Roni Cohen and Sivan Desser discuss methods of defining your most valuable customers and the advantages of identifying this customer segment.
– [Sivan] Thank you for joining the session. In this session, we’re going to talk about the VIP segmentation and we’re going to discuss how to better define your most valuable customers. This is Roni over here. She leads the data science team here in Optimove. I’m Sivan.
I’m a data scientist. And let’s get started. Every business should know who their VIP customers are. Now, there are a few reasons why this sentence is so important. So, the first reason, the very obvious one, actually, is that these are our most valuable customers. They worth most to us and therefore we would like to clearly identify them, we want to make them happy and keep them satisfied in order for them to stay with us and not churn.
But not only that, there’s also a lot of information we can gather and learn from these customers. For example, we can maybe try to identify and look for specific behaviors or characteristics that they have and when we’re comparing them to the rest of our customers’ database, you can get a better understanding of what kind of behavior we would like to encourage and see more of from the rest of our customers’ database in order for them to be in the right path to become more loyal and more valuable to us.
So, we understand that it’s very important to define this VIP segment, but we know that this could also be quite a challenge. On the one hand, we can maybe underestimate some of our customers. We can create a very small segment of VIPs, and by that, we might be overlooking some very valuable and very important customers and we’ll not be interacting with them in the appropriate way.
But on the other hand, we can maybe do an overestimation. We can create a very large segment of the VIPs. And by that, we will be standing very generous offers to customers that are not necessarily worth these offers. So, we understand that this could be quite a challenge and this challenge is exactly what we’re going to address in today’s session. And we’re going to do that by answering the following five questions here in front of you.
So, we’re going to start off with under- standing what attributes should you look at when defining a customer as a VIP? Then we’re going to talk about the ideal segment size. We’re going to talk about thresholds. Should they be fixed? Should they be dynamic? How often do we recalculate and actually updating the list of customers that we have under our VIP segment? And finally, if we have a big enough of a VIP segment, we might want to consider splitting it into sub-segments.
But before digging in those questions, we first need to understand if a VIP segment is even needed. And to answer this question, we created here in front of you a Pareto analysis. Now, I’m sure most of you are familiar with Pareto analysis, but I just want to quickly explain what we did over here. So, here we took customers and we sorted them from highest to lowest in terms of their monthly income.
Then we grouped them together into percentiles, and for each percent of customers, we wanted to see how much they contribute to the percent of monthly income. So, the percent of monthly income is the blue bars and the line is the cumulative monthly income. Now, as you can see over here, the first percentile, the top 1% of customers contribute 9% to their monthly income.
The second percentile contributes 6% to their monthly income and so forth. And what we can see over here is that we don’t have any specific group that really stands out and that is very different from the rest of the groups, and that’s why in this specific case, a VIP segment might not necessarily be needed.
But I can tell you from my experience that those are not the cases that we’re used to see and normally we’ll see a group that really stands out and that is very strong and different from the rest of our customers. And that’s why they will be our VIP segment. So, let’s say that. Let’s say that we do want to create a VIP segment. The first question that we want to ask ourselves is, what attributes are we going to look at when deciding if a customer should be a VIP or not?
Now, when thinking about this question, there are two main points to take into consideration. So, the first point is the metric itself. We have monetary metrics, we have frequency metrics, we have also sort of metrics that define the volume of a customer’s activity. And the second point is the time frame. So, obviously, the default time frame to use is the lifetime values.
Right? Because they represent the whole journey of a specific customer. But when thinking about segmenting a customer as a VIP, we might want to do it based on his most current behavior. So, we might want to examine also show the time frames like last month, last year, and so forth. So, there are different points here to take into consideration when thinking about what attributes are we going to base this decision on.
And this is exactly what we did over here. So, here in front of you, you see a Pearson correlation matrix. Now, in this matrix, each cell represents the correlation between two different attributes. So, over here, we have a list of attributes that we would like to examine. They’re based in all the two points that we’ve discussed before.
So, we have frequency metrics, we have monetary metrics, we’re looking at different time frames here. We have last month’s attributes and lifetime attributes. And let’s take a look at specific example. So, this cell represents the correlation between the number of visits in the last month and the number of items in the last month.
And we can see that the correlation here is 0.6. Now, the numbers here can go between zero to one. Zero means no correlation, one means full correlation. As you can see, for example, I hope you can see, the colors in the gray diagonal over here, we can see the correlation for one attribute with itself. So, this is, of course, a full correlation because it’s basically the same attribute.
If you don’t see the numbers, I can tell you, you can see it from the colors, right? The brighter the color, the smaller the correlation. Okay? So, you can get from the colors. Now, when thinking about VIP customers, we always want to look at the correlations between the different attributes to the total value when we’re looking forward.
