Blue Apron's IPO Filing Implies Troubling Customer Retention

There has been a lot of buzz surrounding Blue Apron’s S-1 for an initial public offering. Many analysts have dived in it to glean insights into Blue Apron’s valuation. However, virtually all the analysis has focused upon traditional financial metrics such as revenues and net profits (e.g., "revenues and losses are both going up!") or surface level analysis of their disclosed customer metrics (e.g., "average orders and average revenue per customer were down year on year!"). 

But what did the S-1 actually say about the true underlying propensity of subscribers to acquire and stay with Blue Apron? There has been far less written about this, because as Tren Griffin and PitchBook Data had noted, churn metrics were surprisingly absent from the filing (in stark contrast, for example, to the treasure trove of customer data disclosed in furniture e-commerce retailer Wayfair’s S-1). By the end of this post, I will show you that we can back out an estimate of what their retention curve looks like from the few figures that Blue Apron has disclosed. While business intelligence firms such as 1010data, SecondMeasure, and Cardlytics have estimated what Blue Apron's and subscription meal delivery businesses' "survival curve" may look like, they are doing so off of panels of users that may not necessarily be reflective of Blue Apron's overall subscriber base. Working with audited data from Blue Apron itself about Blue Apron's entire subscriber base can be a useful complement, despite the limited nature of the disclosures they have put out to the public.

Cutting to the punch, this is what Blue Apron's data tells us about subscriber retention, and more below on how we get here:

Blue Apron Discloses Customer Churn... Sort Of

While Blue Apron didn’t explicitly disclose churn metrics, they did implicitly. By disclosing their “cost per customer” and their historical marketing expenses, we can back out how many customers they acquired between Q1 2014 - Q1 2017. 

Doing out the math, they acquired 2.9MM customers in this period. Assuming they had a very small number of customers at the start of 2014 (a seemingly reasonable assumption), this would suggest that they acquired 2.9MM customers and then lost another 1.9MM of them to finish Q1 2017 with 1MM subscribers. But is this good or bad?

Using Probability Models To "Back Out" Blue Apron's Customer Retention

Here's the fun part - using some survival model math from my paper on customer based corporate valuation paper for subscription businesses with Peter Fader (whr.tn/CorpValPaper1 for download; here for Journal of Marketing version) , we can get the quarterly customer behavior that is most consistent with the customer data that Blue Apron has disclosed... even though Blue Apron never disclosed it! Check out our paper for more on how we do it. Wonkish comments are at the bottom of this note.

The Results: Low Customer Retention

Fitting the model for customer acquisition and retention at Blue Apron yielded some eye-opening findings about Blue Apron’s customer retention: 62%+ of customers churn within 6 months (NB: given the sparsity of Blue Apron's disclosures and the front-loaded nature of the acquisition curve, I would only be confident in the retention curve up to one year out). If retention of future customers is consistent with that of historically acquired customers, then this would suggest that Blue Apron's future success will be heavily reliant upon future customer acquisition – finding many new customers to acquire, and being able to acquire them cheaply. It will be hard to get off the “acquisitions treadmill” because the firm loses customers very quickly after acquiring them.

More fundamentally, it will be harder for Blue Apron to turn a profit in the future because it will need to continue spending large sums on customer acquisition expenses. Blue Apron has spent $94 to acquire new customers on average historically, yet makes around $25 per month in gross profit per active subscriber (using Q1 2017 numbers). Even with very optimistic assumptions about what costs are variable in nature, Blue Apron likely hits break-even when customers have been around for at least 4.5 months.

Moreover, as pointed out by Valery Rastorguev, our model would also suggest that the cost per customer acquisition has been rising lately. While $94 is the historical average amount that Blue Apron has spent between Q1 2014 and Q1 2017, more recent acquisition costs are likely well north of this (and if they aren't, then acquisitions are even more front-end loaded than my model would suggest, implying even worse retention figures). The plot below contains my estimates for CAC over time. I estimate CAC in Q1 2017 to be $169, which would imply a break-even point on newly acquired customers of ~8 months, implying that Blue Apron will lose money on 66% of its customers.

While these data points make it hard to be bullish on Blue Apron’s valuation, they make me even more bullish on “customer-based corporate valuation.” Investors should be demanding metrics such as the number of active customers and the number of customers acquired and/or lost over time. More data would certainly lessen the uncertainty associated with the estimates provided in this analysis, which begs the question - is Blue Apron offering a relatively meager serving of customer metrics because of the troubling implications those disclosures would have for customer retention?

I am very grateful for many insightful conversations with Mark Zubenko, Eric Schwartz, and Valery Rastorguev. Errors and omissions are all mine.

Wonkish Comments on model:

1. I use two hazard models -- for the acquisition and the retention of customers over time -- to estimate monthly customer behaviors most consistent with Blue Apron's quarterly disclosures.
2. The estimation procedure is virtually the same as what I laid out in the customer-based corporate valuation paper.
3. I used "smoothed" monthly estimates of Blue Apron's marketing expenses based on Blue Apron's quarterly disclosures to get estimated monthly CAC figures.