My girlfriend lives a little less than 3 miles northeast of me in Washington, DC; we're on different Metro lines, and no bus runs between our neighborhoods. As recently as a few months ago, visiting each other regularly was an expensive proposition, with an Uber or Lyft costing $8 to $10 each way. But instead, I've been paying less than $5 to make the trip.
That's thanks to Split, a DC-based ride-sharing iPhone app which launched in May (an Android version is coming soon). It works much like Uber or Lyft, but users specify both their location and destination when they request a ride. That lets the company's software match up riders with similar trajectories, letting drivers pick up multiple passengers and collect multiple fares at once. Uber and Lyft have experimented with this carpooling approach — with UberPool and Lyft Line — but both are costlier than Split. In DC, UberPool has a $5 floor for fares; Lyft Line isn't here yet, but it also has a $5 floor in San Francisco. Split's floor, by contrast, is $2. Every ride costs that, plus $1 per mile. If you're traveling 3 miles or less, it's always going to be cheaper than the cheapest competitor.
When I first realized that, I felt less relieved than guilty. Presumably there's no way to cut costs that much without compromising driver compensation, right? But when I've asked Split drivers about why they've joined, the response usually touches on how much better the pay is than for Uber and Lyft. For one thing, Split is at the moment guaranteeing people an hourly wage while they're on duty, regardless of how many rides they get. That, multiple drivers told me, provided a kind of security that the other services couldn't match.
So my guilt gave way to skepticism: If Split is cheaper for consumers, and better for drivers, then obviously it's going to go bankrupt before long. Right? After visiting Split's offices at the startup incubator 1776, I left more convinced that they could actually make this work. The company is insistent that it's putting more rides together per driver, per hour than their peers — and that could mean cheaper fares without lower pay for drivers. But the pay piece here is important, and the company's insistence that it wants drivers who "don't just do it for the money" raised concerns that worker pay could be on the chopping block in the future.
If Split's model works, it has the potential to spread far outside the app's current DC base, creating a tier of transit that's significantly cheaper than Uber/Lyft or taxis but more convenient than buses or subways. That could be a big win for consumers. But it needs to work for drivers as well.
The case for optimism: Carrying more passengers at once lets everyone win
For all their posturing as bold disrupters, Uber and Lyft are remarkably inefficient services. Their drivers spend much of their days idly around, transporting nobody and collecting no fares; see Tim Lee's account of his week as a Lyft driver for an example of what this looks like. Apart from the very recent introduction of carpooling options, the apps generally only allow for one passenger or group of passengers at a time, meaning seats that could be used to transport other users are wasted. And their emphasis on point-to-point transportation means a lot of time spent with drivers and users trying to locate each other or trying to negotiate a drop-off point.
So Split CEO Ario Keshani designed his app to eliminate these kinds of inefficiencies. "The algorithm and the routing and the entire approach is designed from the ground up to make it as efficient as possible," he told me. "At scale, our drivers can pick up way more people." He claims that the company has seen as many as eight rides per hour, per driver; four or five per hour per driver is typical. By contrast, New York taxis averaged 1.5 rides per hour per driver, and according to data released by Uber (warning: Excel doc) on its New York operations, Ubers average 1.47 rides per hour per driver, reaching a max of 1.96 at peak hours. Split, in other words, is more than doubling the efficiency of its competition.
The most obviously efficiency improvement in Split is sharing. "If you've got two people in the car, then you've got two people in the car, and that's two fares," Keshani says. But the company differs from other ride-sharing services in other ways that are meant to minimize wasted time and maximize the time drivers can spend transporting users. While Uber and Lyft require drivers to accept requests from users, Split's algorithm automatically assigns users to cars without requiring approval from the driver. It sends a new destination to the driver as soon as the algorithm has matched her with a user, and the matching process takes less than a second. That saves time, especially when sharing is involved; it lets Split seamlessly add a user to a car that's already transporting someone else. Even when someone's being transported solo, the algorithm will sometimes assign cars a passenger to pick up as soon as their current trip is over, meaning drivers don't have to wait at all for another assignment.
Split also doesn't pick up and drop off at exactly the point that a user specifies. Instead, it has a list of predetermined pickup/drop-off points and uses the closest ones to a user's location and destination. That means users have to walk a bit further, but not much; "They're usually within a block of where you are," Keshani says, which jibes with my experience.
