AI-generated complaints: should you answer AI with AI?

A complaint that used to fit on one page can now run to 20. It cites case law from another country, raises every issue the customer could conceivably have, and arrives from someone who has every reason to believe it’s watertight, because a chatbot told them so.

AI-generated complaints are becoming a familiar sight in collections and complaints teams, and the sector is only starting to work out how to respond. Our recent webinar, Your customers are using AI: is your collections team ready?, chaired by Flexys CEO James Hill, put the question to Mel Chell, CEO of Solaris Law, and Sam Challenger, an operations and collections specialist with 15 years in financial services. This post covers what they said, and where we land on it.

How big is the shift in AI-generated complaints?

Templated complaints aren’t new. Most teams remember the “freeman of the land” letters doing the rounds, and learned to spot them quickly. What’s changed is scale and sophistication. When we polled attendees during the webinar, 38% said more than half of the complaints they receive are now AI-assisted, and three in five put the figure above 20%. Only 15% said under 5%.

These complaints are longer and more technical, though not more accurate. Some lean on US law that has no bearing on a UK agreement. That doesn’t make them easy to deal with. A customer can misquote a case, but the firm still has to read every point and respond to it.

Why answering AI with AI is the wrong move

Faced with a 20-page complaint, the temptation is obvious: paste it into an AI tool and ask for a draft reply. The panel’s view, and ours, is that this is the wrong response. It creates a loop where each generated reply invites another generated complaint, and nobody gets closer to what the customer actually needs.

Underneath the generated argument, there’s usually one thing that went wrong. A payment that wasn’t applied. An arrangement that didn’t hold. A letter that landed at a bad time. The job is to find it.

That might mean picking up the phone rather than replying in writing. It might mean reading the text for signs of vulnerability, which the polish of an AI draft can hide. And where a complaint raises legal points, it means having them checked by someone qualified, not by another model.

AI still has a job here. Triage, and first drafts that a person reviews, are sensible uses. Writing the reply is not.

What it means for customers

The impact isn’t only operational. Some customers are skipping the complaints process altogether and going straight to court, with AI-drafted particulars of claim. Where those claims lack merit they can be struck out, with costs awarded against the customer. The customer ends up worse off, and the debt or grievance is still unresolved.

Many of these customers are already in financial difficulty. AI will tell someone what they could do without telling them what it might cost. That’s a poor outcome by any definition, and the cost of handling these cases finds its way back to all customers eventually.

How teams are adapting

There’s no settled playbook, but the themes from the webinar are consistent:

• Firmer guardrails on AI use: reviewing policies more often, pointing agents to approved internal knowledge, and keeping customer details out of public tools.
• AI for triage and first drafts, with a person checking and signing off every reply.
• Earlier conversations with dissatisfied customers, before a grievance becomes a 20-page complaint.
• Clearer information about how complaints are handled, so customers know what to expect.
• Tracking which complaints involve AI, and how those customers fare, so you can see what’s changing.

That last point depends on where the complaint lives. If complaints sit in one system and the collections activity in another, you can’t see the full picture of what went wrong, or evidence how it was put right.

Keep the human in the loop

Keeping a person in the loop is widely agreed. Where their judgement sits in the process, and at which points, is still open. The picture could look different in six months.

What won’t change is the test that matters: did the customer end up in a better position? An AI-generated complaint answered by an AI-generated reply doesn’t pass it. A person who finds the real problem and fixes it does. That’s better for the customer, lower risk for the firm, and cheaper than a cycle of 20-page exchanges.

Watch the full webinar, Your customers are using AI: is your collections team ready? to hear Mel and Sam’s discussion in full.

This article is for general information and does not constitute legal or compliance advice.