The Network Effects of Uber’s Master Plan
I am going to describe and illustrate what I believe to be Uber’s network effects model. This does not represent their entire business model or encompass all their strategies. It only covers the activities I believe strongly influence or result in network effects. I should note it also doesn’t mean I think execution is guaranteed. It is just a model to help think about how they work and where they may be headed based on the status quo.
Also, if you enjoy this I think you’ll also like my follow-up focused specifically on autonomous vehicle tech:
I like to think about network effects businesses as engines. In the physical world energy is transformed, exchanged, and wasted through the interactions between each part of an engine. When thinking about networks as engines, “value” takes the place of energy. Each interconnected piece transforms value as it is passed along the cycle. While a pipeline business does this linearly, a network effects business recycles the value, resulting in a feedback loop that (when designed correctly) feeds upon itself to create progressively more.
A well designed network will “spin” faster as more value is generated. But, just as an engine can be poorly designed, so too can a network. Pieces that are out of alignment cause it to wobble. In the physical world, this movement represents wasted energy. In a network, this represents an inbalance that destroys value (ex. Uber’s Surge on NYE destroying brand value).
Lack of oil or poor quality parts in an engine results in friction. In a network this represents clunky interactions and transaction friction. In extreme cases misalignment or friction will tear an engine apart, or cause it to seize up.
Getting an engine started requires fuel (capital) and a spark (the entrepreneur). The network operator is faced with trade-offs as the network begins generating value. It can either invest more into getting it running faster or it can let it build on itself organically. Network design and product-market fit is critical in this phase. It is easier to tune an engine before it starts running on all cylinders.
At some point monetization is probably necessary. I view monetization as a drivetrain, translating the output of the engine to the wheels. Two points are relevant here. First, a drivetrain is bound to cause what is called drivetrain power loss.
Drivetrain power loss means that bolting a transmission, drive shaft, axles, differentials, wheels, etc to the engine slows it down due to friction and inertia. Many entrepreneurs are faced with this dilemma. Monetization concepts often generate friction and certainly encounter implementation inertia. Monetization also represents value capture, which means energy is being syphoned from the system. If the amount is too large, the system ceases to be self-sustaining. Analogously, customers that are charged too much and suppliers that are paid too little will leave a network, shedding value.
So, what does this have to do with Uber? In 2014 David Sacks tweeted the following napkin sketch of Uber’s network model, which I found through Bill Gurley’s excellent blog post.
I love this sketch. It makes it easy to break down and analyse Uber’s network model. First, let me put it through my PowerPoint Machine:
So what is this model showing? Each arrow represents a positive value transfer.
- Driver(-partners) are motivated by demand, and so as demand increases, the supply of driver-partners should increase.
- Uber has systems in place to indirectly distribute driver-partners, which means more driver-partners should mean more geographic coverage.
- More coverage means two things by default: faster pickups and less driver-partner downtime.
- The likelihood of a consumer requesting a ride goes up as pickup time goes down, which means shorter pickup times results in more demand.
This completes the main circle. Before explaining the left side, I am going to make an adjustment:
In my opinion (this is really a six in one, half dozen in the other thing) less downtime results from faster pickups, not directly from more geographic coverage.
The driver-partner is only paid while they are giving rides. Driving to pick up a rider is unpaid time. This means the quicker a driver-partner can pick up its next costumer, the more of their total time on the road is paid time. If driver-partners are earning more per hour, this means Uber can charge consumers less, which may hurt driver-partners initially, but should generate more demand demand, raising their earnings, and repeating the virtuous cycle. I say “should”, since any change will generate ripples as it propogates through the system. Some driver-partners may not see their earnings stabilize, or may be unwilling to wait for the ripples to settle.
I will adjust the model a little to reflect that this network utilization results in a lower cost structure, which leads to the ability to charge less while maintaining the same aggregate payout to driver-partners:
In my final draft I took out equations deriving the profit function for the system, the operator, and the drivers after deciding the digression was potentially more confusing than helpful. If it’s something you’re interested in let me know.
