlight all purge all type of mailing”.
Unless some message is really talking to you about something that you’re interested in, you will not open it. That means that all this batch and blast type of a marketing, it’s not really one-to-one marketing. Just because you’re using one-to-one channel and just because you know who you are addressing does not mean that you’re doing the right thing. You’ve got to be more selective and you’ve got to know who the contact is.
Now if you decided that you really want to talk to this person, the second question is what are you going to say, with what offer, what content, what future product? We call that personalization in a nutshell. Without using any fancy words or technology or marketing jargon that’s what this is. Find me a contact and let’s find out what are you going to say or you can offer.
So when you get into this kind of a one-to-one marketing, data-based marketing basically, you’re becoming a data player. So what does that mean? We use a term like data scientist quite casually these days but just because you touch data all day long does it mean that you’re that a scientist or a player. To play this game properly you really have to think about three elements data collection, refinement, and delivery quite carefully.
A few years back this whole notion of big data was a big deal. What was the big data movement at that time? It’s about speed, it’s about the size, it’s about the variety of data. That’s what we talked about. The question is that just because you have a lot of data and retrieve it relatively fast. Yeah sure I can retrieve any transaction that anybody did five years ago within .1 second. The question still remains, okay that’s great but when displaying the history to that person or looking at it yourself, does it really change the way you make decisions about how to contact these people?
Do you know who to contact by doing this or do you know what to say to these people the answer is absolutely positively not? And I think that’s why we don’t say big data anymore. That became a dirty word relatively fast among the real players and in those days but that’s an important part because you’re collecting massive amounts of data. You’ve got to be able to retrieve those data fast and return some basic dashboards (These are how many people who clicked this thing and all that) so you can have a real time dashboard. That’s based on retrieval but does that give you insights in terms of are these hundred thousand people that I must contact for the Back to School mailing that I’m about to do right now? The answer is maybe, maybe not.
That means all this data should go through some serious refinement processes, and by the way if you’re curious why certain elements are in italics, that’s what I call analytics. We’re going to talk about what analytics are next July. But you’re going to go through some serious amount of refinement to make sense out of this massive amount of data. Now here’s the unfortunate truth about the data business, tremendous amount of resources are spent on data collection and delivery, probably because a lot of them are floating around you get to maintain it somehow. So that is a collection. The delivery part is in point because that’s how the users touch and feel the data. This is how we blast e-mail. It’s how we look at the report. This is how we run a campaign or run personalization. There are a lot of tool sets out there and a lot of money spent on the delivery part.
Now, if you do not do that refinement properly including the best analytics such as modeling, segmentation, all those things, the burden of having to break all the rich information into bite size information you could actually use in the campaign management, for example to know that yes this is a pocket of people that I want to talk to right now. The burden is delayed towards the delivery part of it. It doesn’t go away. Without naming actual service providers there are a lot of over promises out there. Like if you buy this email delivery software we will just build a model automatically, we’ll collect the data for you. We’ll even learn from this current campaign and make the next campaign better.
I heard a lot of complaints about how it isn’t working but what can I say that’s what’s being promised and I don’t blame the software manufacturers who claimed all these things. The answer is, yeah is that the right tool to do it and is it even the right steps to do it. I would like to differentiate those things.
So let’s talk about data refinement a little more, and I’m not going to talk about say you know good ole data hygiene all that merge-purge and all of them. But let’s talk about what analytics really mean to you. Well it means a lot of things and depending on who you talk to their definition of analytics is quite different.
So all the major headings I’m sure you heard about the things anyway but let me just go with a high level. The most basic kind of reporting is data analytics or the KPI report dashboard digital analytics all of that campaign performance. You can see that “Oh yeah the best time to send an email is Tuesday morning”, I just made it up of course but you’ve got to do those things based on the data without the reporting you will not have this information. One step further you can get into descriptive analytics and the profiling segmentation is under this heading and at this point you’re describing who is doing all these things.
Let’s talk about, ok so this bunch of customers. Who are they? How old are they? Do they have kids or do they not have kids because they’re too young or to old? What does the journey look like? What is the entry product for these people? Understanding who you’re dealing with, that’s what we call descriptive analytics. Other names are again segmentation, clustering, profiling all those wonderful things.
