05 September, 2015

Uber (Part 2a) - Intuition building for "multi-sided markets" with "network effects"

If you're new to our Uber conversation, it begins here.  This post is out of order.

On Sept 3rd, Wired covered NYC's launch of a "Technical Advisory Group" tasked with developing recommendations on how regulations should evolve to consider on-demand ride services.  Uber reportedly has a seat at that table, as does Venture Capitalist Fred Wilson and academics from NYU and Columbia, among others.

The TAG's recommendations will serve as input to new regulations in NYC, which are likely to cascade influence to other global cities facing the same situation.

Below, I offer a framework from a 2006 Harvard Business Review article to explore why I believe Uber is as much a "regulatory arbitrage" story as it is a "sharing economy" story.  Among my half-dozen recommendations made (so far) has been that Uber should consider pushing to close the regulatory loophole behind them while they're big and competitors aren't.

This is because I believe Uber is in a market that may be fundamentally unwinnable, at least in the manner/value in which Uber appears to be trying to win it today.  In brief, the on-demand ride market has significant network effects on both sides of the transaction, but it isn't really a multi-sided market.

To help readers understand my broader framework for Uber's strategic imperatives (now re-labeled as Part 2b), I'll attempt to make my intuition more explicit.  Sometimes intuition is helpful - if the problem is framed correctly.  Other times, the right answer is counter-intuitive.

One benefit from being this explicit (pedantic?) is because it's not clear that I know enough about the topic to see what I'm missing.  Perhaps it's me, not the (private) market valuing Uber, that's mistaken.  My background in economic analysis and algorithms is stronger than many people's, but I'm not an expert in any particular area.*

Here's Uber investor and board member Bill Gurley explaining Uber's pricing mechanism and business model in early 2014, if you want a quick primer on how Uber works.  My conversation below focuses entirely on the peer-to-peer model of UberX.





Original problem statement of on-demand transportation:  


I find this a helpful way to think about on-demand networks:

  1. The value of an on-demand network (i.e. not scheduled) to the consumer is dependent on the number of unused nodes available for rides
  2. Without centralized employment, these unused nodes only get paid when they become used
  3. Legacy taxi schemes solve this by paying nodes "above market" rates when they become used, then insert various distortions into the free market (medallion caps, unified pricing structures, etc) to ensure drivers end up paid for the time when they weren't used

Without the free market distortions introduced in #3, the driver market could "race to the bottom" with low-cost, unsafe rides that fluctuated between boom-and-bust cycles because of information asymmetry (this is in fact the history of taxi regulation)**

The medallion system is sub-optimal, though.  The gains allocated for transfer to the unused nodes are mostly captured by the medallion holders; high medallion values represent the annual cash flows that accrue to the medallion instead of the driver.  Vehicle cost structures are largely determined by setting a regulatory floor.

Net effect is drivers still end up paid the equivalent of minimum wage, albeit within that distorted structure.  And during peak times, customers can't get a cab because there aren't enough - if there were enough during peak times, there would be too many during off-peak.



An intuition-based Uber problem statement: Uber's peer-to-peer "innovation" breaks #3 without appearing to provide a structure to replace it


Uber has invested an incredible amount of money to encourage the race to the bottom for drivers (benefiting consumers), without a visible structure to protect its own margins.  My larger framework in Part 2b explores the economic implications of executing their plan as I currently understand it, particularly how Uber's cash flow needed to execute it will interact with their $50b valuation.

Even if Uber's long-term plan is to use its knowledge of demand patterns to reduce the importance of #1 and #2, I believe Uber will still need to build self-protection through shaping beneficial re-regulation, and/or evolve toward a different operating model.  Even then Uber may struggle to maintain a substantial portion of its current valuation.

If this thesis is true, all the technical innovations and legal battles Uber is fighting today are arguably the equivalent of rearranging the deck chairs on the Titanic (a favorite analogy of consultants, not always used correctly).

Expert discussion framework: Strategies for two-sided markets


Uber's go-big-and-fast-and-destroy-the-economics behavior often makes sense for a two-sided market.  Sometimes it doesn't.  In Oct 2006 the Harvard Business Review published an article on this topic by Harvard entrepreneurship Professor Tom Eisenmann et al: "Strategies for two-sided markets."

We'll use Prof. Eisenmann's framework and words to illustrate how Uber's structure and behavior seems to defy what I see as the conventional (expert) prescription for the type of challenges Uber faces in the on-demand ride market.
Note: Any errors of commission or omission here are entirely my own; I have not contacted Prof. Eisenmann about his article or this topic
As we progress, you can compare elements of the on-demand ride market with what you know about other successful multi-sided market businesses to see how they're very different from Uber.  For a more recent analysis of peer-to-peer markets specifically, see Jon Levin et al here.

