When the Chatbot Won't Let Go: Timing the Escape Routes in Ten UK Apps

Most app reviews are written on the best day. The login works, the dashboard loads, the payment goes through, and everyone goes home happy. But an app is only as good as its worst day. So in November 2026 we picked ten apps that British households actually rely on — high-street banking, energy suppliers, rail retailers and parcel firms — and gave each one a small, believable problem. Then we started a stopwatch.

The test: one broken thing, one stopwatch

We used three banking apps, three energy accounts, two rail retailers and two parcel carriers. The scenarios were deliberately dull: a card frozen by a fraud check, a bill that had been estimated rather than read, a refund that had not arrived, and a parcel marked as delivered to a place that did not exist. Nothing exotic. Nothing that should require a specialist team. Just the sort of thing that makes a normal Tuesday go sideways.

We timed from the moment we opened the help section to the moment a human being understood the problem. Not a bot. Not a queue. Not a form that promised a reply within five working days. A person, on chat or phone, who could actually do something.

Across the ten sessions we spent 124 minutes on hold, in queues or trapped in loops. The fastest route to a human took 5 minutes 42 seconds. The slowest took 41 minutes, and that was after we abandoned the app and rang the number hidden in an old statement. The average was just over 18 minutes. Only three apps got us to a person in under ten minutes.

The loop: where chatbots go to die

Every app had a chatbot. Most of them were polite. Several were useless in a very specific way: they could answer the question they had been trained on, but not the question we had. One energy app responded to 'my bill is wrong' with a link to a page about smart meters. When we said that was not the issue, it offered to start a live chat. The live chat button opened the same chatbot. We went round four times before closing the app and laughing, which is cheaper than crying.

We counted 31 separate loops across the ten tests. A loop, for our purposes, was any moment where the app sent us back to a screen we had already seen without new information. Four apps never offered a phone number at all. One buried the contact route behind Account > Profile > Help > Contact us > Live chat, and then only during what it called 'busy periods', which appeared to mean always.

Rail apps were the most likely to hide behind a web form. Parcel firms were the most likely to offer a chatbot with a name and a cheerful avatar, as if that made up for the fact that it could not find a parcel. Banking apps were the most likely to have a real human available, but also the most likely to make you prove your identity three times before letting you speak to one.

Dead ends and hidden forms

A dead end is worse than a queue, because at least a queue moves. We found dead ends in six of the ten apps. One bank app told us to 'contact us in the app' and then linked to a page that said we could not contact them in the app. A parcel app let us open a claim, then closed it automatically because the tracking number was 'not yet in the system', then sent an email saying the claim was closed because we had not responded. We had responded within four minutes.

Energy suppliers were the slowest overall. That may be because billing disputes need actual people, or because the apps are designed to keep you in self-service long after self-service has stopped being useful. One app offered a callback, which felt like a small mercy until the callback arrived two days later while we were on a train with no signal.

The apps that got it right

The good news is that three apps made getting help genuinely easy, and none of them made a fuss about it. A challenger bank put a 'talk to a person' button on the first help screen, with a realistic wait time next to it. It was accurate to within a minute. An energy supplier offered a callback window and then actually called inside that window. A parcel app let us open a claim in four taps and gave us a reference number immediately, plus a link to a human chat if the automated claim stalled.

None of these were perfect. The challenger bank still asked us to confirm our date of birth twice. The energy supplier's callback was friendly but could not fix the bill on the first attempt. The parcel app's human chat was only open until 6pm. But they all respected the basic rule of support: do not make the customer prove they deserve help.

What good support looks like

After ten tests, the pattern is clear. Good in-app support has three things. First, a visible route to a human, not hidden in a menu or conditional on the chatbot failing. Second, an honest wait time. Third, a reference number or a next step that survives the conversation. Everything else is decoration.

Bad support has three things too. A chatbot that cannot escalate. A contact form that vanishes into a void. A promise of a reply that never arrives. The frustrating part is that these are product decisions, not natural disasters. Somebody chose to put 'contact us' three menus deep. Somebody chose to measure success by how many conversations the bot deflected rather than how many problems it solved.

The verdict: support is the product

We review apps on the good days because that is when they look best. But the real test is the bad day. On a bad day, a banking app is not a dashboard; it is a way to reach someone who can unfreeze a card before payday. A rail app is not a timetable; it is a way to get a refund without writing a letter. A parcel app is not a map; it is a way to find out where the thing went.

The handful of apps that made this easy did not do anything clever. They just put a human behind a button and told the truth about the wait. That should be the baseline, not a highlight. If your app cannot handle a broken Tuesday in November, it is not really finished.