Uber Eats loses customers at the second failure: when support does not fix the first

The headlineUber Eats’ customers are polarised and the picture is stable; the damage comes from what happens after an order goes wrong

Uber Eats’ 2026 reviews describe a divided customer base. Across 32,471 Google Play reviews written between 1 January and 12 November 2026, 51% are one star and 34% are five stars; only 15% sit in between. The average rating is 2.64. 61% of reviews contain a complaint about a specific part of the service, and 2,900 reviews (89 per 1,000) contain a leave signal.

Where the problem sits matters more than its size. The largest complaint themes are customer support (19.6% of reviews), app problems (13.2%), fees and prices (12.2%) and refunds (10.9%). Customers who complain about support are three times as likely to threaten to leave as those who do not (19.5% against 6.4%), and customers who complain about refunds are 2.4 times as likely as the average review. Put simply, Uber Eats has a recovery problem: the first failure (late, wrong, cold or missing food) is survivable; the second failure, when support, refunds and charges do not put it right, is what customers say they leave over.

This article follows the questions a growth, product or CX team would ask: whether customers are leaving or only complaining, which themes keep coming back, why they threaten to leave and where they go, what they love and recommend, which features they ask the app to add, whether app updates help, how the company answers, and what to test first.

Leaving or complaining?Customers are not complaining more or leaving faster: every monthly line is flat, and April’s apparent collapse is missing five-star reviews, not extra anger

Monthly average ratings stay between 2.59 and 2.73, the complaint rate between 59% and 64%, and leave signals between 82 and 95 per 1,000 reviews. The distribution is U-shaped: 51% one star, 34% five stars. Leave signals are concentrated at the bottom of the scale (16.4% of one-star reviews, 6.6% of two-star, 0.2% of five-star), so the unhappy half of the base is where retention is decided.

The one exception in the raw data is April, where the average rating drops to 2.35 and the complaint rate rises to 69%. The cause is a 13-day window, 21 April to 3 May. In it, five-star reviews fell from about 38 a day to about 4, while one- and two-star reviews fell only from 62 to 46 a day. Complaint reviews per day were lower, not higher, so this is a collection or visibility effect, not a service failure. We exclude the window from every monthly rate in this article and keep it in all totals. It is worth asking whoever owns the Play Console whether review prompts or visibility changed on the two builds released at the time.

The one signal that is moving is delivery delay. Delay complaints rose from 6.4% of reviews in August and September to 8.5% in October and 11.1% in the first twelve days of November (p below 0.001), and cold or poor-condition food complaints rose from 3.8% to 5.4% and 5.5% over the same periods. The two themes move together and are the pattern one would expect before a peak ordering season.

What keeps coming backCustomer support, app problems and fees top the list in every month, and the same themes sit in the same order all year

Customer support is the largest theme at 19.6% of reviews, and it is also the one with the strongest link to leaving. App problems (13.2%) are mostly about the app itself, payments and the ordering flow rather than a single bug. Fees and prices (12.2%) and refunds and compensation (10.9%) complete the top four. A further 8.5% of reviews complain about “delivery” without saying what went wrong.

The themes that matter most for retention are not the biggest. Order accuracy and missing items are only 5.1% of reviews but carry the highest leave rate (22.2%, 2.5 times the average), followed by refunds (21.3%) and unexpected charges (19.9%). Two smaller themes are new enough to name: ads, pop-ups and notifications (1.4% of reviews, 468 reviews) and tipping (1.3%).

Why customers leaveCustomers leave at the second failure: support, refunds and charges appear in 29% of reviews but 62% of leave signals

Of 2,900 leave signals, 2,636 are explicit exits (“deleting”, “never again”), 260 name a rival they are moving to and 4 ask to uninstall. Customer support appears in 42.8% of them, refunds in 26.0% and app problems in 21.4%. Fees, often assumed to drive churn, appear in 18.0%.

The pattern is clearer when reviews are grouped by what went wrong. 27.2% of reviews describe an operational failure: late, wrong, cold or missing food, a courier problem or a cancellation. Where nothing else is wrong, 12.9% of those reviews carry a leave signal. Where support, refunds or charges also fail, the rate is 22.4%; where nothing at all is wrong, it is 3.1%. 56.0% of operational-failure reviews also describe a recovery failure, so more than half of first failures turn into second failures.

Two newer triggers sit on top of that pattern. Uber One and subscription prompts are mentioned in 5.2% of reviews and those reviews leave at 25.7%, 3.2 times the rate of the rest; 13.6% describe a sign-up they did not intend. Automated support is mentioned in 4.2% of reviews, with a 20.0% leave rate and 28.2% who warn other customers, and the mention rate has climbed from 3.4% to 4.9% during the year. A further 700 reviews (2.2%) say they could not reach a person or ask for one, with a leave rate of 22.7%.

How customers compareDoorDash is named in two of every three rival mentions, and leavers who go there cite support and refunds more than price

Only 4.4% of reviews (1,430) name a food-delivery rival, but those that do are the most hostile in the dataset: 79.3% are rated one or two stars and 39.0% carry a leave signal. DoorDash accounts for 929 of them, 65%; Deliveroo (111), Just Eat Takeaway (102), Grubhub (85), Mr D Food (42) and Skip (31) trail far behind.