Even though Uber's press coverage seems to place it under the broad umbrella of "multi-sided markets" subject to "network effects,"  strategies and tactics that lead to winning (differently structured) two-sided markets may not be appropriate for Uber's highly commoditized on-demand ride market.

For example, replacement markets in successful "freemium" style SaaS business models (e.g. Android, Zenefits, etc) have switching costs for users and often more than one source of revenue; other peer-to-peer markets such as AirBnB have much more supply differentiation (location, amenities, pricing set by provider, etc).

An AirBnB property doesn't experience variable costs from cleaning/maid services when unused, whereas a driver has opportunity cost.  Also, in some air BnB properties, people are renting out an empty room - the fixed costs of providing that room are rolled in to their existing living expenses, although that is not true for an investment property.

Challenge 1: Pricing structure depends on whether cross-side and same-side subsidies and network effects are positive or negative


Typically, two-sided networks have a "subsidy" side that is highly valued by the "money" side.  As the platform attracts more "subsidy" side users, the platform becomes more valuable for the "money" side.  Note that it's possible for both sides to pay the platform (as in Android), it's more a question of which side is paying "more" and "less."

When there are positive cross-side network effects, this works a bit like a turbocharger.  The platform provider will initially pay the subsidy side out of pocket - and sometimes even the money side (as in new platform video game development, or signing up new Uber drivers).  Past critical mass, the "money" side pays enough extra that their cash flow funds both the subsidy and the platform.

Platforms with pricing power on both sides determine three interdependent things by choosing two (the third is math based on other two):

  • How much "subsidy" is given to that side
  • How much premium is paid by the "money" side
  • How much margin the platform keeps for matching the parties.  Note that the subsidy/premium decision can be highly segmented, and can change over time.  
    • See for example this 2012 paper on optimal pricing and sequencing heuristics by monopolistic platforms; in some circumstances it's quite challenging to fully "optimize" this decision but heuristics can get (provably) close to varying degrees

Pricing decisions are complicated by same-side network effects.  In some cases these are positive (e.g. more users makes the platform more attractive for drivers, and vice versa).

In other cases same-side network effects turn negative (e.g. too many users creates "surge pricing", too many drivers increases unpaid dead time and reduces earnings).


Eisenmann describes six factors that impact pricing structure for multi-sided platforms.  Uber's design doesn't seem consistent with the demands of #1, #2, #4, and #5:

For the purposes of the below discussion, I'm presuming passengers are on the "subsidy" side, and drivers are on the opposite ("money") side.  Uber is in the middle as the "platform."

In my broader Uber discussion, I discuss what types of changes could put both drivers and riders on the subsidy side, with a fourth party stepping in as the money side.  If such an option exists, I haven't seen it yet.

  1. Ability to capture cross-side network effects.
    • Eisenmann writes "Your giveaway will be wasted if your network's subsidy side can transact with a rival platform's money side."
    • With Uber, because the money side drivers can "wait" on multiple platforms due to their status as independent contractors, and users can "hail" on multiple platforms to conduct a quick price check, there is a tremendous amount of cross-platform leakage that will prevent long-run price-making by the platform

  2. User sensitivity to price
    • Eisenmann writes "Generally it makes sense to subsidize the network's more price-sensitive side and to charge the side that increases its demand more strongly in response to the other side's growth."
    • Uber has proven that riders are very price sensitive, but because rides (unlike software) have marginal and opportunity costs, so are the drivers.  As long as driver networks are essentially "shared" among platforms as independent contractors, there will be a strong incentive for rival networks to compete - they'll transfer platform revenue to drivers once driver revenue has reached the equilibrium floor of marginal cost pricing

  3. User sensitivity to quality
    • Eisenmann writes "High sensitivity to quality also marks the side you should subsidize.  This pricing prescription can be counterintuitive: rather than charge the side that strongly demands quality, you charge the side that must supply quality."
    • This doesn't seem to help us understand UberX; both sides seem to have a minimal expectation of quality (safety, getting paid) that's better seen as table stakes than elastic.

