Lost rides: the nature of shared mobility competition

by Evan Fields, head of data science at Zoba. Evan holds a PhD in Operations Research from MIT, believes aggressively in cities, bakes enthusiastically, and can be found on Twitter at @evanjfields.

Zoba provides demand forecasting and optimization tools to shared mobility companies, from micromobility to car shares and beyond.

Over the last several years, new shared mobility — especially micromobility — companies have blossomed all over the world. Many mobility service operators have found themselves sharing a city with other companies with near-identical vehicles and business models. Understandably, overlapping shared mobility turfs have caused many operators to worry about competition. The COVID pandemic’s tectonic effects on the entire mobility industry have accentuated these fears: many operators have paused operations or withdrawn from markets entirely, and it remains unclear what the mobility landscape will look like once the dust settles. At Zoba, we think the shared mobility industry’s best days are ahead, but we absolutely understand these justifiable concerns about the competitive landscape and thought we’d write a little about how we think about competition in the context of tactical mobility decisions.

Competition for mobility happens on varying timescales. In the short term, there is competition for trips: should I take my bike or the train to work? How will I get home tomorrow evening? Over longer timescales, the primary competition is for habits and mindshare: where will I live, and what travel mode do I expect to be my main way of getting to work? Over long timescales, non-competitive dynamics also come into play. For example, the second order effects of users taking a trip with a given travel mode may be positive for all competing operators within that mode. Consider how biking causes support for bike infrastructure which causes safer bike routes which cause biking. A bikeshare operator benefits from this feedback loop even when a potential user decides to take a trip via a competing bike service. Without discounting the incredible importance of these long-term competitive and collaborative dynamics, in this brief post I’ll focus exclusively on some short-term insights gained from Zoba’s core business over the last couple years. Zoba helps mobility operators with short term tactical decisions like where to place vehicles and moderate term strategic decisions such as which areas of a city to expand into, and this has provided us with a unique vantage of tactical competition in shared mobility.

When evaluating competition, most mobility operators instinctively think of other companies with similar vehicle types (which typically also have similar price structures and target demographics). For example, scooter companies consider other scooter companies their most important competitors. While intuitively appealing, the notion of “like competes with like” overlooks an important dynamic of shared mobility: every ride is a choice, and therefore competition is cross-modal. For example, the main competition for scooter ridership often comes from non-scooter modes a scooter operator may be overlooking.

By definition, the user of a shared mobility service does not own the vehicle used for a trip. Most (typically all) of the cost of each trip — in time and money — is paid per trip. Therefore, every time a potential user is deciding if and how to take a trip, the user is free to choose the mode and provider that best serves this single potential trip; the shared mobility user is not “locked in” to a given travel mode by previous travel choices. As a counterpoint, consider private car ownership. A car owner makes a large upfront purchase of a car, but thereafter enjoys relatively low per-trip costs and is incentivized to use the car so as to amortize the upfront cost. Even cheaper travel modes like personally owned bikes create some lock-in. For example, when I ride my bike to the Zoba office in the morning, I’m super likely to ride it home in the evening because my bike is in exactly the right place to start that evening trip.

A shared mobility user chooses between travel modes without forfeiting anything, and there are inevitably lots of modes to choose from — including not taking a trip at all. A beachgoer looking for a scooter joyride might decide not to take a ride if no suitably close and cheap scooter can be found. Someone running a quick errand around the block might walk instead of using a bikeshare. Longer trips can be well served by public transit, rideshare, car sharing, etc. With users choosing ad-hoc between operators and travel modes, the right question about competition is not “where are the other operators just like me?” but rather “where am I losing potential trips and users?” These questions often have quite different answers; “competition” is any alternative that tempts users away from a mobility service! Not only do users choose between myriad travel modes, but mobility services in a given mode may not directly compete if they target different people, inspire different brand loyalty, have different service regions, and so forth.

This last point is worth expanding on. No two mobility services are exactly alike, and differences between mobility services compound to give each service a unique competitive landscape. As an illustration, consider an imaginary city with two kick scooter companies: LuxeCo provides expensive smooth-riding scooters, and BargainCo provides cheaper but shakier vehicles. Though LuxeCo and BargainCo appear to be direct competitors — they’re both scooter companies in the same city — their effective competition is quite different. LuxeCo competes more with rideshare; BargainCo competes with bikeshare and walking. They also have different demand profiles: LuxeCo has the most demand in touristy areas, BargainCo near college campuses. Further, the fraction of LuxeCo’s total competition that comes from BargainCo changes constantly. LuxeCo’s main competition comes from BigRideHail on rainy days but from BargainCo on beautiful days. Competition for commuter trips differs from competition for leisure travelers. Combined, these factors mean that LuxeCo and BargainCo experience markedly different patterns of competition across the city. Whether the two operators even compete meaningfully with one another is heavily dependent on time and geography. The LuxeCo/BargainCo split reveals another important consideration: competition is typically asymmetric. In an area where most users prefer LuxeCo’s offering, for example, LuxeCo will experience little competition from deployed BargainCo vehicles, whereas BargainCo will experience significant competition from LuxeCo’s vehicles.

