I'll start with the numbers. In every hundred sessions of the Lyft app in which the new premium ride appeared at its highest price, the premium sold 1.1 rides. In the same hundred sessions 0.3 people gave up on booking altogether and another 1.3 moved down from the normal ride to the cheap one.
Lyft is Uber's main rival in the United States and, like Uber, it lets the rider choose between different rides at different prices. British readers may know it without realising, because since July 2025 it has owned Free Now, the app many Londoners use to hail a black cab. In the autumn of 2020 Lyft added Fast Pass, an option that promised a car sooner in exchange for a surcharge, alongside the two it already offered. Standard gave a precise waiting time, while Wait & Save cost less because whoever chose it agreed to wait longer. In one of the screens reproduced in the study, Fast Pass costs $24.10 for a car arriving in three minutes.
As it usually does when it launches a product, Lyft ran an experiment alongside the debut. From 5 November 2020 to 20 January 2021 one group of riders never saw Fast Pass, while the others found it in the app at one of three prices assigned at random, the highest roughly four times the lowest. In all, 7,242,942 riders and 47,616,543 sessions were involved, and nobody knew they were part of a study. Economists call this a natural field experiment, and among the people who made it famous is John List, an economist at the University of Chicago who was Lyft's chief economist at the time and now holds the same role at Walmart. The results came out this month as a working paper of the National Bureau of Economic Research, the American research body where economists circulate their work before it goes through the journals. List signs it with Gregory Sun and two professors from Bocconi University in Milan, Nicola Gennaioli and Matthias Rodemeier.
According to the textbook, those numbers should not exist. Economists call regularity the idea that an extra option can only take customers away from the others and never add any, and as early as 1982 a group of researchers at Duke University led by Joel Huber had shown that real people do not behave that way. In 1989 Itamar Simonson, a consumer behaviour researcher who later moved to Stanford, described what he called the compromise effect, the tendency of people faced with three alternatives to pick the middle one because it is the easiest to justify. A very expensive product therefore serves above all to make the one next to it look reasonable. This is where the three-tier price list comes from, the one you now find on almost every software pricing page, with the top tier placed there partly to steer people towards the middle.
Lyft broke regularity too, only the riders went the other way. In the first session with Fast Pass on screen the effect was small, since Wait & Save rose by about 1.5% on its usual level and the number of people who gave up by 0.41%, figures that on their own would make any marketing manager shrug. As the weeks went by, though, the effect kept growing. Counting all sessions, the shift towards the cheap ride became four or five times larger, and it also grew with the price of Fast Pass, so much so that walk-aways rose most at the highest price, where Fast Pass was chosen by about one rider in a hundred. The option almost nobody bought was the one that moved everybody else the most.
Fast Pass appeared in about a third of the sessions of the riders involved, and the oddest finding concerns the other two thirds. Even when Fast Pass was absent, those who had seen it at the high price chose Standard half a percentage point less often and Wait & Save four tenths of a point more often than those who had seen it at the low price. They carried the price around with them.
The authors ruled out the obvious explanations, including simple inattention, and put forward one that comes from Gennaioli's work with Pedro Bordalo of Oxford and Andrei Shleifer of Harvard, two economists who have spent years studying how attention runs towards whatever stands out. A price that surprises makes you notice every price, and when the gap between the most expensive and the cheapest looks large the customer gives price more weight than before. Memory does the rest, because that high price stays in the mind and comes back every time the app is opened. These were experienced riders, after all, with an average of 124 previous Lyft sessions, who knew the usual fares well and could therefore spot a jump.
The paper does not name the cities where the experiment took place, and there is a calendar detail worth keeping in mind, which is that those eleven weeks fell in the middle of the second Covid wave, when anyone paying for a car may have had more than one reason to watch what they spent. The result stands, but I would bear it in mind before turning it into a law.
Anyone in marketing should look above all at where the bill ends up. Whoever reads the Fast Pass figures sees a product that sells and carries a surcharge on every ride, while the cost is spread across the Standard ride that loses customers, the bookings that never happen and the later sessions, where nobody would think of charging it to Fast Pass. The authors note that a demand model built on the assumption of regularity would have given Fast Pass the credit for increasing demand at Lyft. The study does not calculate the effect on revenue, but it points out that a lost ride costs the whole fare and that the lost rides alone could wipe out a Fast Pass surcharge worth about a quarter of the price. On top of that come the riders who moved from Standard to Wait & Save, who keep travelling and simply pay less.
There is a practical consequence for anyone who tests prices. If the effect starts small and grows over weeks, an A/B test lasting a few days, meaning a comparison between two groups of customers shown two different versions of the price list, records mostly the sales of the new tier and almost nothing of the rest. And if you only measure what happens on the screens where the new option appears, you also miss the part customers carry with them into later visits. Lyft noticed because it followed 47 million sessions for eleven weeks.
Before anyone rewrites their price lists, it should be said that this is a single experiment, still under review, on a very particular kind of purchase, one that is repeated often and always has a free alternative close at hand, namely walking or staying at home. Someone selling a watch or a holiday might see nothing of the kind. The question I take away concerns how many premium tiers have been judged by their own sales alone, across everything we buy every week with the same gesture, from the online grocery order to the gym membership.