The Loyalty Two Taps Should Have Killed

Based on the research ofSandeep Chitla, Maxime C. Cohen, Srikanth Jagabathula, and Dmitry Mitrofanov, "Customers' Multihoming Behavior in Ride-Hailing: Empirical Evidence from Uber and Lyft," Manufacturing & Service Operations Management, 2026

Sandeep Chitla, Maxime Cohen, Srikanth Jagabathula, and Dmitry Mitrofanov built their case on a panel of 1.4 million rides taken by 162,000 users in New York City, a market with Uber and Lyft installed on nearly every rider's phone and switching costs that should round to zero. The intuitive prediction, the one that shows up in antitrust filings and strategy decks alike, is that a two-tap duopoly erases differentiation: if checking the other app costs nothing, price becomes the only thing that matters, and loyalty becomes a rounding error. The data say otherwise. 83.4% of riders took every one of their rides on a single platform. Among the remaining 16.6% who used both, the authors' structural model, which accounts for operational factors, individual rider behavior, and the dynamic back and forth between platforms, finds they actually considered both apps on only 43.4% of their trips. Most of the time, even a self-described multihomer is just opening the app they always open.

The Commodity That Wasn't

The paper's core contribution is showing that this loyalty survives a structural test, not just a glance at raw usage counts. A naive read of "83.4% single-home" could still be consistent with pure price-shopping if riders simply happened to get the better deal from one app every time. The model rules that out. It treats Uber and Lyft as differentiated products, not interchangeable quotes, and finds that differentiation, not inertia or ignorance, is doing the explanatory work. Riders are not failing to notice a better price next door. They are choosing a platform the way they choose a coffee shop: by habit, fit, and accumulated trust in how it has treated them before, with price as one input among several rather than the only one.

That matters because it inverts the standard platform-economics story about multihoming. The textbook worry about a two-sided market with low switching costs is that it tips toward commodity pricing, since no platform can charge a premium when the alternative is one tap away. Chitla and colleagues show the tap is cheap but the decision underneath it is not. A related paper on this same New York City market, by Jeffrey Fossett, Michael Luca, and Yejia Xu, found a price gap of roughly 14% between Uber and Lyft on matched trips and only about 16% of riders opening both apps on a given day. That is a story about search frictions in a moving price. This is a different and sharper claim: even riders who do use both apps are not treating every ride as a new shopping decision. The friction is not just that checking is annoying. It is that most riders, most of the time, are not trying to check at all.

What Riders Say Versus What They Do

The gap between stated preference and revealed behavior shows up elsewhere in the rideshare literature, and it cuts the same direction. Daniel Valderrama and Bruce Cameron's study of rider multihoming in the US rideshare market, published in Research in Transportation Business & Management, surveyed riders directly about whether they use more than one ridesharing app and found substantial self-reported multihoming, well above what a model of actual ride-by-ride behavior like Chitla and colleagues' would predict. Part of the gap is methodological: a survey question about "have you ever used" a second app will always catch more people than a model of what riders actually do ride by ride. But part of it is exactly what Chitla, Cohen, Jagabathula, and Mitrofanov are pointing at. Riders think of themselves as more price-conscious and more willing to shop around than their actual tapping behavior supports. The self-image is multihoming. The transaction log is loyalty.

Riders think of themselves as price-shoppers. Their ride history says they are regulars.

The Driver Side Runs the Opposite Playbook

The asymmetry gets sharper once you look at the other side of the same marketplace. James Parrott and Michael Reich's 2018 report for the New York City Taxi and Limousine Commission, "An Earnings Standard for New York City's App-based Drivers," tracked how many TLC-licensed drivers worked for more than one app company over time. In January 2015, just 5% of drivers drove for two or more platforms. By October 2017, that had risen to 45%: 55% drove for one company, 31.5% for two, and 13% for three or more. Multi-app driving did not creep up. It climbed steadily every single quarter the report tracked, because the economics on that side of the market reward it directly. A driver who is logged into both apps fills idle minutes between rides with whichever platform pings first, and every added app is close to pure upside for utilization. A rider who is deciding between two apps for one trip gets no equivalent payoff for comparing, just a few seconds spent and, per Chitla and colleagues' finding, a service that is not actually identical on the other side anyway.

Xiaotong Guo, Andreas Haupt, Hai Wang, Rida Qadri, and Jinhua Zhao formalize why that asymmetry is durable rather than accidental. Their 2023 paper in Transportation Research Part C, "Understanding multi-homing and switching by platform drivers," builds a structural model of driver labor supply across a ride-hailing duopoly and estimates that both multihoming costs and switching costs are, in their words, sizeable, and that even small reductions in those costs materially shift platform market share and driver welfare. Drivers multihome because the model rewards it and the friction, while real, is worth paying down. Riders single-home because, per the anchor paper, the two services are not actually substitutes in the way theory assumed, so there is nothing to arbitrage by switching in the first place.

What a Platform Operator Should Take From This

The managerial upside in the anchor paper is not "riders are loyal, relax." It is that loyalty this far above the two-tap-duopoly baseline is a targetable asset, not a fixed fact of the market. If 83.4% of riders are already effectively locked in and the structural model can separate them from the 16.6% who multihome, and can further identify which of a given rider's trips are the ones where they are actually weighing both apps, that is a segmentation a platform can act on instead of guessing at.

The instinct to compete primarily on price treats every rider as the Fossett, Luca, and Xu comparison shopper. Chitla, Cohen, Jagabathula, and Mitrofanov's structural model shows that rider does not represent the market; they represent a minority, and even that minority is not shopping on most of their own trips. A platform that can tell the two riders apart gets to spend its retention budget on the 16.6% who are actually deciding, and its margin on the 83.4% who already decided a while ago.

The Part the Tap Count Never Captured

Put the ride-side and driver-side numbers next to each other and the picture of this marketplace changes. Two companies, one app each, a tap costs nothing on either side, and yet one side single-homes at 83.4% while the other multihomes at 45% and climbing. The difference is not switching cost, which is roughly the same cheap tap for both. It is what each side gets for paying it. A driver's extra app fills otherwise idle minutes with money. A rider's extra app offers, on the paper's own structural evidence, a service that is not different enough from the first one to be worth the look, on most rides, most of the time. Two-tap economics was always a story about frictionless switching. It turns out the friction was never the binding constraint. The product differentiation was.