"+1" to let us know.
This is the first part of a 3-part series discussing techniques to calculate Customer Lifetime Value (CLV) in a retail setting. We will discuss the pros and cons of various CLV estimation approaches.
Evaluating techniques to calculate CLV
- Customer Acquisition. Knowing the CLV for a new customer puts a ceiling on acquisition spend, particularly when we calculate CLV for each acquisition channel.
- Customer Retention. Knowing the remaining lifetime value of existing customers is critical for retention marketing. As we previously wrote, retention marketing requires long-term optimization. Lifetime value projections provide a much-needed baseline for measuring the impact of retention marketing strategies.
There are many ways a firm can calculate CLV, and there is no universally accepted standard. If you want to base decisions on CLV numbers, it’s important to test the accuracy of your CLV calculations. Fortunately, there are two simple tests we can use to determine the accuracy of a given CLV calculation approach:
- Acquisition test: we can predict what a group of new customers will spend over their first twelve months. Twelve months later, we can check up on the group and see what actually happened.
- Retention test: we can find a group of customers who joined 6 months ago, and predict what that group will spend over their following six months. We can check up on them six months later to see what actually happened.*
You’ll notice we put a finite cap on the time horizons for these tests. It’s a bit difficult to check how accurate a prediction is when you have to wait… forever. Regardless of which technique is used to calculate CLV, it should be a simple task to make predictions for just 6-12 months into the future.
The problems with ARPU-based CLV
In part 1 of this series, we focus on a common, but often inaccurate, technique to calculate CLV: ARPU.
ARPU stands for Average Revenue per User. In practice, this metric is used on a per month basis: the average revenue a customer spends per month.
- The upside of this approach: it’s dead simple to calculate. Sum up your revenue, count your customers, count the months your customers have been around. Simple division, and you have your number.
- The downside: of all approaches, this is, on average, the least accurate technique around.
The main problem with ARPU is that it treats all customer-months the same. Consider what happens when a group of customers make their first purchase. The average revenue per customer is very high in that first month – heck, everyone is buying something. However, over time, some customers fade away. If you zoom in on month 3, a large number of customers are idle, so the average revenue per customer is lower. Fast forward to month 20 and the average revenue per customer is lower still.
This graph illustrates why ARPU has issues. ARPU is, by definition, the value of the average month. The average value falls somewhere between the high values we see in early months and the low values we see in later months. ARPU will therefore be biased based on the balance of newer and older customers in your data set. If you have many customers in early months, ARPU will be too high – likewise, the opposite is true if the majority of your customers joined a long time ago.
ARPU can create terrible CLV projections
We picked a data set from a high-growth client to illustrate how dangerous ARPU can be, and we ran the two tests mentioned above.
- New customer test: projecting twelve months of revenue for new customers. ARPU came in 55% too high compared with actual numbers.
- Existing Customer Test: projecting six months of upcoming revenue for customers who are six months old: ARPU came in 122% too high compared with actual numbers.
People generally use ARPU as a back-of-the-napkin approach to estimate CLV. Unfortunately, as we see here, ARPU projections can paint a very misleading picture – and worse yet, this is the norm, not the exception. In general, we see ARPU return results like these when:
- Your business is experiencing a lot of growth. In that case, you’ll have so many customers in early, high-value months that your CLV projections will be too high.
- Your business is new – you don’t have enough “late months” yet in the data set, so here, too, your numbers will be too high.
- Your business has changed – revenues or margins are different today than they were in the past, so your numbers will be biased by old trends.
Of course, there are situations in which ARPU provides accurate projections. This happens in the rare circumstance where the balance of early and late months are even for the duration of your projection.
The moral of the story: it’s critical to test your CLV calculations.
Next up: The problems with Cohort-based CLV
ARPU’s big shortcoming is a blending of early months and late months. In the next part of this series, we’ll look at the pros and cons of a technique that tries to mitigate this problem: cohort-based CLV.
Interested in learning more about CLV in retail? Let us know!
*If you don’t want to wait a year to evaluate the accuracy of your predictions you can perform a holdout test. For example on January 31th 2012, we can use data up to December 2010 to come up with a prediction for the customers who made their first purchase in January 2011. Then we can see how the prediction did against the actual numbers.
The Adobe 2018 Digital Trends report was recently released and it gave its readers some interesting insights into this year’s top marketing priorities around data, design, and technology. The report, which consisted of close to 13,000 participants from various industries including retail, e-commerce and the marketing agency sector, revealed the trends that distinguished the […]