A ride-sharing service’s profitability is directly related to the utilization rates of its cars and driver-partners. What this means is the more time driver-partners spend driving customers around, the lower the cost structure becomes, which in turn means prices can be lower while still covering costs. It does not, however, means that prices must be lower. Price (and the share sent to driver-partners) is a dial Uber can adjust to determine how much value it wants to drive back into the network, and how much it wants to harvest (slowing growth).
I should note this model ignores tiered pricing. Uber has segmented demand according to price since its earliest days. Uber Black holds the high end, and in 2014 UberX held the low end. UberPool is now at the low end in most markets.
Each segmentation represents a trade-off between price and quality of service, and each has its own pricing bandwidth defined by customers’ Willingness to Pay (WTP) and driver-partners’ Willingness to Drive (WTD). Uber is doing many things to both increase WTP and decrease WTD.
Aside: Anyone interested in the WTP of ride-sharing customers should check out this recent research from economists at the University of Chicago (yeah!), Oxford, and Uber. It turns out demand is inelastic, which means Uber will earn more if (and when) they can raise prices.
In my opinion, the network effects between the product lines are weak, which is why I choose not integrate the tiered pricing strategy into the model. Strong Uber Black service does not positively or negatively feed back into demand or supply for UberX service in an instructive way.
At the end of the day I see demand segmentation as a product-market fit decision for Uber in each market, not a piece of the network model. But, I am very interested in ideas about how I could be wrong.
It is hopefully now clear that as network utilization rises the cost structure improves, enabling the network to sustainably operate at lower prices. Like any network business, Uber will charge lower prices as it seeks to grow, since it wants value to circulate in the system, sustaining/accelerating growth. It also needs to keep prices low to fight competition and deter entry, which is the next important addition to the model:
I think there are some very smart people at Uber who are skilled at thinking about strategy from a multi-move game theory perspective (see: China, Austin). I think about Amazon as an analog for Uber in many ways, and this is one in particular. The following Bloomberg article talks about how Amazon responded to Diapers.com. Amazon built a reputation early on as a fierce competitor, willing to compete on price without mercy.
Similarly, Uber is building a reputation as a competitor willing to fight hard in price wars for as long as it takes. Over time, as they win more and more battles, their willingness to aggressively target competitors and their ability to sustain low prices due to economies of scale will become an entry deterrence, supporting and protecting demand.
The next addition to the model is my personal favorite: additional services.
Uber has created a distributed local logistics network. While its founding function is transporting people, that does not mean it cannot — and should not — transport other things. Throughout the day there is expected fluctuation in core rider demand as well as unexpected volatility. Just as a well diversified financial portfolio drives down volatility and improves returns, a diversification of demand (riders, food deliveries, etc) drives down network demand volatility and improves overall value creation (assuming costs are managed).
In my final exam I referred to the practice of adding plug-in services as business model “hybridization”. Uber is already experimenting with this through Uber Rush (courier service), Uber Eats (food delivery), and Uber Essentials (online ordering & delivery). The challenge with plug-in services like these is they probably come with higher variable cost and cost of goods sold, which must be balanced against the benefits to the overall Uber marketplace.
Additionally — and here comes my next change— it is important to monitor demand fluctuations and volatility of each additional service versus the core network. It is probably not advantageous to add a service that correlates with core network demand, and it is certainly not advantageous when volatility is correlated. This is where the Uber Data and Algorithm Advantage© (UDAA) comes into play.
If you didn’t skip the earlier section, you’ll have seen a link to a recent paper from economists at the University of Chicago, Oxford, and Uber which discusses the use of Uber’s enormous datasets to estimate short-run demand curves. If this achievement tells us anything, it is that Uber may be very adept at both collecting and using data on everything that goes on in its network with the aim of service optimization. If this sounds interesting, check out this Freakonomics podcast with the UChicago author Steve Levitt.
I’ve added the UDAA to the center of the core cycle simply because it should impact every stage and feed on the continuous “spinning” of the network, pulling in data exhaust. I want to note that I’ve taken some inspiration from Benedict Evans’ (fantastic) write-up on how Amazon works, including this napkin sketch from Jeff Bezos:
By now it should be clear the Uber network model has become quite sophisticated and demand is becoming increasingly diversified. It is now time to add everyone’s favorite disruptive technology: driverless cars.