At this point though you’re still trying to find out what’s going on in the world. Is this the right tool to predict what they’re going to do next. Now that is a very different type of analytics altogether, which is predictive modeling and by the way a lot of folks think that yeah well you know we’ve got to do this in order without reporting I cannot jump to them modeling, the answer is no.
That’s like saying that your doctor if you see a patient do you really try absolutely everything from cheap treatments to the most expensive one in order the answer is no you’re going to kill a patient. Sometimes somebody is in a dire situation or have a significant improvement by doing certain procedure out of order you will just open that patient up. You don’t just wait around until you try every cheap drug that’s out there fresh. So let’s not have this notion that all these things should happen in order. This is just a menu of things that you can do to do what? To meet your business goals.
So the predictive modeling is about the future. What are they going to do? Who’s more likely to do certain things? If there’s a propensity model for personalization who is likely to be a buyer of a expensive home electronics item or who’s more likely to respond to certain offers when they’re not customers yet, so the acquisition effort. Or you want to know the future value of the customer. So you know even before they do a single transaction who is more likely to be a valuable customer or who’s more likely to churn who is more likely to be a dormant customer.
On the whole, demand forecasting predicts the future. Today you can know all the units, clicks, all the value generated. If you know the right number before anything happens you can plan things much better either before marketing, or supply chain management or manufacturing. You need this kind of a predictive model but if you talk to some agency folks they’ll have a very different idea about analytics. The word analytics and they normally talk about optimization.
I have a million dollar budget for marketing. Where should I spend the money to maximize my return investment? Where should that money go and how do I mix the channel that would be our marketing mix modeling or attribution? Okay I did all these things, how did it work? Now of course if you do the old matchback, call it a bottom up approach. But sometimes you don’t have all these one-to-one channels employed and you have to do the top down modeling that’s attribution modeling.
Analytics means a lot of things, I’m just laying out there so that maybe this is a good time for us to start using this word in a different way. Just like we don’t really use the word segmentation casually. So let’s talk about segments for a second. Like I said the title is Ditch Rule Based Segmentation but like I said, you do not teach something that works and segmentation and cores serve a very different purpose anyway. So let’s go down the list.
Segmentation is really more about messaging than targeting. Now it’s a very general statement and of course if you have a very small group of customers that you really want to talk to all of it, well there’s no reason for modeling just build a very simple rule call it a segment. I don’t care. It’s rule based. Talk to all of it. Great. What if you have 10 million customer base though. Is that enough? Because if you look at the second bullet point, what segments do is that it pens a customer or individual into one segment at a time.
Now some people like that mutually exclusive nature of the segment. That you know for a clear cut plan and don’t want any overlaps I would like to know how many people were dealing with. But the danger of pinning a customer into one segment is many. In other words, when you talk about that the descriptions of segment and you know there could be five million people in the segment and the way we describe these things is that yeah, the segment has a higher than average index for luxury car, for example. You end up calling the entire segment into luxury car buyers and that’s the dirty secret behind all this.
To simplify very complex interactions of numbers and figures, we describe these things in a very simple to understand term. I use the example like Golden Age is everybody in that segment rich or old. Maybe they have to live in an expensive area but they maybe have an income problem maybe they retired or lost a job. So we don’t know that. You end up treating everybody in the segment the same way because you’re pinning a target into one thing. Why is that useful? Because there are a number of segments that’s limited and when you run what we call Kamins clustering you have to know and request ideally how many questions you want anyway. Are we talking about dividing the entire country into 70 different pieces or are you just needing virgins of messaging and maybe I just need top 5 so I just need less than 10 segments. For those things you’re not going to create unlimited number of.
And like I said, this is really good for messaging. You are not going to create an unlimited number of versions copy so this is a good way to go about using the segments. Well it’s not easy to recreate how all these groups got together in the first place. As the world changes there’s no guarantee that you’re going to recreate the same set of segments again and again and again what they’re doing is to create a rule and they milk it for a long time sometimes more than 10 years and keep updating the segment itself as new data sets show up.
