Blue Apron CBCV update: a cocktail of good and bad

Given my last proper CBCV analyses on Blue Apron was from November 2018, and the fact that I'm about to teach a few lectures referencing the company through the case study I co-wrote with Eric Schwartz, I decided to dive back in to see how things have evolved since then.

While a lot has remained the same, a surprising number of things have changed. These were my main takeaways from my analysis:

1. Customers expected lifetimes are about 40% shorter than they used to be, and currently stands at an estimated 7.6 months. I believe that this is due to the value of the brand eroding. HelloFresh, Blue Apron's main competitor in the US, has seen stable retention patterns -- HelloFresh's retention used to be worse, but is now even with Blue Apron's.
2. Topline monetization of customers while they are alive has gotten marginally worse as well, both in terms of how much customers are spending when they first become customers as well as their ongoing spends thereafter.
3. On the flipside, CAC appears to have stabilized and currently sits at an estimated $139 on a trailing twelve month basis.
4. Also to the positive, we have seen very dramatic, very impressive gross margin improvement. The company's gross margin used to sit at 32%, and now currently sits at 40%.
5. Taken as a whole, netting these various factors out against one another, I infer an estimated CLV of approximately $54, which equates to a fairly low marketing ROI of approximately 38%. This is probably not going to be enough to generate real value over the longer term, given the company's fixed cost base and the potential for future customer acquisitions.
6. All of the trends noted here are consistent between my own statistical model, and that of my friends at Second Measure. It was nice to see that our substantive conclusions moved in the same direction, and reinforces the validity of the conclusions I've reached here.

Below I discuss the results in more detail...

Model specification

As before, the underlying model I specified is virtually the same as the one I used in my original analysis on them -- there is one model that characterizes the inflow of customers over time, another for how customers are retained after they are acquired, a third for the number of orders they place while they are alive, and a fourth for the revenue associated with each order. I then simply calibrate the parameters of these various models to whatever customer-related disclosures happen to be available, and use this to drive future revenues, profits, and cash flows. For more on this methodology, see my paper with Wharton and LBS marketing professors Peter Fader and Bruce Hardie in the Journal of Marketing.

I was only able to pull together my original analysis on Blue Apron back in June 2017 because Blue Apron happened to disclose a good amount of customer data in their S-1. While most of those disclosures haven't seen the light of day since, the fact that they have continued to disclose figures such as active customers and orders (in addition to their traditional financial information) gives us just enough to continue updating the model and get diagnostic results.

I would caution that the lack of detailed disclosures more recently introduces some uncertainty about the current state of Blue Apron's unit economics (folks at Blue Apron: please disclose more customer data!). It is for this reason that data from Second Measure, who very kindly shared some of their credit card panel data with me for this note, becomes particularly useful. Second Measure has access to data that is very granular (and thus rich), and is current ~up to the present day, making it a highly useful complement to my usual SEC-only analysis. In this particular analysis, I went one step further and incorporated Second Measure's data directly into my customer model (for more on how this sort of procedure can be done, see my paper with Elliot Oblander on data fusion).

Blue Apron model validation

If you've been following my work, you know the drill by now -- as empiricists, we need to make sure that our model -- any model -- validates well before we can trust it. It is very easy to cobble together interesting sounding figures and spin an intelligent sounding story around it, but if the figures/story do not validate well, beware.

Thankfully here, the model continues to validate very well, regardless of the disclosure we happen to be looking at. All of the S-1-specific disclosures fit as well as ever. Below I focus on the charts for which we have more recent disclosures -- namely, active customers, total orders, and revenues:

If anything, these results may paint a slightly over-optimistic picture -- model misses (while generally small) tend to be on the upside more recently. Having passed our validation check, we turn next to the unit economic insights that arise from the model.

Blue Apron's retention has significantly worsened

The first implication from our model is that their retention profile, on the whole, has significantly weakened.

This is the implied average retention curve for Blue Apron's customer base currently:

This curve implies a sharply lower expected lifetime for new subscribers -- about 40% lower -- than the last one I had fit on them. As of the S-1, I inferred an expected lifetime of previously acquired customers of about 12.7 months. This time the analogous figure was about 7.6 months.

Coincidentally, this is also just about the same decline that we see in the Second Measure data. I fit models to Second Measure's retention data for all acquisition cohorts, then used these models to compute the average expected lifetime of customers acquired in 2016 versus 2018. The models fit exceedingly well, giving us confidence in the resulting predictions (happy to share the fits with those who are interested, but some indication of this goodness of fit can be seen below).

