In Part 1 I talked about understanding the strengths and weaknesses of calendar-based first purchase cohort analysis in understanding high value customers.
In this post I want to discuss decile analysis as an important complement to the above cohort analysis and how decile analysis can get to the heart of who truly is ‘high value.’
Quick note on terminology. Purists will say that deciles are just another type of cohorts and they are correct. The reason I use decile analysis is that cohort analysis is now so commonly meant to refer to calendar-based first purchase cohort analysis that I want to avoid confusion.
We all know customers are not equal. Companies have some form of a power law or Pareto distribution to the value of their customers i.e. where 80% of X comes from 20% of the customers.
A decile is just a tenth of dataset. By first sorting our customers by some attribute and then grouping the customers into ten equal groups by that attribute, we gain insights and can take better and faster actions. One thing I really like about this analysis is its extensibility. The better our attributes, the better our insights. We can go as deep as we want. But we can also keep it high level.
The key question is what are the attributes? This depends on the quality of our data and the time, effort and cost at getting better data.
ASIDE
I have seen a bunch of companies get into data dashboard rabbit holes. Massive dashboards lead to data warehouse projects lead to more dashboards and hires and work. I believe another great aspect of decile analysis is that it helps you make the decisions on tradeoffs. Is getting this next attribute really going to be worth the effort, time and cost? Is the juice going to be worth the squeeze?
I am going to next run through some examples of analyses and discuss what we can learn from them. And then end with the outline of how to do this yourself.
The following tables are created using dummy data meant to replicate a successful apparel brand. I am using these to illustrate the types of analyses and their value. This is not meant to be benchmarkable data.
Table 1 is a classic distribution of sales with the top 20% of customers driving 66% of the total sales. If anything, this is probably too light for the top 20%. This table is the easiest to put together and is a good place to start. But it’s a bad place to end. Just as ROAS can mislead because it focuses on revenue and not contribution profit, ending our decile analyses at sales will lead us to bad conclusions and actions.
Table 2 focuses on contribution profit. I am a huge proponent of cash and understanding cash flow. And while getting to cashflow per customer is possible and the ultimate goal, it requires excellent data. Getting to contribution profit is entirely doable and absolutely necessary, so for now, we are going to use contribution profit.
The first thing that jumps out is that the top 20% of customers now account for 76% of total contribution profit as opposed to 66% of total sales. Leaving sales behind and focusing on contribution profit means we are on the right track!
The second thing that stands out is how little contribution profit our bottom deciles add. Our bottom decile is adding just $236K in contribution profit. In fact our bottom 40% of customers are only adding an additional 2% of contribution profit. When I look at that I just see wasted capital or more specifically, misallocated capital. Think of the cumulative ad spend, inventory, interest, 3PL fees, time and effort spent on that 40%.
I added Table 2A because it brings home the stark differences between the best customers and the worst. Our best customers here are adding over $2K each in contribution profit. Our worst customers are adding just $16.
In this example, there are 15,000 people in each decile. So it’s not like we are talking about a handful of people. My first thoughts are around potential:
- If we can get 15,000 to give us so much contribution profit, why can’t get 30,000 or 45,000?
- We should not accept this distribution of outcomes as a given. We can create more high contribution customer and acquire more high contribution customers. What would our business look like if instead of an equal distribution of outcomes across the deciles, we were able to narrow the range and push more customers into the upper deciles?
You already know your product mix by sales. But do you know your product mix by deciles?
Top decile customers aren’t there by chance. They buy differently. So what products do they like? Conversely, look at the bottom deciles. They are almost entirely in tops and have near zero buying of outerwear.
While we have gained insights into what our best customers like to buy and can now better investigate their buying journeys to establish the patterns, we don’t want to write off the bottom deciles yet. Do top decile customers start with certain products and move their way up? Are bottom decile customers just great customers, but early in their journey with us?
Here we will also find products that could be cut. These are the products that customers buy from us once, but then don’t go on to buy from us again.
I am reminded of the backpack story from the CEO of a successful apparel brand. He said for years the company sold backpacks and put a lot of time and effort into design and product development. They also put a lot of ad spend behind backpacks. “Then someone finally looked at the data and realized that the backpack buyers never came back. They bought one backpack from us and we never saw them again.” To his credit, he immediately cut backpacks.
Operators podcast episode 57 has a great discussion about discounting. We are likely familiar with the arguments for and against and some of the clever ways to discount selectively.
