The next caller to your contact centre might not be human 

Maybe a year ago, if I had suggested that customers would use AI to contact organisations on their behalf, it would have sounded a little futuristic. 

Today, I’m sure it seems a lot more practical. 

We’ve become used to talking about the AI that organisations deploy to serve customers. Agentic AI, Chatbots, agent assist, knowledge management and automation are all becoming familiar topics.  

What I think we spend far less time discussing is what happens when the customer arrives with their own AI. 

That might sound unusual, but the building blocks already exist. Consumers are becoming increasingly comfortable using AI assistants in their personal lives, and the technology is improving at a pace that would have seemed unrealistic only a few years ago. If you look at the use-cases of Gemini on Android, and Apple Intelligence in the iOS world, they are quickly enabling the adoption.  

The question we need to start answering now is, when an AI contacts your organisation on a customer’s behalf, what happens?  

The first obvious question is how to identify a bot calling your contact centre. 

The truth is, I’m not convinced that’s the right question – or one to invest the effort in. 

The latest voice models are becoming increasingly natural. Response times continue to improve, turn-taking is becoming more conversational and we’re even starting to see providers introduce human-like speech patterns, pauses and breathing. Whilst some of those characteristics may have limited or no value in customer service deployments, they make it increasingly difficult to rely on traditional ideas of what an AI sounds like. 

There may be indicators. The interaction might be unusually structured. Responses may be highly precise. The caller may focus purely on obtaining or exchanging information without the conversational habits we associate with people. There may be tell-tale patterns in how responses are generated or delivered.  

But I suspect those signals will become less reliable over time. 

The more important question is whether we should be trying to detect these interactions in the first place or instead, preparing for them. 

If a customer’s AI can identify itself, pass through the organisation’s authentication and verification processes and successfully represent the customer’s intent, what exactly is the organisation trying to prevent? 

I think we’ll ultimately need to accept that these interactions are coming. The challenge is deciding how they’re handled. 

For organisations already using AI within their customer service operation, there is a significant opportunity here. If an AI caller reaches an AI-powered entry point, the interaction can be handled in a very different way. The conversation no longer needs the same level of rapport-building, clarification or conversational structure that humans naturally use. Instead, the focus can shift to intent, information, authorisation and outcome. 

That creates opportunities for faster interactions, fewer errors and more consistent handling. It also provides more control. 

An organisation might decide that certain activities can be completed immediately, whilst others require additional verification from the customer through another channel. It may choose to limit the actions available through an AI-to-AI route or create specific processes designed for these interactions. For example, balance checking is an approved use-case, account closure is not.  

In many ways, that feels like the most logical evolution of customer service. 

The alternative is considerably more complicated. 

If a customer’s AI reaches a human advisor, the contact centre suddenly faces a new set of challenges. Human advisors already operate within carefully defined processes, quality frameworks and compliance requirements. They have access to systems, customer data and the ability to make changes on behalf of your customers. 

Adding customer-owned AI agents into that environment introduces another layer of complexity. 

How should advisors interact with them? How much information should they provide? 

What happens if the AI asks questions differently from a human customer? 

What happens when the AI can pass every verification step that the organisation presents? 

These aren’t purely technology questions. Instead, they’re operational questions which I know very few have planned for.  In fact, I think they may eventually require a new set of skills within the contact centre itself. 

We’ve spent the last few years teaching people how to interact with AI, think about all those prompt training videos or posts. Soon, we may need to teach advisors how to interact effectively with AI acting on behalf of customers. 

The organisations that prepare for this now will be in a much stronger position than those who assume it’s a problem for the future. 

The reality is that customers use whatever channel makes life easier. If an AI assistant can book an appointment, make a reservation or resolve a straightforward service request faster than the customer can themselves, many customers will naturally choose that route. 

We’re already seeing this start to emerge. Google has spent years demonstrating the concept of AI acting on behalf of consumers. Initially it felt like an interesting technology demonstration. Today, with Gemini, Android and the broader adoption of AI assistants, it feels considerably more relevant. 

What interested me recently wasn’t the technology itself. It was Google’s decision to make parts of that capability something businesses can choose to opt into. 

On one hand, that makes perfect sense. It gives organisations control over whether they want to participate. On the other, it raises an interesting question. 

How many organisations can realistically afford not to? 

If a customer increasingly expects their AI assistant to handle routine interactions on their behalf, refusing to support that behaviour may eventually become the equivalent of not supporting a digital channel. It doesn’t stop the customer needing the service. It simply makes your organisation harder to do business with than the alternatives. 

In fact, we’ve already seen examples of organisations taking a cautious approach. Financial institutions such as JP Morgan have restricted customers from using AI agents to access account information on their behalf. I understand why. Questions around authority, accountability, liability and fraud remain largely unresolved, and few organisations have developed the policies and operational processes needed to deal with those situations consistently. 

In many ways, those restrictions are understandable. I just don’t think they’re permanent. 

Most organisations aren’t blocking AI agents because they’ve concluded there is no value in them. They’re blocking them because they haven’t yet worked out how they should operate in a world where they exist. This is a very different problem. 

The history of customer service is full of examples where consumers adopted new behaviours faster than organisations adapted to support them. Online banking, mobile apps, digital chat and social messaging all followed a similar pattern.  

At first they were viewed as new channels to control. Eventually they became channels customers simply expected to be available. 

I suspect customer-owned AI will follow a similar path. 

The contact centre industry’s challenge isn’t deciding whether that future arrives. It’s deciding what experience those AI callers encounter when they do. For me, that’s why this isn’t really a discussion about detecting bots, instead it’s a discussion about service design. 

Of course, identifying an AI caller is only the beginning of the conversation. Once an organisation accepts that AI may be acting on behalf of a customer, a much bigger set of questions emerge. What authority should that AI have? What actions should it be allowed to complete? How should policies evolve? And what roles should human and AI agents play in those decisions? 

That’s a discussion worthy of its own article…