According to the Second Measure data, a typical customer from 2018 had an expected lifetime that was about 40% lower than their 2016 customers. This is visually obvious when we look at the retention curves (with model fits overlaid) over the next two years:

While the absolute level of the Second Measure E(lifetime) figures are not the same as our own, this is to be expected -- they are using a panel data set and not SEC data, and because of various reasons, those panel members may not be representative of the broader population. But it is fairly remarkable "model-free evidence" that their data infers a nearly identical shift in Blue Apron's retention patterns across cohorts relative to our own.

Fitting an analogous model for HelloFresh, I did not see this sort of deterioration. Don't get me wrong: things didn't get better for HelloFresh. They just didn't get any worse. My figures for HelloFresh were also consistent with what Second Measure's data suggested. For example, this is a chart comparing Blue Apron's 6-month retention rate to that of HelloFresh:

HelloFresh's 6-month retention rate used to be significantly worse than that of Blue Apron (which is consistent with the conclusion that I had reached in my original analysis of HelloFresh!). However we can see that that is no longer the case -- HelloFresh's retention has stayed relatively flat, while Blue Apron's has crashed down to HelloFresh's levels.

This is not a good sign for Blue Apron, all else being equal, especially because they have been very actively attempting to refocus their business around their very best customers, presumably at the expense of customer acquisition volume (see my comments here about their "getting religion"). If this is the best they can do from a retention standpoint after attempting to refocus their acquisition budget around their best customers, that is a problem.

Blue Apron's monetization has also deteriorated

I also infer that Blue Apron's monetization has weakened since the last analysis. When I re-ran the model, the parameters governing how spending while customers are alive is trending across acquisition cohorts fell. This was also independently reinforced through the Second Measure data. In the chart below, I look at the average amount spent by active customers, 0, 3, 6, 9, and 12 months after those customers were acquired, by acquisition cohort. For example, that light blue line at the bottom represents how much paying customers from different acquisition cohorts spend in their first month with Blue Apron:

We can see a very clear downward trend to that line (I overlaid a trendline to make this trend more evident). One could brush that off as being promotion-related, which, while negative, would be isolated. But we can also see a distinct slight downward trend for how active customers have been spending other numbers of months after acquisition as well. This is an indication that more recently acquired customers are simply spending less -- and that this is true holding aside what we would expect due to customer churn.

The silver lining: CAC has stabilized, and gross margin has significantly improved

While retention and monetization have weakened, CAC appears to have stabilized and Blue Apron's gross margin has improved.

This is a chart of Blue Apron's rolling 12 month average CAC:

From 2014 to mid-2016, Blue Apron was spending about $70 to acquire new customers. In the run-up to the IPO, this figure moved north of $100... and kept moving up. It appears to have peaked in December 2018 at around $157, and has begun moving down. CAC currently sits at approximately $139. This is not a good figure -- it is still higher than it was as of the S-1 -- but it is nevertheless nice to see that the trend is no longer up for now.

The higher CAC is, the better the customers need to be after they are acquired just to break even.

As noted above, though, the real improvement has been in the gross margin. Below, I plot quarterly revenues alongside the gross margin:

Despite the fact that revenue continues to fall, the gross margin continues to move to new highs. It currently stands at approximately 40%, well above the 30-32% they had been doing in late 2016/early 2017.

The bottom line

What does all of this mean for CLV at Blue Apron?

• They spend about $140 to acquire customers.
• Taking into account our projections for retention and monetization, customers now generate about $616 in revenues after they have been acquired -- this is their expected sales post-acquisition value or E(S-PAV). Scale this down by 32% to account for variable profits brings us to a profit-based E(PAV) of $197.
• Netting E(PAV) against CAC, their expected CLV is approximately $54, which equates to a marketing return on investment of approximately 38%.

It is this $54 variable profit that Blue Apron earns on these acquired customers that helps them pay off their fixed costs -- once those fixed costs are paid off, they can then turn a profit. While E(CLV) is positive, in marketing ROI terms it is not very large.

How can we think about Blue Apron's fixed costs? In Q2 2019, they had ~$115M in total expenses, of which we are assuming about $81M are variable in nature. This leaves us with ~$34M in quarterly expenses. Let's be generous and remove D&A from this, as a non-cash expense. This brings us to $26M in fixed costs per quarter. Let's be even more charitable and assume that Blue Apron can wean an additional $5M of these fixed costs away each quarter through operational efficiencies, while still maintaining the current level of "customer goodness." This would bring them to $21M in fixed costs.

One way of thinking about this is that in NPV terms, Blue Apron would need to acquire ~473K customers each quarter just to pay off their fixed costs. This will be really hard to do. Remember, they have 449K active customers in total right now (and have acquired an estimated 4.4M customers to date since inception, an estimated 750K of which were acquired in the past 12 months). Of course, this calculation is no more than suggestive because it doesn't account for the future value Blue Apron will extract from its existing customer base. But it does continue to suggest subpar returns at best.