In short, every operator has a unique pattern of times and places where they have to fight the hardest for rides. From a tactical perspective, competition in a given area matters exactly because it affects when a vehicle deployed to that area (or ridden to that area by a user) is likely to get a ride. At Zoba, we weave this key competition insight directly into our demand and optimization models. These models are designed to help our customers make tactical decisions such as where to place vehicles or how to price them. In that context, competition serves as a mediator in the supply-demand interaction.

Recall from our previous post about demand that at a given place and time, there’s a fixed total demand that a mobility service could serve, and this demand does not depend on supply. However, the mobility service must vie with any competitors present at that place and time for that demand. The amount of competition determines how much demand the mobility service can capture by adding more supply:

• In high competition areas, each additional vehicle placed can grab some demand that would otherwise have been served by competitors. Therefore the fraction of the total demand that the mobility service can serve rises as supply rises.
• In low competition areas, there are few or no competitors to steal demand from. So each marginal vehicle an operator deploys has little to no effect on the total demand servable by that operator.

Here’s a stylized example of this phenomenon. Imagine a single location where the demand is three rides wanted over the next hour. For this example, we assume competitor vehicles are exactly exchangeable with the operator’s vehicles and therefore the latent demand is evenly split between all vehicles present. We consider how the share of the total latent demand the operator’s vehicles experience depends on how many vehicles they deploy and the number of competitor vehicles present. If there are no competing vehicles (shown in blue below), then as soon as the operator deploys any vehicles at all, their vehicles experience the entire demand. (Note that with fewer than three vehicles deployed, the operator can’t capture all three desired rides.) On the other hand, if there are three competing vehicles (shown in red), then the operator’s deployed vehicles must “share” the demand with the competitors. By deploying more vehicles, the operator can grab a greater fraction of the total demand. For example, if the operator deploys only one vehicle, that vehicle is one fourth of the total vehicles and thus gets to “see” one fourth of the total demand. If the operator deploys three vehicles, then their vehicles are half the total vehicles and experience half the total demand.

Of course, a mobility operator can only make tactical decisions about their own vehicles — they can’t decide where to put their competitors’ supply. Therefore, for optimal tactical decisions, the most important feature of competition is how it affects the supply-demand relationship. With this information, one can make even better tactical decisions than with forecasted demand alone — and no other competition information is as fundamental for making optimal tactical decisions.

Zoba’s models estimate the curvature of the supply-demand relationship at each location — and therefore the effective competition — directly from historical trip data. We consider each location at each time to have a supply response function, just as shown above: the supply response function maps the number of vehicles deployed to the fraction of the total demand those vehicles experience. The curvature of a time-and-location-dependent supply response function depends on the competition at that time and location, and we estimate the curvature using historical data. In particular, natural variation in an operator’s number of deployed vehicles in the area allows us to estimate the competition by seeing how demand reacts to varying supply conditions.

No third party competition data is needed for our models. We do not use raw competitor data in our optimizations, though this data is possible to scrape and there are even third-party providers that package it in some markets. Legal or ethical questions aside, we avoid this data because it would not actually be very useful to us. Knowing there are a certain number of competitive vehicles in an area provides little actionable information for operations as it’s not guaranteed that those vehicles are the primary reason for losing rides in the area. Real-time competitor data may provide some high-level insight into market conditions, but it is a suboptimal way of incorporating competition dynamics into an operational optimization. In contrast, we strongly believe estimating competition from historical data is the most principled way to estimate competition. By building our competition estimates on the empirical shape of supply-demand interactions, we remain agnostic about where competition might be, what travel modes create that competition, and how much total competition there is. An operator’s historical data is used to construct an understanding of competition specialized to how their service interacts with the market in practice.

Understanding the unique competition profile for our clients helps Zoba make optimal tactical and strategic recommendations which integrate both demand and competition (as well as other spatial and temporal factors, projected vehicle movements, etc.). Our vehicle placement optimizations, for example, consider each location’s demand, competition, and where vehicles deployed to that location are likely to end up. Perhaps counterintuitively, it can be optimal to deploy some vehicles to high-competition areas where these vehicles can steal rides from competitors, especially if there’s a lot of demand. The greatest improvement to mobility operations comes when we discover heretofore overlooked gems: high-demand low-competition areas — a huge opportunity made possible only by careful demand and competition modeling.

Zoba is developing the next generation of spatial analytics in Boston. If you are interested in spatial data, urban tech, or mobility, reach out at zoba.com/careers.