It has also been proposed that Uber‘s move towards autonomous cars will spell the end of its two-sided network effects. In my humble opinon, what they fail to recognize (or give appropriate weight) is the following:
- The rollout of driverless tech will not be a step function. It will vary across markets, and within them be gradual, and at times halting. Regulation will play a big part in determining the various adoption paths, in addition to infrastructure, local capital markets, culture, and several other factors.
- Driverless tech is most dangerous to Uber when it hits the “off the shelf” stage. This likely will not happen any time soon, giving firms like Uber a head start, particularly as regulators wake up and start mixing things up. Caveat: I do not doubt comma.ai’s talent and ambition.
As mentioned, I see autonomous cars as a hybridization of the network on the supply side which drives down volatility, stabilizes liquidity, and better matches demand and supply. Demand should be expected to be heterogeneous. One portion will be most appropriately and economically supplied by driver-partners, another portion will be better off supplied by autonomous vehicles. The remainder will be indifferent and flexibly supplied, and I suspect it will end up a pretty large portion of Uber’s total demand a few years down the road.
Most interesting is the potential autonomous cars offer rural markets. Providing a reliable supply of rides in these markets is a challenge due to low supply, inconsistent demand, and a geographic dispersion of both. An investment in autonomous vehicles solves this problem (cost issues aside), or at least makes it more manageable, which brings me to my second to last point (almost done!): tie-ins with municipalities.
We are already experiencing the first cooperations with municipalities as we speed towards Travis Kalanick’s vision of transportation as reliable as running water.
Each time Uber ties-in with a municipality, the following happens (demonstrated in the model):
- Demand is generated.
- Geographic reach is extended.
- Credibility to the tie-in model is strengthened.
While I think the model is getting a little overcomplicated at this point, the ciriticality of municipal tie-ins is so important and unique to Uber’s long-term operations that I think it’s important to integrate it.
I should also note that there might be a weak network effect between municipal tie-ins and autonomous vehicles. My expectation is autonomous vehicle roll-outs will in many cases require (or at least benefit from) support from local governments coming in the form of subsidized infrastructure upgrades and regulatory changes. The more municipalities Uber can build relationships with, the easier it will be to roll out autonomous vehicles, at least in the early years. This may be stretching the definition of a network effect, though.
So, where has all this gotten us? This is where I think it’s important to turn to a framework discussed in some form both in Goolsbee’s class and in the book Platform Revolution. The holy grail of any network effects business is to consolidate and hold a market, because in doing so competition can be defended against and value can be sustainably harvested. The framework suggests that a network market’s consolidation potential can be evaluated along the following four factors:
- Economies of Scale
- Demand Homogeneity
- Nature of Network Effects, Spillover
- Switching Costs/Multi-homing
If any of the four are weak, consolidation will require some work. If more than one is weak you’re going to have a bad time.
In ride-sharing the biggest challenge is switching costs/multi-homing. Multi-homing means that a consumer regularly uses more than one substitute service (like Uber and Lyft, or Amazon and Walmart). Amazon is once again an excellent analog.
Note: Demand homogeneity may also be a future issue but consumers’ preferences are in the early stage of their evolution. Something to monitor.
There is little structural reason for users not to price compare Amazon. In fact, the nature of the web browser makes comparison shopping even easier than on mobile phones where Uber operates (except for the driver-partners that have multiple phones set up).
The key strategy Amazon uses to overcome this is what I would call “benefit-bundling” with its Prime subscription. Users pay an annual fee for free shipping and other services like music and video streaming.
The primary advantage of benefit-bundling is to set users up to cognitively “discount” total transaction cost to a level at which the marginal benefit of a possibly cheaper transaction is less than the value of the time required to comparison shop. I suspect this can be explained with Richard Thaler’s (also UChicago, the koolaid tastes good) Mental Accounting concept, though I admittedly haven’t given it extensive thought.
So, even though there are no switching costs for moving between Uber and Lyft, a cognitive barrier could be constructed to simulate switching costs.
This is where Uber Prime comes in.