Four in five leavers name no destination: 558 of 2,900 leave signals (19.2%) name a rival, and 358 of those name DoorDash. Among DoorDash-naming reviews with a clear preference, 72% prefer DoorDash, and the ratings follow: reviews that prefer DoorDash average 1.14 stars while those that prefer Uber Eats average 4.05. The Uber Eats supporters are a real group (152 reviews), but they are outnumbered about 2.6 to 1.

Why customers name a rival is a service question rather than a price question. Complaint reviews that name a rival carry more support complaints (43.8% against 31.3%), refund complaints (23.0% against 17.5%) and delay complaints (16.3% against 12.6%). Among the 358 reviews that threaten to leave and name DoorDash, support appears in 44.4%, refunds in 29.1%, app problems in 20.1% and fees in 16.5%, which is below the 18.0% for all leavers.

What they loveCustomers praise the app and the delivery, but the few who recommend Uber Eats talk about price and range

Among reviews that praise something specific, 47.5% praise the app experience and ease of use and 32.9% praise delivery. Price and promotions (13.1%), food quality and variety (11.5%), couriers (4.7%) and support and refunds (4.5%) follow.

Recommending Uber Eats is rare (266 reviews, 0.8%), and what recommenders say differs from general praise. They mention price and promotions 1.9 times as often (24.4% against 13.1%), food quality and variety 1.8 times and support and refunds 1.8 times as often, while speed (0.7 times) and ease of use (0.9 times) do not set them apart. Convenience is the entry ticket; value and range are what turn a satisfied customer into an advocate.

The same people sometimes defend Uber Eats against rivals: 110 positive segments praise it as an alternative provider, and the 152 DoorDash-naming reviews that prefer Uber Eats average 4.05 stars. Their reasons are concrete, such as cheaper prices, more deals, easier use and help that resolves problems.

What they ask forFew customers ask for features (2% of reviews), but the requests are specific: rate the driver, control the tip, cancel without penalty, filter by diet and confirm the address

681 reviews (2.1%) ask for something the app does not do, and 46% of those are too general to act on. The classifiable requests are narrow and practical. They cluster into nine topics, led by driver choice and rating, tip control, promotions and search, with a long tail of smaller asks.

The detail matters more than the topic. Customers do not ask for “better drivers”; they ask for a button to request a new driver when theirs is idle, a separate rating for the courier, the restaurant and Uber Eats, and a limit of one order per driver. They do not ask for “better tipping”; they ask to tip once the food has arrived and to stop tip suggestions being calculated on fees and taxes. The table below sets out each request in the customer’s terms.

Ranking the nine topics by volume, helpful votes and their link to leaving puts driver controls (score 71), promotions (64), tip control (56) and location safeguards (55) at the top. Cancellation without penalty ranks sixth overall but has a 21.3% link to leave signals, close to payments and far above tipping or search. Tip control and search are cheap and visible but barely linked to leaving, so they are goodwill moves.

The feature labels understate the largest unmet need, which customers describe as a complaint rather than a request: reaching a person. 700 reviews (2.2%) say they could not reach one or ask for one, and they leave at 22.7%. Likewise, clearer refund rules for missing and wrong items appear in complaints far more often than in feature requests. Both are covered in the recommendations.

How customers behaveComplaining customers act: one in five warns others and one in seven threatens to leave, and failures arrive in pairs

Among the 19,879 reviews with a complaint, 18.5% warn other customers, 14.3% threaten to leave, 12.1% ask for a refund and 9.5% compare Uber Eats with another provider. In other reviews these behaviours are close to absent (0.3%, 0.5%, 0.1% and 3.2%). Warning others is strongest where trust is in question: 62.5% of reviews in the trust and scams theme tell others to stay away, against 18.5% for all complaint reviews.

Failures also cluster. 20% of delay complaints also describe food that arrived cold or in poor condition, four and a half times the chance rate; refund complaints appear with order-accuracy complaints four times more often than chance; and a quarter of support complaints are also refund complaints. These are not separate problems but single bad orders described in several ways, which is why fixing one link in the chain (a visible refund after a missing item) can move several themes at once.

A star-rating dashboard would miss part of the picture: 1,086 reviews rated four or five stars (8.7% of them) still contain a complaint, and they make up 5.5% of all complaint reviews.

Do updates work?A new build ships about every week and none of them moves complaints or ratings

The app moved from build 6.301 in January to 6.347 in November, roughly one release a week. The first eight well-reviewed builds (n = 5,559) show 56.5% of reviews with a complaint, 11.7% with an app complaint and an average rating of 2.86. The last eight (4,134 reviews) show 57.3%, 12.6% and 2.80. That is no improvement and no deterioration.

Of 44 comparisons between consecutive builds, 5 differ significantly in complaint rate (1 better, 4 worse), slightly more than the two or so that chance alone would produce at the 5% level, and the first-eight against last-eight comparison shows no net change. App complaints stay between about 9% and 16% on every build, so the app theme (13.2% of reviews) is a standing problem rather than a release-specific one.