Total value is the right column highlighted in blue. Now, the reason we are doing that is because we want to find the attributes that will enable us to understand who our most valuable customers are. So, we always want to look for the highest correlations with the total value. And as we can see over here, we have two attributes with very high correlations.
So, the attributes are the purchase amount in the last month and the purchase amount lifetime. Now, before choosing to take both of these attributes, we first want to say that it makes sense for us to choose both of them. So, we’ll do that by looking at the correlation between them. So, this cell represents the correlation between the purchase amount in the last month and the purchase amount lifetime.
And we can see that the correlation here is very high, meaning that there is no reason to take both of these attributes and it will actually be redundant. So, to take one of them will be enough. And at this point, we will choose to take the purchase amount in the last month because it has the highest correlation with the total value. Next one, we’re looking at the total value column. We can see that we don’t have any more high correlations, and that’s why in this case, we will stop and we will choose the purchase amount in the last month to be the attribute that, based on it, we’re going to decide if a customer should be a VIP or not.
Now, once we’ve chosen our attributes that we’re going to base this decision on, next, we want to understand what should be the VIP segment size. Now, here in front of you, again, you see Pareto analysis. This is very similar to the example shown before except that here, first of all, we based this Pareto on the monthly purchase amount, the attributes that we have chosen.
And second, we can see over here that it’s very clear that a VIP segment is needed, right? So, a Pareto analysis not only helps us understand if a VIP segment is needed or not, it can also help us understand what should be the VIP segment size. So… And we’re doing that by actually looking for percentiles that contribute to the monthly purchase amount in a very disproportionate way.
So, basically, we’re looking for big drops in the bars that you see over here. So, as you can see over here on the left, the first percentile contributes 25%to the monthly purchase amount. So, this is very disproportionate, right? They’re are only 1% of the customers, but they contribute to 25%, almost a fourth of the total monthly purchase amount.
So, they are very strong customers and they most definitely should be in our VIP segment. Now, when looking forward, we might want to consider taking also the second percentile, maybe even the third percentile, to be in our VIP segment because we see that they’re also different from the rest of our customers’ database, but they’re not as strong as the first percentile.
So, if we were to do it, we might want to consider splitting our VIP segment into sub-segments, and by that we will interact with each group in the appropriate way. But for this example, it will be enough for us to take the first percentile and we will say that our VIP segment will be our top 1% customers. Roni. –
[Roni] So, now that we understand what is the attribute we will be using to define who is a VIP customer and also what is the segment size you will have when understanding who is a VIP customer, the next question we’re going to ask is, “Should the threshold be a fixed or dynamic one?” Being the threshold that a customer has to reach for us to decide, “Is he going to be a VIP? Yes or no?Are we going to basically set it in stone at this stage, or is it going to be dynamic?”
And I think the graph over here will help to better illustrate the exact question that we’re asking. So, what we see in this graph over here is based on the attribute that we’ve chosen, the monthly purchase amount, and also the segment size, the top 1%, we study one year back and we look every single month what was the threshold the customer had to pass in order to be in the top one percentile?
What was the minimum amount a customer had to spend in the past month to be in the top 1%? So, if you look at the first bar over here, June 2017, you see that in that month, for a customer to be in the top 1%, they had to spend at least $926. But if we move forward a few months and we look at October 2017 over here, you see that in this month for a customer to be in the top 1%, they had to spend at least $1,318.
So, much more than we saw over here and we see how this trend can gradually change. If we study this trend one year back, we see that in the past year, on average for a customer to be in the top 1%, excuse me, they had to spend at least $1,000. So, when we ask the question of, “Should this threshold be a fixed or dynamic one?” we’re asking, “Should we take that $1,000?” and that will be the criteria from today moving ahead to define if a customer is going to be a VIP or if at every single point in time that I want to understand who are my VIP customers, do I look to see what is the threshold at that stage, and then that will be a threshold that will continuously change.
There isn’t necessarily a right or wrong threshold to use here. There are pros and cons to each one of them. I’ll start off with the fixed threshold. Main advantage over here, it’s easy for us to understand. We have a clear understanding of what it means to be a VIP customer. Also much easier for us to measure and track VIP customers in different points in time, if the criteria to be a VIP customer does not change.
However, we’re losing the relationship between that top-tier group and the full customer database. Because when we’re looking at that top 1% we’re basically able to extract that very unique group of customers that is much more valuable than the full customer database. When we don’t continuously adapt our threshold, we might lose the distinction between these two groups.
So, we can answer that by using a dynamic threshold. When we use a dynamic threshold, basically, at every single point in time, they want to know who is a VIP customer, we look at what it takes to be in the top 1% in that day, and by that we’re able to promise that we’re going to continuously adapt to the changes in our customer behavior. The flip side of that can be that it might be a bit more challenging for us to measure and track the behavior of this VIP segment. In terms of deciding what is the right threshold, there isn’t necessarily a right one.