  4. Output costs
    • Eisenmann writes "Pricing decisions are more straightforward when each new subsidy-side user costs the platform provider essentially nothing.... when a giveaway has appreciable unit costs, as with tangible goods [rides], platform providers must be more careful.  If a strong willingness to pay does not materialize on the money side, a giveaway strategy with high variable costs can quickly rack up large losses."
    • Uber began their ramp up by subsidizing both sides - and in the process eroded drivers' ability to subsidize the riders.  Unless Uber can find a new side to this network, it will have to subsidize the riders themselves.  
      • In the long-term, possibly by replacing drivers with autonomous vehicles.  
      • In the short-term, there are few options that don't first begin with making at least some of the drivers employees (e.g. to use detailed knowledge of ride demand patterns to preposition cabs in order to reduce the number of empty nodes required to operate the on-demand network at a given service level)

  5. Same-side network effects
    • Eisenmann writes "Suprisingly, sometimes it makes sense to deliberately exclude some users from the network.... platform managers must assess the possibility of negative same-side network effects, which can be strong."
    • Uber's driver count is uncapped and often "wait" on multiple networks.  The ability for drivers to jump onto the UberX network erodes earnings for other drivers, creating negative network effects beyond a certain point.  
      • It's probably not a good idea for Uber to exclude drivers from its own network unless it can first ensure the drivers it lets on aren't still "driving" on multiple networks at the same time
      • There may be incentive mechanisms (decline rates, ratio of "regular" driving to permit "surge" driving) that could at least partly accomplish this goal without hiring drivers full-time

  6. Users' brand value
    • Eisenmann writes "The participation of "marquee users" can be especially important for attracting participants to the other side of the network... a platform provider can accelerate its growth if it can secure the exclusive participation of marquee users in the form of a commitment from them not to join rival platforms."
    • Factor doesn't seem relevant to Uber, outside of regulatory agreements (eg only Uber at the airport) and perhaps one-off event partnerships


Bottom line for Challenge 1: Uber's pricing and design doesn't seem consistent with #1, #2, #4, and #5 above.  In part, this is because hewing to these principles would be inconsistent with Uber's decision to use independent contractors as a second side to the platform.

In the historic case of taxis, even though drivers are often "independent" that's a legal fiction in lieu of supervision of unmanaged workers that should be considered separate from how the market operates.  In the case of Uber, adopting that legal mindset doesn't change the underlying economic forces of what they're trying to accomplish.

While the use of contractors may have saved Uber some money in the short run, and made it faster/easier to scale, it makes it challenging to design an optimal network-based pricing structure.  

Perhaps the control/coordination requirements of an efficient on-demand ride system means it's not an appropriate candidate for a true multi-sided market playbook.

Examples of similar commoditized matching markets exist - Home Depot's Installation Services and UHaul's Moving Helpers.  Both of these are revenue generating in the sense that they likely don't cost the platform more than the revenue they take in, but both of these exist to sell the underlying service - paint, doors, and windows at Home Depot, and rental trucks at UHaul.

I doubt if HomeDepot cares much that contractors are also listed at Lowe's, or even that repeat customers can contract outside the platform.  What HomeDepot wants is to ensure that anyone who wants to buy a window from Home Depot can do it, even if they need help installing it.

Challenge 2: Winner-take-all (WTA) dynamics.  The only reason to bet-the-company on winner-takes-all is (a) if the market is destined to be served by a single platform and (b) the risk-adjusted cost of the fight is less than the monopoly profits accrued by winning it

2.a Eisenmann's three factors argue against a single platform winner in the on-demand ride market


2.a.1  Your market might be WTA if... Multi-homing costs are high for at least one user side 


  • Eisenmann writes: "Homing" costs comprise all the expenses network users incur - including adoption, operation, and the oppotunity cost of time - in order to establish and maintain platform affiliation... When multi-homing costs are high, users need a good reason to affiliate with multiple platforms
  • Uber homing costs for drivers and riders are essentially zero.  Homing costs could be increased by forcing drivers to pick a side, or by charging customers an "access fee" to the platform.  It's possible this could be done while maintaining contractors, but more likely means employing at least some of the drivers.  This factor argues against a single platform winner under the current peer-to-peer industry structure


2.a.2  Your market might be WTA if... Network effects are positive and strong - at least for the users on the side of the network with high homing costs

  • Eisenmann writes: When cross-side network effects are positive and strong, those network users will tend to converge on one platform.  A small-scale platform will be of little interest to users unless it is the only way to reach certain users on the other side.  The odds of a single platform prevailing also increase when same-side network effects are positive: for example, when users of a software program need to share files with one another
  • Uber network effects are in conflict: Cross-side network effects are strong to a certain limit, then decrease.  Beyond this point, each side's cross-network preferences are in opposite preference to same-side network effects: riders benefit from more drivers but few riders; drivers benefit from more riders and few drivers.  This factor argues against a single platform winner

