Uber’s Internal Labor Market Through a SaaS Lens
Uber released some data on its own labor market in 2015. I thought it would be interesting to see how many more metrics I could derive on top of what they’ve shared.
Even though they’re not a SaaS business, I quite like the paradigm of “churn” and “lifetime” as a way to determine total lifetime value for Uber drivers. You can read more about CAC and LTV here, I’m cribbing some simplistic models from that to try to apply those concepts to their driver revenue data. Even though drivers will likely stop and start driving on an individual basis given their flexible hours, looking at total population adherence to driving should provide some useful insights about the expected lifetime activity of an individual driver.
Where no tabular data was shared, I used WebPlotDigitizer to translate graphs into estimated data points for analysis.
Here are my 3 source charts:
Total Active Drivers
New Drivers Added
Driver Retention
Data nuance:
- Chart 1, Active Drivers, only includes drivers who completed 4 rides
- Chart 2, New Drivers, includes all drivers
- Chart 3, Driver Retention, only includes drivers who completed 4 rides, and refers to early 2013’s cohorts
Retention
Based on these three alone, you can triangulate a few figures. One is what their active driver total should be given their example driver retention from 2013, when they were adding ~1,500 drivers / month.
Expected total active drivers in November 2014 based on churning out new cohorts using churn rates from early 2013: 225k total drivers.
Total actual active drivers in November 2014: 160k
A key difference in definitions is whether or not a driver has made 4 rides or not. So it’s possible that only 71% of new drivers make 4 rides. Driver churn could also be 2x worse to derive that total (simple monthly churn from 8% to 16%). Either of those modifications to new cohort additions will bring the estimated November 2014 total down to the actual number of 160k.
It seems intuitive to me that both of these are true to some extent, so to split the difference I’ll estimate that 87% of drivers provide 4 rides, and that churn is about 1.5x worse in recent cohorts (13% instead of 8%).
Cost to Acquire Customers (Drivers) — CAC
Cost to acquire the drivers is a big blind spot in this publicly shared data, it makes sense that Uber wouldn’t want to share their sales and marketing costs. That said, there’s some public info I can use to gather a range. Lyft’s average CAC was leaked as $530. Uber’s aggressive referral bonuses of $600 are a form of pay per acquisition deal as well. It’s tough to say what side of their CAC distribution that lies on, but I’m operating on the assumption it’s on the high end. For some businesses, referrals may reflect the lowest tier of customers/users. Purely on a hunch, a driver referred by another driver could probably be a very active one, so Uber might stomach an above average CPA for them, putting their actual average more in the $450–500 range.
A “fully loaded” CAC requires not just marketing cost to acquire, but also sales and overhead as well. This whole section is going to be filled with assumptions, so I’ll assume these modifiers bring up the total cost by 1.5x, based purely on my own experience.
I think $700 is a fair amount to estimate for CAC. I’ll use a range around that value later on in this analysis.
Driver Lifetime Value — LTV
They present driver earnings segmented by activity level, I can derive weighted average monthly earnings per driver from that. Then using their revenue share by segment, estimate Uber’s lifetime revenue generated per driver acquired (assuming unrestricted supply of riders as they scale).
So $1,600 per month, of which Uber collects ~22% (weighted average of UberX/Black revshare), so $350.
Simple inversion of previously calculated monthly churn of 13% puts the average lifetime at 8 months (1/.13 = 7.7), so I’m estimating LTV as 8 * $350, or $2800.
Payback
Layering this data in as average revenue per driver against expected marketing spend levels for an example cohort of November 2013, you get an interesting picture.
So at different CAC levels, we see decent sized differences in marketing payback months. Remember, these are Cost per Driver with 4 Active Rides to stay consistent with previous definitions.
So CAC is a highly sensitive subject then, and I’m sure Uber has some tight controls around it to optimize for the payback period they’d like on their marketing spend.
Total Active Driver Ceiling
In high churn businesses, there’s a mathematical equivalence point at which constant marketing acquisitions balance out with churn rates and total active users ceases to rocketship up and plateaus out.
Assuming 50k new adds per month based on their recent data ($20m in marketing per month, or $240m yearly) at a roughly constant rate, and a ~13% churn rate, they theoretically top out at ~384k drivers.
This is all extrapolation from end of 2014 data, so it’s possible my estimate of 50k monthly adds (higher than their 38k monthly adds at the end of 2014) has already been outstripped. Changing that value to 70k could move it up to about an 538k estimated plateau.
BLS claims there are <200k taxi drivers in the US employed today. Seems highly likely UberX will increase that market size by quite a bit, but how much I’m not sure. Seems like greater than 2x growth in market size is pretty optimistic, so 538k in sustaining active drivers on the platform within a few years seems pretty reasonable to me.
UberX’s driver earnings premium on top of median taxi driver earnings ($17 UberX vs. $11.16 Taxi via BLS) seems unlikely to stick. As UberX starts cannibalize current taxi drivers with lower wage expectations, I’d expect downward pressure on earnings per month per driver as they scale. By how much is tough to say. $320 in Uber’s revenue coming down to ~$250 doesn’t seem out of the question though based on that wage decrease.
Even with that, based on the above quantities, that’s $1.2-1.6B in expected revenue yearly in the US alone. It’s a flagship market at the moment but I imagine they are aggressively trying to make that not the case, as I’m sure the global taxi market dwarfs the US one, even if not on a per capita basis.
Conclusions
Like many others, I view Uber and other marketplaces as systems with both positive and negative effects. We’ve surveyed ZipRecruiter users who have experience working for companies like Uber, and while people are thankful for the flexible work opportunity, they’re also dubious of their long-term career prospects. This seems borne out in the average working lifetime of drivers, the flexibility leads to drivers leaving within a few years.
I was skeptical of Uber’s publicly stated driver churn numbers, and the purpose of starting this analysis was to verify that they were at least internally consistent. Any time I see differing definitions in data, I start questioning if those are to someone’s benefit or not. Turns out it does indeed worsen the picture a little to use consistent definitions. That said, it doesn’t worsen it that much, and Uber is able to finance the increased expense of driver churn because their overall profitability metrics look great.
Viewing drivers as subscribers that have churn, lifetimes, and average revenues is a somewhat useful way to view the world, and leads to a pretty profitable view of Uber’s business.
- LTV:CAC of 4x (greater than 3x usual SaaS benchmark)
- Payback of ~1–2 years (within SaaS benchmarks of 2 years)
- Plateau of total active drivers from 400–600k (2–3x higher than current active taxi drivers)
Uber seems to have released this data as part of an effort to understand their own driver behavior and defend several key points they’ve made about driver earnings. They may have expected interested readers to make the logical leaps I did, and I’m glad to see the transparency they have around studying themselves and the market they’re creating.
I certainly find it pretty fascinating to analyze a segment of the workforce that previously had little to no motion, and is now seeing explosive growth. Hopefully they continue to release periodic reports like this so I can update this with 2015 and 2016 data! Their continued explosive growth last year and this year are sure to be fun to dig into.
5/16/16 Update: I’ve iterated on my LTV projection using an SQL methodology, check it out for some updated estimates. Main takeaway is that a more nuanced churn curve reduces the impact of steep initial churn, leading to a $3.6k LTV rather than this post’s estimated $2.8k LTV.