What I would actually recommend is build a graph like we’ve seen in the previous slide and understand how your threshold behaves. If you have a threshold that’s fairly volatile, meaning that the business is very much affected by some sort of seasonality, in those cases, I would definitely recommend to look at a dynamic threshold. But if we see that we have a fairly steady threshold, in those cases, having a more simpler fixed threshold might be the case to go. The next question we’re going to ask is, how often do we recalculate?
Meaning, how frequently, every how many days do you want to refresh the list of our VIP customers? So, what we see in this graph over here is based on the criteria we’ve chosen up until this point. We look at a specific month, and every single day of that month, we asked, “Who is a VIP customer? Who can be found in my VIP group in that day?”
And every single day, we want to understand what percent of customers have recently joined that VIP segment. That’s what we see over here in the blue graph, and what percent of customers have left the VIP segment? So, if I compare my VIP customers today and yesterday, what percent of customers have left and can’t be found in the group anymore. That’s what we see over here in the in the purple line. So, you can’t really see the numbers over here, but basically, in terms of percent, we’re seeing that around 0% to 7% of our customers either leave or join the VIP group on a daily basis.
So, nothing major, but we understand that this group isn’t 100% stable. There are some changes. So, it does make sense for us to ask the question every once in a while, “Should we refresh the list?” So, we can either decide to do this on a daily basis. If you refresh the list on a daily basis, we’re promising that we always have the most accurate list. Every single day I know who are my VIP customers today.
Main advantage of that is that I’m going to be able to locate newly acquired customers that are showing VIP behavior from day one. So, from the first day that they’ve joined, I’m able to understand that they’re much more valuable, I’m going to interact with them in the appropriate way to treat them as a VIP customer and make sure I build that strong and loyal relationship from the beginning. Now we can also decide to do a periodic refreshes of this list.
So, do it maybe every week or every month. This will allow us to have one stable list to work with and then it’s easier for us to implement our marketing strategy because we know we have a closed group of customers that we can continuously follow. What we would actually recommend over here is not having to choose but actually do what we call a grace period. And basically, with the grace period, we’re making the best of both worlds.
The idea is that we’re going to recalculate lists on a daily basis. We’re going to close customers in the VIP segment for a minimum of one period. That will be their grace period. So, every single day, new customers can join this group, but they’re not going to leave immediately once they don’t reach the VIP criteria anymore. And then by that, on the one hand, once the new customer join and is very valuable, I pick up on that right away and treat him accordingly, but also, I’m not worried about giving my customers mixed messages because I know I have a certain amount of time to implement my marketing strategy and they’ll be in that closed group for a while.
So, now we want to talk about, “Should we further split our VIP group into sub-segments?” So, up until now, we’ve talked about defining that VIP group, but I might want to decide to even further segment that into sub-segments. If you remember at the beginning when Sivan showed the Pareto analysis, we saw that we have a very strong top 1%, but we also saw that we had a fairly strong second and third percentile.
These customers were much more viable than the full customer database, but they were definitely not like our top 1%. If you have the ability and the capacity to actually manage different marketing strategies for different tiers of VIPs, we might want to have a larger group of VIP customers, and then further segment them into different tiers and then personalize the right messaging, the right offering to each customer, based on the tier that they are in.
So, for example, we have a VIP group of our top 5%, we might want to further segment it into three tiers. Tier one will be our top 1%, tier two will be our second and third percentile, and tier three will be with the remainder of the customers. And then by that, I know all these customers are much more viable than the rest of my customer database, but even between them, some might be more valuable than others and I’ll treat each one based on the tier that they are in.
We can also take advantage of this and actually track the highest achieved tier of every customer. So, every customer regardless of the tier that he is in today, we will know what’s the highest achieved tier he was in sometime in the past. And then by that, we know what is the potential of every customer. Say, for example, we have somebody that is in tier number three today, but in the past was in tier number one, that’s very beneficial information for me to know about this person because I know he has the potential to actually be much more valuable.
I can then interact with them in a way to see what we can do to push them back up to the to the top tier and basically have them fulfill their full potential. I think we all understand we’re going to be treating our VIP customers, our top tier group in a very special way, and for us to see the best results from that strategy, we have to make sure that we’re accurately defining what does it mean to be a VIP customer.
So, today, we’ve covered five different questions of how we can do that. First, what are the relevant attributes for me to look at? Then, what is the segment size that I’ll be looking at? What is the group size of my VIP customers? The threshold itself, is it going to be fixed? is it going to be dynamic? How frequently do I want to refresh this list?
And finally, should I further segment this group into sub-segments?