Farfetch IPO: Strong Unit Economics Support Its Projected Valuation

For a Q2 2019 update on Farfetch stock price performance, read our post “What happened at Farfetch in Q2 2019? A CBCV Perspective“

Online luxury fashion retail marketplace Farfetch IPO’d today at a valuation of $5.7 billion, selling shares for $20 each. Taken at face value, this is a relatively steep valuation – it would be selling for ~12x revenues despite a -25% EBITDA margin over the past twelve months.

A lot of the discussion – bullish and bearish – about the company’s post-IPO prospects have revolved around surface-level measures like revenue volume and growth, profit, whether to view Farfetch as a marketplace or a more traditional e-commerce retailer, and secular macro trends (Dealroom, SeekingAlpha, Bloomberg, Forbes).

What struck us the most when we peeked inside its pre-IPO F-1 filing, however, was the richness of the customer data that it provided. While it disclosed a number of informative customer-related data points, this chart showing total marketplace GMV by cohort was the one that really got our attention:

The thickness of each of those colored bands over time was a heartening sign that repeat customer activity is strong here. As unabashed Customer-Based Corporate Valuation junkies, seeing this customer data gave us an irresistible urge to fit statistical models to it so that we can go beneath the surface-level metrics to better understand the unit economics which lie beneath. We want to know how much Farfetch was spending to acquire new customers, how that cost has evolved over time, how much repeat business those customers were bringing in after they were acquired, how this all rolls up into customer lifetime value (CLV), and ultimately, into the overall valuation for Farfetch as a whole.

The data that Farfetch disclosed allows us to answer all these questions.

What we found through our analysis surprised us in a good way. Overall, we saw very strong customer monetization, marginal contribution profitability and stable CAC, resulting in a healthy marketing ROI and total valuation in line with the IPO range, assuming the company is able to generate operating leverage. These were our main findings:

1. After customers are acquired, they have very long lifetimes and spend a lot while they are alive. The net present value of all future revenue a new customer generates for the firm is an eye-popping ~$4K.
2. If we were to assume that Farfetch earmarked all demand generation expenses for new customer acquisition, Farfetch is spending about ~$100 for each new customer it acquires.
3. CAC has been very stable, if not slightly declining, over the past few years.
4. Our most conservative variable profitability assumptions imply Farfetch customers generate approximately $500 in marginal profits after they are acquired, in net present value terms. We suspect their post-acquisition customer values may actually be closer to $1,000 with more realistic assumptions about their variable cost structure. Netting out the CAC that Farfetch spent to acquire these customers, this implies a customer lifetime value of ~$400-900, a ~400-900% return on their customer acquisition spend.
5. Farfetch has attractive cash flow dynamics. Its current non-financial working capital ratio to sales stands at -17%, dramatically reducing their cash needs as they grow and making them less beholden to the capital markets as they grow their way into profitability in coming years.
6. If Farfetch is able to continue acquiring customers with unit-level performance comparable to what we have seen to-date and if it is able to bring the G&A margins down to the level of its peers,  we forecast that Farfetch’s current valuation is entirely justified on a discounted cash flow basis. Our fair value estimate assuming its unit economics remain stable is ~$20.
7. However, and this is very important, we would caution investors that this valuation depends very heavily upon profits that may happen many years in the future. As such, their valuation is highly sensitive to WACC and to expectations of the longer-term competitive landscape. Farfetch’s valuation uncertainty is very large.

In sum, Farfetch has done a very commendable job from a unit economics standpoint. Its marginal return on invested capital (read: customers) far exceeds their marginal WACC right now, which is what will allow it to grow into its relatively steep valuation. Newly acquired customers are like annuities, which will allow their revenues to remain stable or even expand after it moves past peak customer acquisition (think Amazon). The annuity-like nature of its customers should allow it to generate more operating leverage than their peers, as existing customers who are cheaper to maintain become a larger and larger proportion of its overall sales base. Those long lifetimes will also give Farfetch the opportunity to learn a lot more about its customers than it otherwise could if customer lifetimes were shorter, allowing Farfetch to better personalize its offering to individual customers over time. This could create a virtuous circle of value creation…

… but again, all of this will only be possible if it can continue bringing in high quality customers for many years to come. This is a big if.

We discuss the model next before elaborating upon these key results in more depth.

