Your segments could be sabotaged by any one of these sneaky mistakes, but luckily they’re all easily preventable — if you know what to look out for.
If you’ve been following our series on segmentation best practices, then you already know why behavioral segmentation is more useful than demographic segmentation, you know how to define a “good” segment, and you know how to enrich your segmentation with marketing experiments that enhance your existing data.
Even armed with that knowledge, there are big mistakes that crop up among even the most segmentation-savvy marketers, so today we’re going to run through the most common that we see.
Apples and Oranges
We’ve probably all heard the statistics joke about when Bill Gates walks into a bar — suddenly the average customer is a millionaire.
This is comparable to the first and most common mistake that we’re going to get into.
Let’s say you want to run a comparative analysis. You’ve got a segment of amazing customers, let’s call them your Stans, as in our sock example.
All you know about The Stans right now is that they order something new every week, i.e., 52 orders per year. So you run some analyses on your entire customer base, and your plan is to compare that analysis with an analysis of just The Stans.
The problem is that including The Stans in your initial analysis makes your entire customer base look a little more Stan-like and a little less average, much like the many false-average millionaires in the Bill Gates Dive Bar. If everyone who doesn’t order once per week (i.e., non-Stans) orders only once per year, the analysis won’t say that — because The Stans threw the average off.
In other words, if you want to find out what makes Group A unique, only compare them to non-members of Group A. If you dig into a pocket of users, make sure you remove those users from the comparison group. This can be a very simple thing to overlook, and, fortunately, it's a very simple thing to correct.
The second common mistake is something we call "has ever" segmentation. There are some use cases of "has ever" segmentation that are perfectly valid and appropriate, but there are others that can really trip you up.
Here's an example of using "has ever" segmentation.
You got some new banana socks in stock.We don’t want to email all of our customers about every single product in stock because we don’t want to risk fatiguing them, so what do we do? Who might you want to notify?
Here are three customers:
Here, we see two customers who have bought banana socks before and one customer who hasn’t. iI we’re thinking in a “has ever” frame of mind, we might see Customer #1 and Customer #3 and think we should send them that banana sock email.
But here’s a tricky thing about trying to understand higher-value customer groups, which, relative to Customer #2, Customers #1 and #3 are.
You can’t ask a question like, “How valuable are customers who have ever done Event X?”
Why? Higher-value customers, by definition, are almost always higher-frequency shoppers. Since they’ve made more orders, they have a higher likelihood of having done Event X or Event Y or Event Z.
Looking at the top row, you can see that Customer #1 bought seven items and only one of them was a pair of socks. Customer #3 bought five items and only one of them was a pair of socks. Customer #2 only bought two items, but no socks. The problem with this “has ever” segmentation — as you can see — is that it isn’t normalized.
Think of it this way: When a customer makes an order, they're reaching into a big bin of different potential items that they can buy. If they've had 10 turns or 20 turns reaching into that bin, the chance that one of those times they pulled out banana socks is going to be a lot higher than someone who only reached into the bin twice.
In other words, the higher a customer’s lifetime value (CLV), the higher the likelihood that they fit into any given “has ever” segment, but that doesn’t mean that they’re the best or even good customers to email on that basis alone.
Obversely, the CLV of anyone who has ever bought socks is going to be higher than the average CLV of the average customer just because high-value behavior will encompass more events than low-value behavior.
Which means that you could potentially dangerously conclude that people who buy banana socks are your best customers, when really, it’s just that your best customers happen to snag some socks in their long, eventful customer lifecycle.
Consider asking the same question about “Who bought pants?” or “Who bought shirts? ”
You can imagine that if you have many SKUs, the answer is always the same: high-value customers.
It is more likely that someone has bought any product if they've purchased many products, which is why you should be careful about assessing the value of a segment with a thing like “has ever.”
So how do you get around this? You normalize the inquiry by instead looking at first purchase to gauge product affinity. For example, how valuable are customers whose first purchase was a pair of socks versus those whose first purchase was shirts versus those whose first purchase was pants? That's normalized because it’s set in a fixed timeframe.
If it turns out that the people whose first purchase with your brand was banana socks are also the type who often return over and over and over again, now we're talking about an interesting segment that you should investigate further.
No Difference, Unnoticed
When you do any kind of segment experimentation, you want to set up a control group to get a baseline. Otherwise you won’t know what kind of effects your efforts are having.
For instance, let’s say your sock retailer runs an email campaign segmented by males and females. The female conversion rate is 10% and the male conversion rate is 5%. Now, don’t jump right into thinking that this email campaign was necessarily twice as effective with females than males.
With a control group (i.e., people you don’t email at all) set aside, you might see that women always convert at 10% and that men always convert at 5%.
So what you thought was some insightful victory turns out to have been ineffective, i.e., it made no difference whatsoever.
If you have that control group, you learn something, and it becomes a valuable exercise. If you don’t, you might end up deploying a ton of emails that don’t move the needle on driving customer value but do squander your marketing and creative teams’ time and resources.
If you're looking for ways to increase the relevance of what you're doing with the customer population, you want a baseline so make sure you test your segments and compare it to the baseline.
If you haven’t already checked them out, go back and read our previous posts on proper segmentation:
- How to Slice the Pie, Part 1: Demographic Versus Behavioral Segmentation
- How to Slice the Pie, Part 2: Design Data-Rich Customer Segments
- The 4 Layers of Effective Buyer Personas
And stay tuned for our next post on how segmentation can help retail marketers vanquish their biggest challenge: getting those first and second repeat purchases — or as we like to say, getting the second date.