Your call is important to us

Last Tuesday I spent fifty-one minutes in conversation with the chatbot of my electricity company, and I am pleased to report that we have grown very close. She is called Ellie. She is, her greeting informs me, 'always happy to help', a claim she then spent the better part of an hour putting to the test.

My problem was simple. A bill had arrived for a flat I moved out of in March. Ellie asked me to choose from five options, none of which was 'a bill for a flat I moved out of in March'. I selected 'Something else', whereupon she asked whether I would like to hear about her great-value tariffs. I would not. She asked again in slightly different words, which I took for a sign of character.

At minute twenty I typed 'human'. Then, in a spirit of compromise, 'person', 'anybody' and 'help'. Ellie replied that she had not quite understood and invited me to rephrase. I rephrased with feeling. She thanked me for my patience, which by then had become a work of fiction, and asked whether she had resolved my issue. I said no. She said she was glad.

Companies defend all this on grounds of efficiency, and a bot can, to be fair, tell you your balance faster than any clerk. But there is a difference between efficiency and an elaborate machine for making complaints go away. If a bot's real job is to wear you down until you give up, the saving lies not in staff costs but in all the refunds that are never claimed. A system built to be unhelpful has, after all, a flawless record.

I am aware that this sounds like the grumbling of a man who has never been helped by a machine. I have been. When I ask my bank whether a payment has arrived, the bot is quick, accurate and mercifully brief, and I would not trade it for a queue. The trouble starts with the next kind of question, the kind that does not fit a drop-down menu: the double charge, the cancelled flight, the mistake nobody has yet thought to anticipate. For those, a human being is not a luxury but the only part of the system that can actually bend.

Some firms have worked this out. One airline, I am told, lets customers reach a person after a single failed attempt, and finds that its bots are used more, not less, because people trust a door they know they can leave by. It is a modest idea, but it requires something the cheerful bot's designers seem not to have considered, namely that a customer who gets stuck is not a problem to be contained. She is information. Every angry transcript is a map of where the system is broken, and a company that reads them is cheaper to run than one that merely buries them.

In the end a real person called Dawn appeared and sorted out the matter in four minutes, with the sigh of someone who has done it a thousand times. 'She does that to everyone,' Dawn said. I hung up feeling oddly sorry for Ellie, who will never know what it is like to be useful. Then the company emailed to ask me to rate my experience out of five. I gave Dawn five. I looked for a way to give her six, and the form, with the dogged cheerfulness I had come to expect, told me it was happy to help.