The Model

The model that we use is almost the same as the one that we used in our paper, forthcoming in the Journal of Marketing Research (JMR), on valuing non-subscription businesses (SSRN), with one key difference.​

Whereas our JMR paper estimated its model upon three customer-related metrics over time – customers acquired, active customers, and orders – we generalized our estimation framework here to be able to learn from the different data points that Farfetch made available in its F-1 filing. Farfetch disclosed total active customers, total orders, and total platform GMV. While it did not disclose total customers acquired, we could infer it and more from the cohort-by-cohort marketplace GMV data that it provided above. [We used pixel-tracking software to infer what the data in this chart implied actual cohort-level marketplace GMV to be.]

Otherwise, our modeling approach is the same. We posit models for the flow of customer acquisitions over time, how long acquired customers remain with the firm after they have been acquired, the number of orders they place while they are alive, and the amount they spend on those orders. We estimated the parameters of our model that imply model-based expectations that are most consistent with the observable data. As it turns out, our model fit the data remarkably well along all of the various data points that Farfetch disclosed.

Here is a chart of platform GMV versus our model-based expectation of it:

These are the corresponding plots for active customers and total orders over time:

Finally, this is a comparison of actual marketplace GMV by cohort (top chart) and the corresponding model-based estimates that we infer:

We can hardly tell the difference between these charts.

While the data noted above is all that we trained our model upon, there were a few other data points that Farfetch had disclosed which we deliberately set aside during model development so that we could validate our model on them:

1. Farfetch noted that it had acquired “240K+” customers in total by year-end 2013. Our model had inferred this figure to be 248K.
2. Farfetch noted that it had acquired “2.3M+” customers in total as of the date of their F-1 filing. Our model had inferred this figure to be 2.4MM.
3. Farfetch said it had acquired their millionth customer in 2016. Our model-based estimate of the date was June 2016.

This should be even more impressive given that we did not estimate our model on any customer acquisition data at all – we only triangulated our way to it through the cohort-by-cohort GMV data, the active customer data, and the total orders data.

In sum, we are very heartened that our generative model of how customers behave is so consistent with all of the observed data, despite our model’s relative simplicity. For those interested in accessing the underlying data and our fits, download this spreadsheet.

Results: extremely strong post-acquisition revenue and profitability

Given the strong empirical validity of the model we have fitted, we can explore what it suggests about the underlying unit economics of the business next.

Our model suggests that customers’ loyalty to the firm differ considerably across the customer base – many customers have very low loyalty, while others will likely remain with the firm for many years to come. Because of this, there is a fairly sharp drop-off in overall cohort activity right after a new customer cohort is acquired, as the low loyalty customers churn out. After that initial shakeout, the number of customers who remain with the firm remains relatively steady.

This smaller group of highly loyal customers are very valuable not only because of their long lifetimes, but also because we infer their overall spending with the firm tends to grow the longer they remain with the firm. We infer that customers who maintain their relationships with Farfetch spend about 20% more per year. To avoid extrapolating this growth too far out into the future, we assume this basket size expansion completely ceases after 5 years.

The combination of initial churn-related customer shakeout, stabilization of the customer base post-shakeout, and growing basket sizes from those who are retained implies that the average revenue per customer from a newly formed acquisition cohort will be very high, fall sharply, stabilize and then grow slowly thereafter. Below, we plot the expected revenue per customer for an average customer from the June 2018 acquisition cohort:

Their spend begins at a very high level, falls, stabilizes, and slowly starts to increase again. This dynamic is somewhat evident from the cohort-level platform GMV chart that we showed at the beginning of this note… but we can say exactly what is going on from a revenue standpoint, and how that revenue is coming about.

Turning from revenue to variable profitability, we make the following assumptions:

1. Farfetch’s gross margin remains constant at its current level of 51.4%.
2. We forecast that Technology and G&A expenses are partially fixed, and partially effectively variable in nature. We assume that over the next 5 and 10 years, these line items will both fall to 9% of sales. These long-term expense ratios are assumed to be effectively variable costs. G&A is currently running in excess of 40% of sales but has been falling sharply, and other more mature e-commerce marketplaces like eBay currently have G&A levels at this level (accounting for stock-based compensation properly).
3. We assume stock-based compensation will remain at 5.3% of sales, again consistent (if not conservative) relative to Farfetch’s peer set.

D&A is a non-cash expense which we treat as fixed in nature, while demand generation expense we assume represents the cost to acquire these customers in the first place, and thus should not factor into Farfetch’s ongoing variable profit margin. This implies that Farfetch’s steady-state variable margin is 28%, which we apply to the revenue curve shown in the plot above to arrive at our post-acquisition customer value estimate of $1K, after discounting those profits back at Farfetch’s steady state WACC of 10%. This result is not sensitive to WACC – at a 12% WACC, the post-acquisition value falls to ~$900.