2.a.3  
Your market might be WTA if... Neither side's users have a strong preference for special features

  • Eisenmann writes: If certain users have unique needs, then smaller, differentiated platforms can focus on those needs and carve out niches in a larger rival's shadow
  • Uber's growth depends on special features:  There is a limit to the size of point-to-point on demand rides that is much smaller than Uber needs to support its valuation.  Additional demand at lower price points will be drawn from carpool-like and bus-like services, which Uber has already launched.  It seems like vehicle cost structure will be different enough in each of these markets to make them specialized.  This factor seems unsupportive of a single platform winner.


None of the above three factors in 2.a indicate the on-demand taxi market is winner-takes-all


2.b  Eisenmann says that only winnable battles should be fought, and even then only fight if your win would enable monopolistic pricing that covers the cost of winning


Eisenmann says "to fight successfully, you will need, at a minimum, cost or differentiation advantages."  

With Uber relying on third-party contractors, there is no clear cost advantage.  Any differentiation feature that appears successful can be easily copied by a competitor - at least at this stage of the game.  Uber arguably has an edge in brand recognition, but it's not clear that this will translate into any pricing power (ask airlines about this).

An exclusive tie-up and volume advantage once the industry shifts to electric cars, say with Tesla, could help cost differentiation.  But that's still several years out - and the factors in 2.a still suggest that a single winner won't dominate even at that stage of the game.

If the battle were winnable (which 2.a suggests its not), and if Uber had the minimum 2.b requirements to fight successfully (which it doesn't, at least today), Uber appears to have at least two of the three factors required to win the battle on its side:

2.b.1  Pre-existing relationships with prospective users would help win a battle
  • Eisenmann writes: Often in related businesses
  • Uber doesn't have pre-existing relationships with drivers or passengers, which in part is what drives the high startup costs in a new city.  While Uber recently announced a partnership with Hilton, that is more about mutual reinforcement of brand image rather than a concrete reduction of Customer Acquisition Costs, which is what this element focuses on

2.b.2  High expectations generate momentum, which helps win a battle
  • Eisenmann writes:  A reputation for past prowess helps a great deal
  • Uber has high expectations and momentum.  No doubt about it - fundraising, press... CEO Kalanick will actually be among of Stephen Colbert's first guests (the second night, along with Elon Musk) when the new Late Show premiers next week

2.b.3 Deep pockets matter when winning battles
  • Eisenmann writes:  Eisemann writes nothing, possibly because it's self evident that wars of attrition are very expensive
  • Uber has deep pockets... for now.  One of the biggest unanswered questions about Uber is how optional their $500m/month burn rate is.  If Uber ran into a cash crunch, are the "mature" cities cash flow positive enough on their own that Uber would simply scale back their expansion rate?  Or would Uber be forced to re-trench in a smaller number of cities until the funding started to flow again?


In summary of Challenge 2, Uber's prize may not be worth the chase: 

  • Eisenmann writes:  
    • First mover advantages can also be significant, but they are not always decisive... 
      • When the market evolves slowly... late movers, may, for example, avoid the pioneer's positioning errors, be better placed to incorporate the latest technology into product designs, or be able to reverse engineer pioneers' products and beat them on cost.
      • Google, which lagged Web-search pioneers by several years, avoided portals' clutter in favor of a simple, fast-loading home page.  It also copied and then improved on Overture's paid-listing model for generating revenue from searches
    • ...First and late movers alike will feel strong pressure to amass users as quickly as possible
      • In most cases, this urgency is appropriate
      • Racing to acquire can be a mistake under two circumstances
        • Business isn't readily scalable
        • Platform-mediated networks are prone to boom/bust valuation cycles... managers need to be sure that funding will be forthcoming should capital-market sentiment turn negative

Uber is spending a lot of money to (a) win something that doesn't appear to be winnable, particularly at this stage of technological evolution; and (b) depends on significant external funding - which can be challenging for any platform business to weather but particularly for one with such a high existing valuation.

Uber's scaling seems to involve a lot of payments to drivers, passengers, regulators, and lawyers that to date have benefited the whole industry, not just itself.

The way Uber, spending so much money to win what seems to be an unwinnable market, makes sense is if you believe building a global network of marginal-cost-priced rides is a means to some other business end.  My framework in Part 2b goes into these possibilities in more detail.

