Your complaints team already knows what’s wrong with your organisation

Our dog has fleas.

When I realised this, my instinct was to find the sheep-dip equivalent for adorable puppies, throw her in, then burn the house down to be on the safe side. This lack of patience is why I don’t work in customer service.

Happily, I took a deep breath, did some research and called a company that will de-flea your house while you’re out, establishing that I could outsource at least part of my problem. What followed was an entirely practical conversation about how they would make sure that, having treated the animals, I did not spend the next six months providing bed and breakfast for the fleas’ extended family.

I quickly realised — because this is my job — that I was talking to an AI bot. But it sounded remarkably human and answered every question without becoming irritated when I asked a slightly different version of the same thing three times. My mum wouldn’t have noticed. She would have thought she had encountered the most patient and well-informed customer service adviser in the land. She’d have told her about her friend Susan’s dodgy knee, for good measure.

This is not everyone’s experience of customer service automation. We have all sworn at the chatbot that responds to “My parcel has been delivered to the wrong house” with “It sounds as though you want to track a parcel”. Older systems often felt less like talking to a person and more like arguing with a menu. But newer ones can be very good — particularly when the question is factual, the source information is reliable and there is a quick route to a human when things get complicated.

Better than a human — at some things

There are now plenty of examples of AI performing at least as well as people on some customer service measures. When Klarna first introduced an AI customer service assistant, the company reported that it handled 2.3 million conversations in its first month. Customer satisfaction was reportedly on a par with human agents, queries were resolved in under two minutes rather than 11, and repeat inquiries fell by 25%.

Lyft says its AI customer care assistant has reduced average resolution times by 87%, while Octopus Energy’s chief executive reported that emails drafted using AI achieved an 80% customer satisfaction rating, compared with 65% for those written without it.

These are company-reported figures, not a universal law of customer service. Humans remain more useful when a problem is unusual, emotionally difficult or requires someone to take responsibility and exercise discretion.

The ideal customer assistant handles predictable questions well, recognises when it is out of its depth and makes it easy to reach a person who can help. But AI can also make that person better at their job. A study of more than 5,000 customer support agents found that access to an AI assistant increased the number of issues resolved per hour by around 15% on average. The greatest gains were among less experienced staff, suggesting that the technology helped spread the knowledge and practices of stronger performers.

But speedier answers are only half the opportunity.

The people nearest the public know things first

When I worked in government, ministers would regularly return from visits or constituency surgeries and tell officials what they were hearing “on the doorstep”. We officials could be a bit sniffy about this. The minister may have spoken to a bloke in a community centre; we had a survey containing several thousand responses.

But, with experience, you learn that ministers are often picking up something real and more current. They hear the language people used outside a questionnaire and notice which issues made someone well up. Local councillors do this too. They may not have a statistically representative sample, but they have an unusually sensitive early-warning system for things that are beginning to go wrong.

The private sector equivalents are the people in call centres and sales teams. They know which feature customers cannot find, which promise on the website is misleading and which new policy sounds perfectly sensible until it meets an actual person. The problem is that this intelligence often travels upwards as anecdote. A good manager might listen; a busy senior team may ask for numbers. By the time somebody has commissioned and analysed a customer survey, the sales team has been fielding the same objection for months — and found a new one.

From anecdote to insight

I recently came across a council that produced its annual complaints report by asking every team, once a year, to fill in a form logging what complaints they had received. It had some information from the main corporate contact point. But if a complaint had arrived through another inbox, been sent directly to a service or emerged during an ordinary exchange with a resident, its inclusion depended on somebody remembering it and volunteering the details.

The information then had to be gathered, cleaned up and written into a report. By the time senior leaders reviewed it, some complaints could be well over a year old. It was less an early-warning system than archaeology. Not because nobody cared: collecting the information required a separate manual exercise in the absence of a simple, affordable alternative.

In an AI age, we can do much better. With the right safeguards, organisations can analyse complaints and feedback as they arrive, giving teams information that is useful to them now. And that creates a useful virtuous circle. If frontline teams can see that the information helps them spot problems, make a case for change or demonstrate that an improvement has worked, they are more likely to contribute their own insight. The technology can identify a pattern; the people doing the work can usually explain it.

Most organisations already possess a vast, messy and largely unused dataset about how their services are working. It sits in complaint letters, emails, call notes, chat transcripts, online reviews, returns data and the reasons sales did not convert.

AI can read across that material, identify recurring themes and distinguish an isolated complaint from an emerging pattern. It can also help test whether things are improving. If you change your returns process, rewrite a confusing letter or give staff clearer guidance, do complaints disappear? Do repeat contacts fall? Does satisfaction improve? That is the difference between counting complaints and learning from them.

Complaint numbers alone can mislead. A low number might mean customers are happy, or that complaining is difficult and dissatisfied customers have gone elsewhere. A rise might reflect deteriorating performance, improved access or better recording. The useful questions are more specific: What are people struggling with? Is it happening repeatedly? What is causing it? Who is affected? What did we change as a result?

The complaint is not the product

In my last blog, I argued that AI-powered complaints are basically a good thing — and, in any case, we have to accept that they’re coming in thick and fast. They allow people who lack time, confidence or “complaints literacy” to explain what has happened and ask an organisation to put it right. Deal with it.

But the growth in AI-assisted complaints means organisations need better tools on the receiving side too — or a lot more resource. Leading AI’s Write My Reply assistant reads a long complaint, extracts the actual issues, checks them against the organisation’s policies and relevant legislation, and drafts an evidenced response for a person to review. Without approved tools, overwhelmed staff will inevitably reach for public AI on their phones — tools that may invent sources and recycle your data.

The immediate benefit is time saved. But the larger opportunity is to treat customer contact as evidence about performance. Which promises are we repeatedly failing to keep? Which processes generate avoidable complaints? What are customers telling us that has not yet reached the board?

People deserve proper answers. But a system that does nothing more than answer and close each case wastes most of what its customers are telling it. AI’s more interesting role may be helping organisations listen while there is still time to act.