It’s really not re-development now. I’m not saying that it is always bad because for the planning and messaging purposes if you really set up something that you have a set of content that works for say old rich people. People who come in and out of the segment who are different but what you want to say to these people do not change so there is honor in keeping that consistency. So if the the consistency is more important. We’ve got to stick to the segmentation. But is that a good targeting tool? Well that was talked about person in that case.
The fourth bullet point I talked about already, so let me be brief here. Group the person and describe them later. That’s exactly what happens. You throw all the data into a big blank wall, see who sticks together, and then draw the line depending on how many segments you want and they end up describing their behavior using all the dominant characteristics of them not just individuals in the segment but the entire segment.
For example, yes this segment has higher than average index value for luxury goods or higher than average index for income whatever it is or homeownership rates this list goes on. Then you end up treating the entire segment the same. This leads to and again let’s go back to the first page. What is one-to-one marketing? It’s about knowing who to talk to. You can do that with a segment. That this segment sounds like a good money generator and a lot of rich customers or valuable customers of ours are in that segment. Let’s go after them.
But if you want to really pinpoint the behavior of who’s likely to respond to a certain offer, who’s likely to say buy a certain type of product or fall for a certain offer. While that’s something that you want to build an individual model for because the number one reason, not everybody is just one attribute. In other words, when you use a segment you pretend that everybody in the segment is the same way but if you look at me I’m made up a lot of different crazy behaviors. I’m the cheapest person when I’m buying things online, price shopping and bargain seeker, but funny enough not when I buy things offline, it is a very emotional purchase. Am I an online or offline guy? Hey I could get both it doesn’t really mean anything. So therefore you have to really understand where you go about doing these things and in that case you have to really go after one attribute at a time.
The nice thing about personalization is you can score high in multiple categories because each person is essentially separate in the model anyway. Therefore it’s easy to update you don’t have to update a whole set of models just because something is not working you just pinpoint the model that’s not working anymore and build one model at a time or update one model at a time. And these things are absolutely ready for multichannel marketing because, again not everybody is one dimensional. So, if you’re going after certain things and you have to think about what you want to do now. Are you pushing certain product at a bargain price or conversely you’re looking for full price purchaser’s not everybody buys things at a full price. That means you’re talking about a propensity to pay the full price. Describe pitiful price once, I’m going to treat him accordingly. Doesn’t work that way.
Now, when you do this, you can be a luxury buyer and a bargain seeker at the same time. So I’m going to use more examples later on but there’s a fundamental difference between segmentation and personas. I don’t think we should use these words interchangeably and I’m not saying this as a techie by the way. Even in a planning session we should really treat these two things as two different techniques and two different things for different purposes.
Let’s go back to our original title about Rule Based Segmentation. Like I said, if you are using it and you have a successful segmentation technique already, I’m not saying that you should ditch it today. I’m not saying that at all. Something has proven to be working by any means, keep it but the question that I have is, “is that the best we can do?” There are many reasons why, I just summarize it through about three major characters.
One is variable selection. When you build a rule we tend to think about in terms of the variables that we are accustomed to like RFM variables. Where if you engage the non transaction base. In other words you’re prospecting so you don’t have any RFM variables that you rely on income and age. Now if you called any respected data vendors, you know how many demographic variables that they sell minimum three to 400 variables that is just on a demographic side. That’s not even counting lifestyle variables, census variables, your demographic variables, where they are on model targets. You’re talking about hundreds of variables here. If you just use two colors thats like you have a box of crayons and 80 colors. You keep using two colors. That’s not the best usage of the data and also how do you know there are certain things that you are trying to do again, let’s be objective oriented here.
What are you trying to do? Are you trying to find the best value? Are you trying to find the best responsive guys or high value customers or people who are about to leave you? What are you trying to do? Now how would you know that income is the best predictor for that differentiation? The joke that is this, do you really believe that the difference between a Lexus buyer and a Mercedes buyer is income? Not millionaires. They’re all rich so therefore the variables that we instinctively choose may not be the right selection.
That is the first problem. Now if somebody did stumble into something that works fine. Income doesn’t work so I added more selectors like say a presence of swimming pool or size of the house, or presence of the fireplace or all that. Fine. Great, but you stumbled into it. The variables didn’t present themselves. When we use mathematical techniques. It will come out for you.