I debated how to show the effects of an Uber Prime, and decided it was so important that it should encompase the entire network. I see products like Amazon Prime as being similar to a myelin sheath.
A myelin sheath is an insulating layer that wraps around nerves in our brain. The myelin sheath protects nerves while serving the crucial function of speeding up the messages being sent.
Similarly, Amazon Prime results in consumers transacting more, transacting more easily, and being less likely to transact elsewhere.
Uber Prime will do the same, building an artificial moat around their business which both incentivises users to single-home in the Uber ecosystem while simultaneously encouraging them to transact more and more often.
Uber — like many great new businesses — is adept at experimentation. My expectation is they will continue to experiment with both add-on services, pricing strategies (like flat-rate, being tested), and royalty programs (through partners) before rolling out their killer Uber Prime app. At that point, it will be game over, at least until the next disruption sneaks up on Uber where they aren’t looking.
So what does this mean for competitors?
Knowing how the Uber network functions gives competitors a sense of where Uber’s weak points may be, and what their future strategic actions may look like. In terms of specific prescriptions, I’ll leave that for another post.
What about driver-partners?
No one outside of Travis Kalanick’s inner circle can know his true thoughts on this, but my suspicion is if and when Uber consolidates the market with a defensible moat they will seek to maximize profits and equitably share them with their driver-partners. Driver-partner supply will likely be elastic due to multiple substitutes for their time (both paid and unpaid). The growth and maturation of the gig economy will only increase this elasticity.
While many suggest Uber could try to replace the majority of its supply with driverless cars, I see a startup born as an asset lite organization that would be very reluctant to own giant fleets of cars. I think they will find creative ways to keep most of the driverless cars off their balance sheet, requiring an equitable profit sharing both towards driver-partners and whatever system they set up to manage driverless cars (example: municipalities lease driverless cars from Uber’s burgeoning car financing operation, which are then licensed for operation by Uber). Yes, this argument is weak, and I am eager to hear others’ unconventional ideas.
What about for valuations?
I suspect Uber is typically valued using some sort of unit model based on target mature markets with discounts for execution risk adjusted based on the competitive landscape of each country/market. Additionally, I wouldn’t be surprised if the “additional services” component may be broken out for separate valuation based on relevant comparables.
However, if you think about the additional services component primarily as a stabilizing force in the network and not necessarily as a profit driver there may be other ways to integrate it into the unit model. Along the same lines, I would be surprised if a valuation of Amazon would involve breaking out Amazon Prime Video to value it using Netflix as a comparable (yes, I know Prime Video and Uber Eats are economically not comparable). I’ll have to think about this more, and am interested in ideas.
What now?
This is the first time I’ve put my ideas out on the internet beyond tweets and LinkedIn posts. I welcome both feedback and pushback on my writing and ideas.
Thank you to Graham Zimmerman for editing support.
If and when I get interesting questions I will add my thoughts on them here:
In writing one of the answers below I came across this excellent article, which is a nice follow-up to my piece:
This is also fantastic, which I had already read. It influences my thinking.
So what about all the flak uber and other such platforms are getting for paying some drivers below their costs? How does all that fit into this?
In the short run (weeks) I can imagine drivers operating below cost, but I have a hard time believing they would (as a whole) do so for very long. Even the Uber financing allows drivers to cancel with one month notice at no penalty, or at least this is their perception based on conversations I’ve had with them.
So, it appears it is more complicated than just a “subsidy”, involving driver incentive programs and charging passengers below the stated rate. Either way, the fact is that they are definitely breaching minimum WTD for some drivers as they have lowered rates. That is how a supply curve works.
As for how it fits in:
- Lyft (for example) needs to grow its network to get all the effects running sustainably.
- The biggest and easiest dial they have is price (especially considering capital market conditions – so much slosh), so they turn it. It’s like adding NOS to the engine (TOO FAST TOO FURIOUS).
- Uber both needs to build a credible reputation as a deterrence (like Amazon) and because services are not very differentiated yet (no Uber Prime) if they don’t match price they will steadily lose riders (network density/availability is a differentiator but it’s a leaky seal)
Question is: can Uber kill Lyft (or convince them to be happy in niches at reasonable prices) before a) drivers find a better substitute than operating at minimum WTD or b) regulators step in on a broad basis?
