Don’t be average! Follow this advice for success in business analytics
In the last article, What is a cohort and why should I care?, it became ultra clear why you should always approach reporting with cohorts in mind, and to mistrust every time series you see, because it doesn’t have the context that made up those numbers. It’s like a head chef who doesn’t know the recipe!
Similarly, if you were the head chef at a highly rated restaurant, the last thing you’d want to hear is that your dishes are AVERAGE. As a manager, you are the chef of your company, and when you think about performance metrics, the last thing you want is to be average!
First a little recap, before we learn how to avoid being average.
What does it mean to be “average”?
One way managers at companies fail is by looking at their data all combined up into a big soup. For example, look at this time series (which was also shown in the previous article):
This view into a company is less than worthless — it’s actually dangerous. We could jump to all kinds of conclusions without knowing anything about how the numbers came to be.
So, in order to get more context, the next step is always to split the time series into a cohort triangle, like this:
Now we’ve unlocked valuable information about “how” the time series came to be. But this is still not valuable, and could never be used for making decisions, because it’s too “average”.
Cohorts are a good start, but not good enough
What we have right now is the 2-D cohort view of every customer in the company rolled up into one big group, which is not helpful to a chef like you. It’s time to introduce customer segments.
Segmentation is key
The “overall” cohort triangle we built for Active Customers is interesting, and far more useful than the rolled-up time series of the same information. But you won’t use it to make decisions. We need to divide the data to separate out different groups of customers because they spend and purchase differently.
Covariate segmentation
The simplest version of segmentation is for variables we have in the data. We can partition different groups by grouping database columns (called “covariates” by statisticians). We often segment on product, region, marketing channel, demographics, or any other grouping of customers that our data makes available and we think may be a valuable view into the customer population.
A “segment” of customers is anybody with a specific variable assigned to them. If you’re segmenting on “region”, group customers in each region into a cohort. If you have a long tail of regions (more than 10 makes it harder to compare groups) then put the long tail into an “other” category.
- Create a Cohort for each segment of customers. E.g.
- Graph and compare the growth and retention for each segment. Are some segments growing faster or retaining better than others? Do you know why?
- If you have relevant marketing or finance metrics, it now becomes interesting to understand unit economics for each segment. Are you in the black for every segment? Are there some segments which aren’t profitable? Do you know why?
- On the LTV cohort, how long does it take to break even on CAC?
Here’s a hypothetical graph for comparing revenue for Subscriptions vs. Gift purchases.
This view takes you from “data” to “information”. You probably already know this information at a gut level, but now it’s clear and quantified. E.g. one product grows faster than another, or one region retains better than another. But now you have the numbers to back it up and you can make more informed decisions about where to allocate resources, where to focus marketing efforts, what needs optimized, and possibly which segments to abandon altogether so you can focus resources elsewhere.
Quintile Distribution
A quintile segmentation is splitting your customer base into 5 equally sized groups. Order your customers by spend and cut them into 5 groups. For each of those groups, create a cohort for Active Users, Revenue, and any other metric which may be useful.
You may see patterns that are interesting, especially distribution of spend (Pareto principle) and activity patterns.
RFM Segmentation
To take segmentation to another level, conduct an RFM segmentation and label each customer with a unique category. You will have a mutually exclusive, collectively exhaustive segmentation of your best/worst/other users, and you can use this segmentation to create cohorts.
Image credit to Putler
Note: If you did this segmentation again next week, some customers would move to different segments and your cohorts would be different.
The RFM segmentation will clearly show how different categories of customers develop over time. The Champions purchase early and often. The Lost customers have not purchased in a long time.
Now we go from “information” to “strategy”:
Graph each segment’s cohort performance against the Loyal Customers cohort performance. Find the inflection points where different groups begin to act differently and develop a thesis on how to treat them differently than the others. Perhaps an outreach effort in month 2 is appropriate? Maybe a coupon campaign in month 3 for the Potential Loyalists would be smart?
Statistical Regression for Health Score
Another data driven method of segmenting customers based on the data you have available is well described in this article by Ed Powers of Service Excellence Partners. He uses covariates in the data to do a multiple regression analysis and group customers based on expected revenue growth.
In this segmentation example, you could create three individual cohorts to see the performance of each segment separately.
Plating your dish
Now you have 10’s, maybe even 100’s of cohorts and associated graphs. You started with the “overall” segment and then separated the customer base into segments. You made right-aligned and left-aligned cohorts. You visualized the data in graphs.
Now it’s time to show it to somebody.
Anybody who has seen a cohort triangle before knows exactly what to look for. If they don’t know what a cohort is, you should explain it to them the same way I explained it to you in this (and the last) article.
No matter their experience level, the cohorts are only “information”. It’s up to you to translate meaning from the data. The final product is often shown as a combination of a deck and a spreadsheet. The deck includes screenshots and descriptions of the relevant cohorts and graphs. The spreadsheet includes everything, because once somebody realizes the power of this information they love to dig in.
Sample cohort spreadsheet
Here’s a sample spreadsheet you can use to get an idea of what a detailed cohort analysis looks like. This is a fictitious company in the “box of the month” space, growing fast, with segmentation and all the chef-y goodness from this article.
How to get help
Cohort analysis is part of Acquitention strategy at Epic Labs. Acquitention is “acquisition + retention” strategy, and includes cohort analysis.
We’ve analyzed billions of dollars of customer revenue, and helped improve companies all over the world. A simple Acquitention Heuristic can be done in days, and follow-on strategy implementation is available as well. We often increase retention by 5+% percent, improve marketing conversion by 10%, and we’re willing to structure our fee based on your success.
Get in touch with us at https://epic.so to talk about your company and your growth goals.