Second issue. Just because you picked a bunch of variables at work does not mean that those variables are created equal. Some variables carry a lot more weight than others. Not to mention the fact that these variables interact with each other. In some cases some are negatively related, for example, let’s say you want to go after the high value customers and then you go well high-value customers are always multi-buyers. They should come back to the store a lot. But if you look at the average value of a customer sometimes you find that that those frequency variables or a dollar per transaction of value per customer, are inversely related.
That means that these whales are not so frequent visitors but when they show up they spend a lot. I’m one of those guys. I hate shopping but the point is you don’t know that you cannot just imagine interactions between variables like that and more times when you’re wrong. I have worked with very smart people for many years and even the smartest individual you cannot imagine an interaction between more than two variables in their head. There is no way. Statistical techniques look at hundreds of variables and pick 20 and they know what’s more important than others. That’s what it does.
Third problem. Great you want to go after high income families, fantastic. Let me ask you that question. How high is high enough for what you’re trying to do? And by the way a lot of times the middle group could be most responsible. Very rich individuals can not open email, because they are bombarded all the time. The ideal range for me could be between say 80,000 a year and under 150,000 dollars. Now how do I know that?
I don’t know that I would have to rely on some techniques and talk to some real statistical analysts either internally or externally to find out how high is high enough. A lot of people put down numbers like our income is over a hundred thousand dollars. Well that’s what you imagine that’s human limitation. Why not eighty eight thousand dollars? Why not $105,000? Those numbers are not intuitive so people don’t use it.
And when it comes to categorical variables, oh my god it’s even harder. You know how many. I’m not sure how many of you guys are B2B marketers. Do you know you have more than 100,000 variations of the SIC industry code, and industry targeting is the most important targeting in B2B prospecting? Well you’re not going to imagine the combinations of all this SIC codes in your head so this becomes really really difficult. This is why rely on the gut feeling only goes so far.
If it works, fine. If you have a very small base, that’s fine too. You don’t have to reinvent the wheel if it’s working. All I’m saying is you should consider using modern techniques. I’m not going to go down the whole list. I wrote a lot of articles about this so if you wanted to dig deeper, by the way I write monthly for target marketing magazine so if you search my name in target marketing you see a lot of modeling related articles. Otherwise, all these best practices in marketing and database marketing anyway.
Bottom line is this, finding all the hidden patterns, finding the right brand forget branding and finding a right set of variables. All these things are much much better done by using machines. You know we talked about machine learning and all that by the way even if use the machine it goes through the same stuff they don’t use every single variable in the database. They find the interaction, they select the variable first and the final algorithm may not look like what humans can understand. But the point is that’s what they do. They could do it right because they have a lot of time and energy and increase accuracy. They have a lot of benefits by doing this. The only downside I think that has gotten in the exchange is that it is too small of a universe. You may not want to do this. I said that already. Predictive data not available. I hear this a lot and that makes me cry actually because this is the age of data.
Not having enough data means you’re not you’re failing at a data collection stage. Remember we talked about collection and all that deployment and delivery. You’re failing at the collection at this point. So we’re going to have a very different conversation about that but, not enough data? I don’t think so. 1-to-1 channel not in plan? Well that was the opening line. Every channel is a 1-to-1 channel now. Even the glasses that you wear on your on your face will be a 1-to-1 channel in the future. I mean Google failed but that does not mean that somebody is not going to do a better job later? On the phone they are looking at that is a 1-to-1 channel. High budget that’s a legitimate way to see things. Yes of course you have to have an ROI you can’t build an expensive model for nothing.
You’ve got to consider these things and there are a lot of ways to simplify things as well. There are a lot of automated modeling out there so you should really consider this. Lack of resources is another one, think ooh wow that’s great but who do I call? There are people who can help you there. In other words, not everything should be expensive. And I joke around like you, all the CRM market by the way like there’s only Rolls-Royce and nothing else, that is changing. We have a lot of cheaper options and automated options out there.
So how do we get to the answer? That’s another benefit of modeling, which is okay you have a thousand variables. Okay fine. You don’t want to use three variables again and again. The best benefit in the age of big data is this, it packs a lot of information into a simple to understand answer to questions. For example, the raw data on the left hand side if you add up all the number of variables. It’s going to be more than, or close to, thousands demographic data alone can be a few hundred transactions that are some data.
