Your customers are sending AI to deal with you. Is your service ready?
Ofcom's research on AI in telecoms is about one sector, but it describes a shift every regulated, customer-facing business will face: AI on both ends of the conversation.
On 25 September, Ofcom published a discussion paper on how AI is changing the way people deal with their phone and broadband providers. The numbers are modest for now. Just over half of online UK adults use AI in some form. Eight per cent have used an AI tool in relation to their telecoms services in the past year: 6% to resolve an issue, 3% to compare prices or products, and 2% to make a complaint. Ofcom expects that share could grow quickly.
The paper's real value is that it describes two different things happening at once, and most firms are only thinking about one of them.
Direction one: your AI talking to customers
This is the familiar one: chatbots and AI-assisted agents handling customer contact. Ofcom's position is that its existing rules protect consumers effectively, so it isn't writing new AI rules yet. But its expectations are clear. Customers should be able to tell when they're dealing with AI, and there should be a clear route to a person when the AI can't resolve the problem.
The research explains why. People were comfortable using AI for simple research and routine troubleshooting, and much less comfortable with billing disputes, outages or anything complex. A common concern was being forced to use AI with no way of reaching a human. Some participants loved it; others found it cold. Both groups exist in every customer base.
The same pattern is showing up in financial services. In a September speech, the FCA highlighted its research showing that almost half of young people who use AI for investment decisions wrongly believe AI-generated financial information is regulated. Customers are already trusting AI more than the rules justify, which puts more weight on the firm to be clear.
Direction two: your customers' AI talking to you
This is the one most firms haven't designed for. Ofcom's paper looks ahead to personal AI agents that compare services, negotiate contracts and act on a customer's behalf. Participants could see the time it would save. Most also worried about their personal data and privacy, and about verification and authorisation, and many believed they'd still get a better deal from a human.
That word, authorisation, is the crux. When an agent contacts you claiming to act for a customer, how do you know the mandate is real, what it covers, and whether the customer will stand behind what the agent agreed? Most customer service processes authenticate a person. Very few have any concept of a delegated agent with limited authority.
Two decisions to make now
The first is about your own AI. Decide, journey by journey, where a human route must always be available, and make it findable rather than hidden behind three failed bot attempts. Complaints, vulnerability, disputes and anything with financial consequences are the obvious candidates. In regulated sectors this isn't just good service; it's how you'll evidence fair treatment if a regulator asks.
The second is about your customers' AI. Decide what an agent acting for a customer is allowed to do in your services, and how it proves it. That might be nothing for now, a read-only allowance, or a defined set of actions with explicit customer consent recorded. Whatever the answer, deciding it deliberately beats discovering your policy the first time an agent cancels a contract the customer didn't mean to cancel.
Both decisions start in the same place: what the business is willing to stand behind, not what the technology can do. Get that clear, and the design of the bot, the escalation path and the agent policy follow from it.
Putting AI in front of regulated customers?
A short conversation on human routes, agent authority and what your regulator will expect to see.