This also greatly reduces confusion about where the trip is beginning and ending. "I feel like every time I book an Uber or a Lyft, I wind up getting on the phone with the guy and saying, 'Where are you? Oh, you're at 14th and U? I'm at 14th and W! That's where I put the pin!'" Keshani explains. "It results in missed connections and wasted time. That time is money for the driver. That time is lost efficiency in the system." Split also gives passengers less time to hop in than Uber. "If you're late for your pickup, we're not going to wait for you, especially if there's another customer in the car," Keshani says. "It's not like I'll call the car and he'll wait for me for five minutes so I can brush my teeth or finish that last beer."
The company is also hugely proud of the algorithm it's using. As my colleague Joseph Stromberg explained in a piece on Split and other services this past July, the algorithm was purchased from a Finnish company after it was used to create an on-demand bus system pilot in Helsinki. "It's 25 or 30 man-years of engineering," Dan Winston, Split's director of strategy and business development, says. "These guys were at a research university in Helsinki, just focused on this problem, had the idea, and worked with the transit authority there to scale up a little pilot. We heard about this and said, 'Wait a minute. This sort of sounds like the same vision from a different angle.'"
It makes sense that the idea would come from computer science researchers. Split's trying to solve a variant on one of the most famous problems in computer science: the traveling salesman problem, which asks for the best route to take between a number of cities so as to visit all of them in the least time. Calculating every possible route and comparing them would require far too much time and computing power, so an algorithm to find an approximate solution quickly is necessary.
That's especially true when you're solving these problems in real time, like Split is. "It's the traveling salesman problem but way more difficult, because it's real time and it's multiple traveling salesmen," Keshani says. And the company has an algorithm that can produce pretty good solutions to that problem in less than a second. That's a big time savings, especially compared with Uber or Lyft, where a given car's path is determined ad hoc by individual drivers.
Even the service's biggest downside — a limited service area that doesn't cover all of DC — makes sense from an efficiency perspective. A smaller service area means drivers have to circulate over less space, and improves the odds that routes will overlap with each other.
The case for pessimism: Will the pay really be different from Uber and Lyft for long?
I left Split convinced that the company actually knows what it's doing, cost-savings-wise. I can't independently verify Keshani's claim that cars are averaging four to five rides per hour, but if it's anywhere close to true, the company is operating more efficiently than its competition and extracting meaningful cost savings, which it's passing on to consumers. But I also left more skeptical that the system will keep working out for drivers.
For one thing, while Split drivers I talked to suggested that they thought of themselves as being hourly workers, they're actually independent contractors, and Keshani suggested that the hourly pay is a temporary measure. "Our goal is to have a system where we take a cut of each fare and others take the rest, similar to the other operators," he told me. "But we're still new, so we're guaranteeing the drivers that they're going to make enough."
He insisted, though, that the company remains committed to paying drivers well. "We have a set of strictly defined core values that tell us and remind us all the time that our drivers should be treated well and compensated well," he continued. "It's not about money for them — it is about money, obviously it's always about money, but it's not just about money. It's also the fact that we interact with our drivers all the time, that we treat them like partners, which they really are."
This all sounds nice, as does the company's practice of letting drivers pop by the central offices whenever they like to talk to the CEO or other staff. But when I pressed Keshani on what this commitment to workers means in monetary terms, he demurred. He declined to say, precisely, what drivers make.
"Will you be happy if I tell you that it is a good enough number that the drivers feel it's fair?" he asked. "I don't want to give it because there's a chance that it will change or modify."
When I asked whether Split pays for its workers' costs — their cars, their car insurance, their gas — his answer was basically no. "What our drivers can make more than compensates for all those costs," he said. "We're not telling the driver, 'Hey, you spent $20 on gas, here's $20.'" He insisted the company helped in other ways — it partnered with Hyundai to give drivers discounts on new cars, for example — but in this respect it's not that different from Uber or Lyft.
Both Keshani and Winston insisted that a focus on pay missed the point: They're trying to recruit good drivers who aren't just doing it for the money. "If we're talking to somebody and their first question is, 'All right, what am I making for this?' I'm not actually sure that's the person we'd want," Winston told me. "That's not our first selling point." But as admirable as "we want workers who aren't just in it for the money" sounds, it can also serve as a convenient excuse for not raising wages.
The Split team does not look like vicious cost cutters who are eager to screw drivers over as soon as their network is big enough. But my hopes that they were trying a compensation model that was fundamentally different and more generous to workers weren't really borne out.