Are cohorts the future of public company analysis?
I originally published this post in my weekly-ish newsletter “Drinking from the Firehose.” Sign up here to receive columns like this (and more) in your inbox: tinyletter.com/ataussig.
The ground truth for modern consumer businesses is cohorted purchase data. I’ve written extensively about this topic on my blog. Cohorts demonstrate how much value a customer can produce over time and inform a business’s willingness to pay to acquire a customer. All reasonable financial models that make predictions about future revenue and earnings are effectively “roll ups” of new cohorts of customers.
Today, publicly traded firms only disclose the roll up. We occasionally gain some insight into cohorts in an IPO prospectus, as we can see below from Blue Apron’s recent S-1 filing:
Yet, these cohorts are rarely updated on an ongoing basis in a 10-Q or 10-K. Because public market investors lack access to this granular, ground truth data, corporate valuation techniques instead rely on somewhat arbitrary forward projections of a business’s revenue, earnings, and cash flows. If only we had access to a company’s cohort data, we could make better predictions about the future and create a more “customer-based” corporate valuation model.
A few researchers at Wharton and Emory have recently attempted to bridge the gap between public financial data and cohort data by estimating the latter from the former. What may sound like magic comes from a rigorous, statistical formulation. The results have had strong implications on several stocks, including Blue Apron’s. In an article published a few days before its IPO, the researchers revealed declining cumulative net revenue in recent cohorts. I extracted some of their data from this Dropbox link:
With an estimated contribution margin of 26% and cost to acquire a customer (CAC) of $147, this analysis implies a marketing payback period of 13 months. That’s not too bad for public company with hundreds of millions in revenue.
However, because the October 2016 cohort performed 25% worse than the last 3-year cohort in April 2014, the best estimate for 3-year cumulative net revenue for the October 2016 cohort is $693. That would imply a lifetime value to CAC ratio of only 1.2x, less than half what was disclosed in the S-1 (~2.6x over 3-year period). Alarmingly, new customers are only slightly more valuable than CAC.
Subscription businesses like Blue Apron are only a small segment of the consumer universe. A subscription represents a contractual obligation between a customer and a business, and as such the concept of “ending customers in a period” is meaningful. However, most consumer businesses do not have the luxury of knowing, with high accuracy, whether or not a given customer will return a few months later. As such, public consumer businesses normally disclose “active customers,” which can be defined on an arbitrary backwards looking time horizon (commonly 12 months). Unpacking unit economics from these businesses is therefore more challenging.
That’s why, this week, I was delighted to see these same researchers apply a similar analysis to several non-subscription businesses: Overstock and Wayfair. The paper is linked here and below. They modeled repeat behavior and churn probability as two separate statistical processes. Both are related to the probability of survival of a population during a given waiting time. This more rigorous mathematical approach to repeat purchasing and churn is a better way for companies to model these important metrics. It also resulted in unit economics that seem to match reality:
My big takeaway from this paper is that more companies — both public and private — should begin modeling customer engagement as a multi-step, statistical process. It should lead to a deeper understanding of customer behavior and help predict future cash flows from customers.
In addition, I hope that cohort analysis gains traction as a standard practice to evaluate publicly traded companies. Perhaps papers like these will place pressure on firms to release ongoing cohort data themselves, which would create a more transparent marketplace for consumer public stocks.
“Customer-Based Corporate Valuation for Publicly Traded Non-Contractual Firms” (Link)
As investors realize these many uses for customer metrics, the demand for their disclosure will continue to grow. This would not be the first time — one of the most commonly disclosed and tracked retail metrics, same store sales (SSS), became popular after a Wall Street analyst used it to uncover the true underlying financial condition of a fast-growing retailer in the 1970’s (Blumenthal 2008). Customer metrics like the ones discussed here allow investors to track the quality of existing customers much the same way that SSS allows investors to track the quality of existing stores. With physical stores ceding share to internet-based retail, the need for such metrics is more important than ever.