If you formulate all this actionable variables out of it, the last transaction date, the first transaction date, average value of a transaction, the retail value of whatever, categorical spending, or channel spending. the most popular banding. The price ranges for that person. I mean the list goes on and on. I reread routinely use like 600/700 variables. That’s a consideration by the way.
I am not saying that we should mashup all these things and I’m not gonna go down the whole list but this this is a typical list of all the variables if you’re doing things right. Ok so let’s not give up. How do we maximize the value of all this data? I would say be selfish and come up with the questions first, what do you want to try to do here? I mean what are you trying to do? Are you selling a very exotic luxury foreign vacation? Well let’s ask the question that way then. Why you want it is a marketing decision. How you want to push your product is your decision. We’re here to answer your question in a very simple answer to use and these answers likely to take a foreign vacation.
What it looks like at the end of all of this, a number between 1 and 10. One brings the worst and 10 being the best or it could be zero. I doesn’t matter what it is. In other words, all you have to remember is higher the better then you can plan a lot of things. Now this is the job of a data scientist or a statistical analyst or any kind of analytics job.
All we’re trying to do is to get out of the mode where you automatically go to the three or four popular variables and start using absolutely positively everything and get to these answers fast that when you make a decision you don’t datamine like thousands variables at the time of the decision making. You want to know who is more likely to be a high value customer. You look at one score where nine is good and zero is bad. By the way, sometimes the modelers may be able to give you the actual projection of the dollar spending for that person. That model could be based on 20 variables selected out of a thousand variables, but who cares. We don’t really have to know these things. You have to ask the right question and just know that this is entirely possible thing to do in todays game.
So let’s talk about personalization for a second because this is again what is personalization is really the second step of one-to-one marketing deciding who to talk to for the high response model fine target those guys. But what are you going to say? Now we have to think about what he’s the guy about is by luxury, is he about convenience. For example let’s say that you’re selling a laser printer. Knowing that if somebody is a photo enthusiast or has a big family, or has a home office, or just prints a lot. Or maybe a purchasing manager for some small outfit. I don’t know but you want to know if you don’t know then you want to know the possibility of him being a home office guy. Yes let’s have a model for that and all the things that I’m saying other personas.
If there’s a dominant characteristic for that person, and by the way some people score high in multiple categories. He has a big family and he just prints a lot. Point is, yes you decided you want to talk to him anyway. What are you going to say? Are you going to show the same old laser printer all over the place. Or show a different picture that will resonate based on his persona.
First statement: Personalization is about the person not about channel. It’s not about product. It’s not about your division for sure. It’s not even about the brand. It’s not even about the company. You should think, “ok I am trying to resonate with this particular target so that he thinks that out of all the 100 emails that I got today this email really talks to him”. I don’t know why.
Well probably because it is all the things that he cares for. How do we know? We don’t know for sure, but he’s very likely to be that guy. That is a very different approach than say I am pushing this product through this channel. So that’s a channel and product-centric mindset and that’s how you end up over-promoting to a lot of people. So let’s just put it out there personalization is about the person. I wrote a lot of articles about it if you read more about it please type in my name and personalization and you will see tons of articles about it.
If you change your mindset about it then there are a lot of other things that can help you. And let’s have an open mind about how you’re going to employ all this technology. What about the 90 percent of your base who didn’t raise a hand and say “I like him”? Well you know what other things that they like and this is where the modeling comes in. Modeling not only consolidates lots of information, it’s also easy to use. Score one score and two. It has no missing value.
You’re just less likely, or I’m not sure it’s not a high school year guy. But in the age of gold the model of OK she’s a science & a movie fan. Do I know that for sure? No. Can I say that? Yeah. 70 percent sure that he’s that guy. That’s all we going into now. Important thing is that there’s no curb going down and I don’t have any score for that either. There’s no such thing in a world of modeling. They’re just less likely to do certain things but by doing so you’re expanding the universe that you can talk to and the menu that you can offer that because I don’t know for sure if it is an action movie guy or drama guy or red comedy guy. I don’t know that for sure let’s say that you’ve built three models just for that.