Reputation and content auditUber Eats answers about 1 in 270 reviews publicly, while 20 reviews carry 21% of all helpful votes

Only 119 of 32,471 reviews have a public company reply (0.37%), and 97% of those replies go to complaints. The replies are formulaic: 86% point the customer to a help form or page, 39% apologise and none thanks the reviewer. Yet reviews are read as a source of advice: 23.2% have at least one helpful vote, 95.0% of all helpful votes sit on complaint reviews, and the 20 most-helpful reviews hold 21.2% of all votes.

Those 20 reviews are almost all complaints (all 20 carry one, 19 are rated one or two stars) and they speak about the themes the rest of the data points to: app problems (10 of 20), customer support (10), fees (6) and delays (5). Two of the 20 have a reply. The reviews that shape what prospective customers read are, for the most part, the ones the company has not answered.

Content ideasCustomers ask the same nine questions in their reviews; each is a help article or in-app message waiting to be written

Nearly three in every hundred reviews contain a question, and nearly half of those fit nine recurring topics. Charges and prices lead, followed by courier behaviour, refunds and the difficulty of reaching a person. Several of the questions are really complaints in disguise, but each points to a moment in the order where a short, well-placed explanation would prevent a one-star review.

Where to startSix moves for the next quarter, sequenced by the strength of the evidence

The analysis points to six actions. The first three address the second failure, which the data link most closely to leaving; the last three address seasonal risk, reputation and customer-requested features.

1. Make refunds for missing, wrong and undelivered orders automatic and visible

   Refunds appear in 10.9% of reviews and carry a 21.3% leave rate (2.4 times average). When a recovery problem joins an operational failure, the leave rate is 22.4% against 12.9%, and refunds appear in 29.1% of leave signals that name DoorDash.

   Owner

   Operations and support policy

   Measure

   Refund-complaint share (10.9%); share of operational failures that also carry a recovery problem (56%); leave signals

   First test

   For one cohort, credit a missing item when it is flagged in the app without a photo requirement and show a status line with a time; compare refund and support complaints after four weeks.

2. Put a person within reach after one failed automated contact

   Support appears in 19.6% of reviews and 42.8% of leave signals, with a 19.5% leave rate against 6.4%. Mentions of AI or automated support rose from 3.4% to 4.9% of reviews and carry a 20.0% leave rate; 700 reviews ask for a real person.

   Owner

   Customer care

   Measure

   Support-complaint share (19.6%); AI and bot mention share (4.2%); leave signals citing support

   First test

   Offer a call-back or live chat after the second unresolved automated turn on orders with an open refund request, and compare complaints with a control group.

3. Separate Uber One offers from checkout and make cancellation one tap

   Uber One and subscription reviews are 5.2% of the total, leave at 25.7% against 8.0%, and 13.6% describe a sign-up they did not intend. Customers ask for the order button before the subscription button, a confirmation screen and a “never show again” option. Recommenders value promotions, so protect the discount value.

   Owner

   Growth and membership product

   Measure

   Uber One mention share (5.2%); leave signals; complaints citing unintended sign-up

   First test

   Move the free-trial offer off the checkout confirmation screen for a cohort and add a renewal reminder; compare complaints and trial conversion.

4. Prepare Q4 delivery capacity and give customers a free way out of long waits

   Delay complaints rose from 6.4% of reviews in Aug and Sep to 11.1% in early November and cold-food complaints from 3.8% to 5.5%; the two appear together 4.5 times as often as chance. Order cancellation carries a 19.0% leave rate and customers explicitly ask for penalty-free cancellation and limits on stacked orders.

   Owner

   Operations and logistics

   Measure

   Weekly delay and food-temperature complaint share (baselines 6.4% and 3.8%); cancellation complaints

   First test

   Allow free cancellation when no courier is assigned within a set time or the ETA slips past a set limit, and re-quote the ETA; operations to set the thresholds.

5. Answer the reviews others read and publish the nine answers

   The 20 most-helpful reviews hold 21.2% of helpful votes and only two have a reply; overall reply coverage is 0.37% and 86% of replies are a form link. 942 reviews contain a question, and nine help articles would answer about half of them.

   Owner

   Customer care and content

   Measure

   Reply coverage on the 50 most-helpful reviews; share of reviews that ask a question; warn-others rate

   First test

   Reply to the 50 most-helpful reviews with a specific resolution path and publish the first three articles in Exhibit 25 (charges, fees, courier behaviour).

6. Ship the small requests customers ask for by name

   Customers ask to tip after delivery (12 reviews), to confirm the delivery location (5), for Google Pay and Apple Pay (8), for allergen and halal filters (16) and distance sort (11). They are cheap and visible but have weak links to leaving, so measure ratings and request volume, not churn.

   Owner

   Product

   Measure

   Feature-request share (2.1%); ratings of reviews that mention the shipped feature

   First test

   Release an optional tip after delivery and a delivery-location confirmation prompt to one market and track mentions.

Data and methodAbout this analysis