The fact that Farfetch is able to acquire these customers for only $100 is highly impressive.

Results: CAC has been stable, if not slightly declining over time

Assuming that all demand generation expense is allocated to new customers, Farfetch’s CAC has been very steady over the past few years, hovering at about $100 per customer. If anything, its CAC has been in decline over the past few years. Note that Farfetch internally uses only demand generation expense attributable to new customer acquisition to calculate CAC, but unfortunately, they do not disclose these numbers.

This stands in stark contrast to an analysis we had done earlier of Blue Apron, where CAC was sharply rising in the period leading up to their IPO. There is no evidence whatsoever of Farfetch “gunning it” on marketing spend to make revenue growth look artificially strong going into the IPO – if anything, the reverse seems true. Again, a very heartening sign indeed.

Results: if unit economics remain at current levels, its IPO price is entirely justified

As we have now done a number of times in the past, we use this fitted model for the flow of new customers acquired, their repeat orders, and the amount spent on those orders to generate revenue forecasts over time. In addition to the expense line item forecasts we mentioned in the last section, we made the following additional assumptions to generate long-run revenue forecasts and translate those revenue forecasts into free cash flow estimates (and ultimately, the overall valuation of the firm):

1. We assume that Farfetch’s currently enviable non-financial working capital percentage of sales falls from -17% to -5%, to be conservative (and to fall in line with its peer set).
2. We assume that capex will fall in line with D&A by 5 years from now, at approximately 3.4% of sales.
3. We assume that demand generation expense will fall to 10% of sales. If the repeat behavior of customers remains as it is, this may end up being overly conservative.
4. We assume the total number of people who may eventually be acquired by the firm is 100MM. While this number may seem arbitrary, we believe it is a reasonable figure in light of known statistics for the global luxury market, and we include a parameter in the model to allow the total number of eventual acquirers to fall below 100MM. This way, the data tells us what the effective size of the market is.

If these unit economic trends persist, we should see very strong operating leverage from the business. The combination of margin expansion, negative working capital, and revenue stability will allow Farfetch to lever its balance sheet, bringing down its cost of capital to the 10% level that we are assuming. All told, this would give Farfetch a DCF valuation of ~$20 per share.

This is a heartening sign, particularly given the relatively high valuations other high-growth companies are receiving. As noted at the beginning of this note, though, this valuation rests entirely upon Farfetch’s ability to continue acquiring customers over the next 10 years as valuable as the ones that it has acquired over the last 10 years, including its ability to continue significantly improve G&A margins.

Time will tell, but its good track record to-date should be grounds for cautious optimism.

Wonkish Comments
Farfetch disclosed an LTV to CAC chart in its filings, and it is helpful to lay out how this disclosure reconciles to our results. This is the chart that Farfetch disclosed in its F-1 filing:

While our figures are consistent with theirs (both over time and across acquisition cohorts), our LTV / CAC figures differ from theirs in a few notable ways:

1. Our LTV figures include profits that these customers will generate more than 24 months out, and as we can see from Farfetch’s own cohort-by-cohort charts, there is considerable additional value out there. This value should not be ignored.
2. We discount future expected cash flows back at Farfetch’s steady-state WACC, which we estimate to be 10%, while Farfetch’s figures do not apply discounting.
3. Farfetch assumes variable profits are equal to their contribution profits less demand generation expense allocated towards existing customers. This is different from what we have assumed in two ways:

• We believe operating expenses that are effectively variable in nature should also be deducted (namely, G&A, technology, and stock-based compensation). We deduct these expenses.
• Farfetch did not disclose how much of their demand generation expense (DGE) was allocated to new customer acquisition versus existing customers over time. Because of this lack of data, we allocate all DGE to new customer acquisition.

September 21, 2018

Daniel McCarthy and Peter Fader are co-founders and Val Rastorguev is a Director at Theta Equity Partners, a company specialized in Customer-Based Corporate Valuation services — valuing firms by forecasting what their current and future customers will likely do.

This article is for informational and educational purposes only, you should not construe any such information or other material as investment, financial, or other advice. Nothing contained in this article constitutes a solicitation, recommendation, endorsement, or offer by Theta Equity Partners to buy or sell any securities or other financial instruments.