Next time you promote something related to movies or books, you know what the dominating characteristic is if you just rely on real data or collect real data. You wait until someone raises their hand and tell you that I’m a romantic comedy guy. Well then this is true to you that you have a very small number of people who have that work and the rest of them are empty this I’m going to skip because we’re running out of time here but I wrote a article called elements of successful personalization.
Please read it. You have to be good at all the things and today we’re going to talk about data and analytics and try to expand the horizon for you so they can have multiple options not just some segmentation techniques you have to nail the technology so they show different things to different people. Because what good is a segment if only show one picture to everybody anyway? And also you have to have a good content library again what good is a bunch of segments and personas if you have only one picture of a product? But this page is an important page. You go like “Aw how do I start?” and this actually the page where I say segments is a very good thing to start with.
What is one-to-one marketing? Knowing the target and knowing what to say. So therefore, when you want to deploy any kind of one-to-one marketing, it could be web or email. Let’s just say this obtains some capability to be able to show different things to different people. That’s number one. I’m not that guy by the way, I’m a data analytics guy. But the point is without the technology you cannot do anything. That’s the first.
The second one, you’re going to test drive this engine now and you’re not going to build like 20 personas from day one. I’m not even going to recommend that anyway. This is where, I don’t care if it’s a simple rule based segment. Let’s say that you were selling cosmetics and you know the target age groups and gender and probably some regions where your brand is popular. Build that rule, don’t hire a statistician from day one.
That’s phase two and make sure that’s based on the rules that you just set up. You can show different things to different people and then let me go back to the previous page so that you know the technology, content library, data, and analytics can work together. Not all of these things will be perfect from day one but you start by testing phase 1 and 2. I’m going to go into more statistical space. I have more variables now so I’m going to use a have a full use of an entire database. I’m going to build some personas and propensity models that’s phase 3.
Today my goal is to show you the lay of the land and say that with your commitment all the technology that enables all these things out there. The biggest hurdle in all this ineffective marketing sits in marketers habit that you’ve been using this segment for a long time. You inherit those segments from somebody else in your company in the past. I don’t know how it happens. I have seen a lot of cases like that. All we’re doing here today is to expand the horizons so that you know the realm of possibility given the fact that yes this is the age of abundance of data, the technology is cheap. Just go out and buy this software. If you are any kind of a by the way e-tail conference or any kind of a DMA conference or can you imagine. I mean you’ve been DMA conferences, this is a DMA production anyway. Hundreds of vendors will offer this. All I’m trying to say today is to know that this is all possible. It’s a matter of putting those elements together on a previous page. And started this way and modeling and segmentation is a big part of this whole conversation.
Now to do this though you need a single customer. You tend to think that all these people. Yeah he’s an online guy or he is a call guy. He’s a store guy because he showed up in a POS database. This is a very channel-centric mindset and what’s the first thing that I said when I start talking about personalization? It’s about the person. It’s not about division. It’s not about your department not about your brand. It’s not about your channel.
Certainly not anybody is an online person by the way. I’m looking for a golf club or somebody generated my interest in golf or by sending me an email. I don’t need a golf club but I got an email. I’m curious what’s the first thing that idea I searched for it on google to find out Titleist has a new driver, is that a good driver. Then can I buy online right there. Yeah I could do that. I want to hit some balls I go to the store. Let me ask you this. Am I an online person or offline person? You get to put it all together so that you have a clear view of a 360 customer view.
This is another term other than big data and all this there is like us a 360 customer view, personalization, those are popular terms. But what it is that you need to do to get to that point and this is a good chart when you deal with all these I.T. vendors and technology people you should differentiate. You should check all the boxes here and create analytics friendly environment and make sure that all those things that are listed here are checked off.
If you dumped the raw data to modelers looking at all the green part of the right hand side. It will take a sick amount of time and this is when you realize, “oh my god he told us that you need a model and all that but it’s so hard and it’s not working and why is that?” It’s because your data is not model-ready but if you know what to do here on the left hand side of the data categorization because of vision and my vision so you know and have information on individual level instead of just raw transaction data you know somebodies not just anybody. Someone specific, his last transaction date. His average value, his product composition, pretty much all these different categories, channels, dollar bands what did he respond to, is it free shipping guy? Was he a buy one get one free guy? All these things are everything that I just said here by the way is about the person not about the transaction. How do you go about doing this?