Learning from failure in China… can they compete where they are not the fiirst to enter market and have to fight to build market share?
I think Uber’s foray into China was far from a failure and this Bloomberg article sums up the reasons up pretty well.
“Perhaps even more valuable will be the 20 percent stake in Didi that Uber is getting out of the transaction, which will value Didi at roughly $36 billion. In time, that stake could prove to be one of the best financial windfalls in corporate history. In short, Uber lost a couple billion dollars in China but won a $7 billion stake in a company likely to be worth much more in the future, especially now that Didi doesn’t have to compete with Uber.
“Uber can also lick its wounds with the knowledge that the company made its mark in China, even if it didn’t win the battle with Didi. The Chinese company started out essentially as a dispatch service for conventional taxis. Uber’s entry helped push Didi into on-demand rides with regular people doing duty as semi-professional drivers, plus carpooling services and buses. Competition with Uber made Didi better, and it helped reshape the Chinese transportation market.”
Competing in China as a non-chinese firm can be incredible challenging, especially in winner take all markets. Travis Kalanick said it himself:
“We were a young American business entering a country where most U.S. Internet companies had failed to crack the code.”
It was not for a lack of effort, foresight, or time. Uber soft launched in Shanghai in 2013 when its biggest competition had not yet pivoted to ride sharing, and were still small startups. This FT piece published prior to the Didi deal provides a lot of history and perspective into Uber’s evolution.
Now, in doing a little research to write this answer, I came across some new information.
It appears that while Uber received a 20% stake in Didi, there is a dual share structure to the deal, with the 20% describing the economic rights (17.7% for Uber, 2.3% for Uber investor Baidu), not control rights. Uber seems to only have about 6% control rights in the combined company. This means Didi has (no surprise) a complex ownership structure with a wedge between control and economic rights. More on Didi’s complex structure here:
While the structure of the deal does dampen my thoughts on the deal being a success, I still think what Uber achieved should been seen as a positive outcome considering the challenges of operating in China. I also think it should give competitors in other markets pause. Uber has demonstrated an ability to intimately understand local markets and a willingness fight tooth and nail for market share, which will be beneficial as it wages battles in other markets. Nevertheless, success on any level is certainly not a given.
How could they use consumer surplus research to negotiate with regulators about a future with driverless option where WTD is not influencing the decision?
This question is referring to this research, cited earlier.
TL/DR Uber can use its datasets to estimate short-run demand curves, which in turn can tell you what consumer (and producer) surplus is.
I’ll bet the paper focuses on consumer surplus because it is the positive story right now. Low prices and inelastic demand mean consumers capture a lot of the surplus. The datasets can probably also be used to estimate the short-run supply curve. My guess is drivers likely have more elastic demand than consumers (more substitutes) and so are capturing very little surplus right now.
To the question, my thought is that consumer surplus being so high right now means consumers love the product, which incentivizes municipalities to cooperate. Uber can cite this data in negotiations, giving backing to a feeling. What local legislator doesn’t want to get in on making their constituents happy? Uber helps them transfer tax dollars into consumer surplus, while getting a cut for being the middle man.
On the flip side, knowing how much (or little) producer surplus there is should help Uber when dealing with regulators who feel compelled to respond to driver-partner concerns. Regulators must of course be careful. Data can be massaged to show anything.
I was talking with [name omitted] the other day about whether it’s clear Uber has any special advantages once the world moves to autonomous vehicles. We weren’t so sure it would. What do you think?
This question required a long answer, so I have posted it separately. You can find the answer here:
I recently finished an electives course on the microeconomics of platforms called Platform Competition taught by Austan Goolsbee (@Austan_Goolsbee) as part of my MBA in the University of Chicago Booth School of Business’s Executive Program. The inspiration for this article was the prep work I did as part of the class.
Hacker Noon is how hackers start their afternoons. We’re a part of the @AMI family. We are now accepting submissions and happy to discuss advertising & sponsorship opportunities.
If you enjoyed this story, we recommend reading our latest tech stories and trending tech stories. Until next time, don’t take the realities of the world for granted!