But have you looked at your discounting by deciles? This is an extreme example, but unless your brand is constantly discounting and that’s the way people usually buy you, I bet you will find big differences between your best and worst deciles.
Discounting can serve a valuable function in recovering cash. This is especially true in apparel. And the temptation is huge. Also, who hasn’t hit the sale button when revenue was coming up short?
I believe this temptation should be avoided and discounts should be used sparingly and selectively. Pricing power is a sign of a strong brand. Two indications of pricing power are:
- Your sell through at full price
- Your ability to raise prices
Discounting is pernicious. As JC Penney found out, once you train your customers to buy on sale, it becomes a massive effort to get them to buy at full price.
Table 5 looks at your returns by decile. Like discounting, you are going to find surprising differences in the return behavior of your best customers from your worst. Some of this is clearly familiarity with the quality of the brand. And especially in apparel, getting the fit right across new products and vendors is super hard. Apparel brands also often train their customers to high return behavior. I see it in my own house where three dresses show up and two go back.
So what to do about returns? This is a great topic for a future post. For now, the deciles help us understand the impact of returns on our contribution profit and helps us identify the truly best customers. I think it also helps us think of creative ways to address returns so we can capture more contribution profit.
Hopefully you have seen the power and value of decile analysis. How else can we extend it?
- Lifetime by decile.
- Our deciles are for contribution profit over a defined preceding period i.e. past 12 months. How long have have our top deciles been with us past the last 12 months?
- LTV by decile
- We can extend the analysis period to include lifetime contribution profit per customer
- Note: this is going to be much harder than it sounds as it requires going back through the unit economics of previous products. For most companies, I suspect this level of effort is not worth it. A simpler way to get enough data is to just use average CP margins by year and then take revenue by customer multiplied by the average CP for that year. While not exact, it’s likely more than enough to give you a sense of the big delta in LTV between your top deciles and your lowest.
- CAC by decile
- This is definitely a topic for a separate post.
- How do you know a priori which decile your newly acquired customer will be? You can’t. But these are people. If you look at your top two deciles, what commonalities do these people have? We know what they bought first and their subsequent buying patterns. Our attribution tools give us some voodoo magic idea of which channels they came from. People far smarter about customer acquisition than me can come up with a ton of other signals.
- The point being, are we going to accept that customers are customers and we can’t really target acquiring higher value customers? If we believe we can, then shouldn’t we also accept higher CAC’s for higher value customers?
“Perfect is the enemy of good.” - Voltaire
Like any data exercise, good insights quickly are more valuable than precise insights months from now. Once you set this framework up, there will be endless ways to improve it. Just get it set up.
To do this analysis, we need:
- Our active customer file. We want to focus on active customers. Let’s define an Active Current Customer as anyone who has purchased in the past 12 months. Sort out your Active Current Customers.
- We want to see all items purchased for each customer. You could do this by month which will open new analysis, but you can also just put the items and their quantities purchased over the past 12 months.
- A table of contribution profits by item. Here again, you can get harder and more precise by doing each specific item. Or you can generalize products into categories and then do average contribution profit by category.
- So now for each customer, we know what they bought and their cumulative contribution profit. So we can now stack rank them top to bottom by their past 12 month contribution profit. So the customer with the very highest Contribution Profit over the past 12 months is #1 and the Active Customer with the very lowest is #xxxxxxxx at the bottom.
- Now divide this sorting into deciles. Each decile represents 10% of your active customer base. If you have 100,000 active customers, then each decile is 10,000 customers. So the top 10% of your customers once sorted are your top decile and so on.
Here’s the dummy data I set up for the illustration tables above.
https://docs.google.com/spreadsheets/d/11LpYTNHjzvRb08RJGojY8-D3wcx9ya6yhA42ck3GHl4/edit?usp=sharing
Previous posts
Understanding High Value Customers Part 1
https://elephantherdconsulting.com/blog/blog-database/understanding-high-value-customers-part-1
Defining Brand
https://elephantherdconsulting.com/blog/blog-database/defining-brand
Future Posts
Understanding High Value Customers Part 3
The Paradox of Product Expansion
The Accounting is Wrong
If you need help with this or have questions, please reach out.
I’m a gun for hire and provide financial consulting to premium brands in the form of short term, high ROI projects designed to help teams inflect, reach a new level and maximize their chances of success.
DM’s are open or schedule directly: https://calendly.com/elephantherd/video-with-ben