I know that is a little daunting. I know that it’s a bit too much of a technology type of a thing but let’s just keep this in mind that you need to think about the data strategy not for the convenience of the maintenance and just storage you’ve got to create a central place, which is customer-centric, not brand or division-centric, customer-centric so that all the customer information is one click. And what is a customer really just look at it from a different way behavior means that somebody does something, you bought, you browse, you visit. Whatever you did the right side of the channel you did something. I sent them a catalog. I sent my email. Did he respond to it? If you summarize all that in a centralized way.
I’m going to use one example today. If you look at the right hand side the tuna can looking grey cylinder there. I have a wording like adequately promoted, overpromoted, and under promoted. That does not come from the fact that I know that I sent 40 emails to that guy last year. Now if I say 40 you think that that’s a lot I know retailer who sent ever sent 40 e-mails per person per week. 40 is not a lot but for some other people 40 is a lot. That’s all relative. If you responded five times out of those 40 maybe you should send more. It’s all relative.
But if you cannot even put those things together in one place the email vendor has a email history. Your CRM database is responsive, all I’m saying is you’ve got to put some effort, make some commitment to do these things and I’m not saying you should do this like before you do anything. All I’m saying is your segment your modeling and the speed in which you deploy the things improved by 20-fold improvement by the way in time things that use take like two months to build one model you can build one model in one afternoon.
Now I’m not saying they should need all these things. What I mean is that this is one of those things that you have to check off to get data model ready so that even if you build a rule based segmentation, you do this right you’ll realize it “oh my god it’s a rule based segment”. I know but the call is much much more colorful. I can use a lot more than just income age and number of purchases. So that’s what you’ll realize. So therefore I mean it’s a short session. I’m going to go deep into it but I just want to share that with you.
The rest of the presentation is pretty much the summary of what presonas look like. This is your reference I brought some sample personas you can imagine. And I put them samples because this is when you become selfish. That’s what I mean by selfish all this time I’ve been talking about customer-centric marketing and customer-centric data for about an hour. The point is this is when you decide what personas do I need it to sell what I’m trying to sell so that depends on what industry you’re in.
I’m going to leave this deck behind anyway and you can always contact me to brainstorm by the way. That’s what I do anyway and I brought two pages full of examples. Depending on where you try to sell. Let’s not imagine in terms of income, age ,and where they live and all the things that we do all the time. Let’s forget all that. Let’s just define we want to go after you. Is it about fashion, about scale, about skin, the frequency of the purchase? Is it about what she is going to respond better in terms of offer free shipping versus 10 percent off or buy one get one free? What is it about? This is entirely up to marketers. All we need is a sample individual who showed some behavior like that as a sample. That’s all we need to build models after that so of course I’m not going to talk about how to build a model in this session but it’s it’s not as difficult as you think. But knowing what to build is the hard part and so this is an example of how you start doing these things.
Key takeaway. Modeling one-to-one marketing is about targeting and personalization so let’s not forget that and everything should be based on buyer-centric view. It’s about the reader it’s not about you, certainly not about technology.
Yes, of course you need all the things that you have to be worked in conjunction with all these different things. Second bullet point. But again the governing rule is still buyer-centric. Rule basic segmentation has significant limitations and I think we covered enough. If you forget, remember those three things were able to choose do you know what’s more important and also what banding is. I mean how high is high? How low is low? You want to go for frequent cruiser on a cruise ship how high is high enough? More than three times in past five years or just once a year? I mean you can make up any rules but let’s not think like that anymore.
That means you’ve got to invest in analytics. You should at least start talking to a real analyst in or outside your company before you formulate all the questions. In fact formulating the questions is part of their job anyway. Any good console or consultants will start formulating the questions before you give a single answer and the database must be optimized for analytics and modeling and this is exactly the job of a data scientist these days. But the point is they also complain all about dirty data and what I say to them “hey that’s part of your job. Do you think this is school? You’re not going to get a perfect data, deal with it.” Point is, you don’t have to deal with it but you have to commit to pay for certain things so that it’s data ready. I mean analytics ready. All I’m saying is expand the horizon please.
