Amazon Restaurants

» see primenow.amazon.com/restaurants as it lived, on the wayback machine

~ the last words ~

amazon restaurants launched in seattle in september 2015 as a feature inside the prime now app, which was itself only nine months old at the time. you opened prime now, switched to the restaurants tab, and ordered from a participating local restaurant for delivery in under an hour. the service was bundled with prime, no extra membership, no surge pricing, no minimum order. amazon was treating restaurant delivery the same way it treated everything else: lose money on the unit economics, win on the membership lock-in.

it expanded across about twenty us metros over the next two years (new york, los angeles, chicago, austin, dallas, atlanta, miami, seattle, portland and a long list of others) and into london in september 2016 as the only non-us city it ever reached. the partner list peaked around 7,601 restaurants in 2018. that number sounded large in isolation but the comparison numbers were brutal. grubhub had north of 115,000 restaurants by then. doordash and uber eats were both growing faster than amazon was adding partners.

amazon shut down the london service in november 2018, citing market saturation. the us shutdown was announced june 11 2019 and took effect on june 24, with amazon redirecting prime now users to its grocery business via whole foods and amazon fresh. about a month earlier amazon had led a 575 million dollar funding round for deliveroo. the message was that amazon would rather own a piece of a winning delivery operator than try to be one itself.

nothing about this service touched turkey. by 2015 the istanbul / ankara food-delivery layer was already crowded with yemeksepeti (founded 2001, sold to delivery hero in 2015 for 589 million dollars) and getir's eventual entry into the food-delivery market, and amazon turkey did not even exist as a marketplace until september 2018. i read about the shutdown the way you read about any small piece of the amazon graveyard: it was the part of the bezos era where amazon kept losing in operations-heavy markets where the incumbents already had a four-year head start.

~ the record ~

~ why it failed ~

the food-delivery business is a two-sided network problem. you need enough restaurants to satisfy every search a hungry user does at 7 pm on a tuesday, and you need enough users to make it worth a restaurant signing up. grubhub started in 2004 and seamless in 1999. by the time amazon entered in 2015 those two had merged and had a ten-year head start on the supply side. doordash had been growing aggressively since 2013 with a different model (it ran its own driver network and onboarded restaurants without their permission via screen-scraping). uber eats launched in 2014 with the unfair advantage of an existing global driver fleet.

amazon brought logistics scale that did not actually translate to restaurant delivery. prime now used contract couriers in branded sprinter vans, which works for groceries (one stop, one big drop) and does not work for restaurants (many small trips, hot food, narrow time windows). the math was bad for amazon and visibly worse than the math was for the competitors.

~ the postmortem ~

the announcement on june 11 2019 was short. amazon told employees there would be reassignments, told partner restaurants the june 24 cutoff was firm, and told the press that the company was focused on amazon fresh and whole foods delivery instead. grubhub stock jumped 7 percent on the news. by late june prime now's restaurant tab was gone from the app and the restaurants.amazon.com domain was redirecting.

the deliveroo investment a month earlier reframed the exit. amazon was not leaving the category. it was admitting it could not win the category directly and was buying a position in a regional winner instead. that pattern (kill the in-house attempt, take equity in someone who is already winning) was the same one amazon would later run in india with shadowfax and in the middle east with careem-equivalents.

~ what got buried ~

the most-interesting bit of operating data from the four-year run was the courier-utilization curve. prime now drivers averaged something like one to two restaurant deliveries per shift versus a doordash driver in the same city running four to six. the geekwire post-mortem and the business-of-business analysis both pointed at the same dynamic: amazon had built a fixed-cost delivery network designed for predictable big-basket grocery drops, and restaurant orders made that fixed cost look bad on every metric. the prime now infrastructure survived. the restaurants tab did not.

@}-->--

              )
             (_)
            .-'-.
            |   |
            |   |
            |   |
            |   |
          __|   |__   .-.
       .-'  |   |  `-:   :
      :     `---'     :-'
       `-._       _.-'
           '""""""

« previous  ·  random  ·  next »