Or you might believe that at some point, the size of Uber's ride network will let them use fewer empty cars to serve the same rider base - giving them a cost advantage that is very expensive to replicate.  Particularly if then augmented with exclusive access to a purpose-built (autonomous?) electric vehicle.

My intuition says that scale cost advantage is possible for on-demand rides, but likely not worth what Uber is paying (and will need to continue to pay in order to accelerate the process).  I'll develop the economics of this argument further in the upcoming Uber (Part 4c).

It would be challenging to evaluate the cost-reduction potential of wait-time reduction without a deep dive on Uber ride-level data across multiple cities.  That is outside of my current scope

Challenge 3:  The threat of envelopment means your "win" might not stay "won"


Eisenmann writes:
You can do a great job addressing pricing and winner-take-all challenges and establish a successful new platform yet still face great danger... Your platform may be "enveloped" by an adjacent platform...
... Platforms frequently have overlapping user bases.  Leveraging these shared relationships can make it easy and attractive for one platform provider to swallow the network of another.
The real damage comes when your new rival offers your platform's functionality as part of a multiplatform bundle.  Such bundling hurts the stand-alone platform provider when its money side perceives that a rival's bundle delivers more functionality at a lower total price.  The stand-alone platform provider cannot respond to this value proposition because it cannot afford to cut the price on its money side and it cannot assemble a comparable bundle
Networked markets - especially those in which technology is evolving rapidly - are rich with envelopment opportunities that can blur market boundaries.  This burring is called "convergence"... in many cases a stand-alone business facing envelopment has little choice but to sell out to the attacker or exit the field

Uber's threat of envelopment seems low at the moment; if anything Uber are the ones seeking to blur the boundaries around other ride services - personally owned cars, carpooling, bussing, and potentially even mass transit.

In the interest of completeness, I offer Eisenmann's prescription for companies facing envelopment:

3.1 - Change business models.  Switch money sides, add new sides.
3.2 - Find a bigger brother.  Ally with groups whose interests align with yours, including with whom you share enemies
3.3 - Sue.  "Anti-trust law for two-sided networks is still in dispute.  Antitrust law was conceived to constrain the behavior of traditional manufactuing firms and does not fully reflect the economic imperatives of platform-mediated networks.  For this reason, dominant platform proivders that offer bundles or pursue penetration pricing run the risk of being charged with illegal tying or predation"

TL;DR Uber is spending lots of money to win something that may not be winnable under current conditions.  This raises questions about the strategic imperatives of their valuation


The on-demand ride market doesn't actually seem like a two-sided market when you look at it closely.  Undifferentiated drivers as contractors seems like a (useful) legal fiction.  Treating them as a "money-side" to the market seems to undermine some of the operating aspects of the on-demand business.

It would be an interesting exercise to determine if this is rational - in other words, if the extra cost of drivers as employees is more than the extra cost savings that would accrue to Uber.

There's an old joke: Two economists walk past a $20 bill laying on the ground, one asks the other why he didn't pick it up.  The second one says "If it was real, someone would already have grabbed it."

Most likely, as long as drivers are willing to work for very low wages (which is where this structure heads in equilibrium if it's not there already), Uber's incentive is to keep ride prices low to maximize the total dollar value of rides and hence their platform fees.

Once that equilibrium is reached, though, perhaps there would be room to use predictive analytics to take some empty cabs off the street.  Could drivers spend less time waiting, and make the same total wages paid hourly, without raising customer pricing?  This would also limit competitors' ability to piggyback on the network.

Said differently, could Uber make better use of a limited number of "medallion 2.0s" than their competitors if the industry were re-regulated?

Depending on how many empty cars could be pulled from the street without changing service levels, there might be enough extra money to raise Uber's margins in a way their competitors would be challenged to follow.

Would employing their drivers also enable Uber to implement a two-tier pricing structure - fixed fee for access to the Uber network at marginal cost pricing, or no-fee access to higher rate rides?  Uber could probably do this without employing the drivers.

Is two-tier pricing possible under competitive conditions?  Would that reach as many users as Uber needs?



Thanks for reading,
Greg




*I'm not considered an expert on any particular topic, but I have substantial first-hand experience working for world-class experts in economics, computer science, and operations research.  Which is why the Uber topic puzzles me so much.  More about me here and here

**Many operational strategies derive value from the optionality of unused capacity.  In some circumstances, the optimal utilization rate is as low as 60-80%.  In other areas, it's upwards of 95%

No comments:

Post a Comment

Focused, topical critiques are welcome though my responses may be limited. Off-topic comments, trolling, spam, or anything not pulling its weight will be deleted.