It’s not just about, I’m going to personalize when I get to have a like hardcore data that this guy like something. Let’s go one step further and have a goal like personalized all the time for every once every touchpoint. Tall goal, I understand but it starts with like 1 email channel that’s a good start actually or a banner. Whatever your favorite one to one channel. Start somewhere please and we do this. Think about all that data, content, technology, and please don’t forget the analytics.
One parting line: nobody is one dimensional so let’s not just automatically fall back to segments. You’ve got to have a different purpose for personas and segments. You need both by the way. But let’s not just go back to the good ole 3 variable based segments by habit. So that’s the conclusion of my session today. So thanks for listening to me for 50 minutes. So now we will take some questions.
Q: Can this be applied in the nonprofit world for things like likely to be interested in local politics or likely to be interested in environmental causes or pet owner can use this to predict who would like an environmental message versus one about voting rights?
A: Not only can it be applied that way I have done it myself actually in fact. Funny story. You know everybody in politics they want the middleground people because people on the far right or far left they made up their mind already we all want to go after the middle ground. Yes you can go model.
In fact, let’s just say that past few elections were determined by model. And I’m not saying this lightly by the way some presidential campaigns were done. I’m talking of the elections done eight years ago. These guys knew exactly what number to put in as a headline for each email. Is that a ten dollar guy or a five dollar guy? What is the headline? This is about environment that there about women’s right, civil right, is about taxes, about gun rights? It could be anything it could be. You know what I’m saying is this. Absolutely positively yes.
If you want to brainstorm about how to go about this thing we can always talk about it. Another thing, donation is not that different from say purchase. You’re committing your dollars to a certain cost. It’s not that different from say all I am buying this high end Samsung, great or your high end electronics. Technology wise is that any difference from predicting who is going to be the more than 200 dollar donor for certain election no. Environment absolutely. Green is another thing. In fact, when I was in a big data provider company we built a lot of models that we can simplify all these transaction. The green models. And by the way it was not easy because if you just rely on survey they will say oh yeah sure why not I am green. Have you ever done anything green and actually spent money being green? That is a very different question. In other words did you change your light bulbs to change some?
I don’t know. Did you install solar panels? Did you donate to a green cause? Whatever it is. All you need is hard evidence of somebody doing something like that. You don’t have it and you can also rely on that survey. Now it’s not my first choice but you can use that. The beauty of all this if you can project models and behaviors that way.
Q: What are good examples of platforms for retail that do a great job at building and implementing personas.
A: Unfortunately, there are not many customers that have dealt with who do really, really well but let’s just say that some of the I cannot say the name because I’ve seen this in confidence some of the home how do you call those not home furnishings but Home Improvement kind of retail stores. I’ve seen retailers who built up a hundred and twenty personas just in the home improvement category. So when they sent out emails. Is it about small jobs big jobs? Is it more likely to be a contractor or is it about lumber, paint, outside, electric, whatever it is the oldest propensity to do certain things they do not wait until somebody does something they build models with a bunch of people who actually have shown such behavior. Build a model and wait and when you deploy email campaigns or whatever that you send out they show the relevant content.
So it could be the same kind of a campaign to promote traffic to their stores. But the picture is that people see it will be different. That’s one example. There are some guys in the travel industry who try to do this but they’re still relying on good old segmentation techniques. I’m not saying that that’s bad. By the way though the first step is to really be able to show different things to different people. And also hotels are very good at acting on the data that they collected. But yet this guy liked certain things and I’m going to do that you care for a point where he careful logic. He cares for no easy reservations and all those tendencies are hard coded.
The next step that showed that some people are doing is to predict who’s likely to be a luxury traveler who’s likely to be a heavy hardcore business traveler. That’s what they’re trying to do with a model. So the limitation is almost none. We’re talking about everything not just retail, travel, finance, banking. You talked about nonprofit. All those things can be the same effect only major difference is for B2B it’s a little bit different. Not because of the fundamentals of the segmentation and molding are different is the same but the nature of the sales process is different because you talk about multiple steps somebody being a qualified lead and all that that’s different. And the channel uses is different. And most importantly the nature of the data that you collect on a B2B space is different. So that’s the only difference. But the fundamentals of what we talk about is really applies everywhere. Thank you.