Can a customer's AI call a contact centre?
The short answer
Yes, and it already works. In a Neos Wave field test in March 2026, an AI agent acting for a real customer called a UK retailer's live support line, passed identity verification, and got a missing-delivery case logged and escalated, with follow-up promised in 24–48 hours. The human agent worked it as a routine call.
What the test involved
A real customer, a real order and a real problem: part of an order left on the doorstep and then missing, another part delivered to the wrong address. The AI was given the customer's details, the order context and a brief to resolve the issue, but no script. Across three calls, averaging just under two minutes each, it provided the order number, name, contact details and address on request and answered the agent's questions in conversational English.
The test ran in a live commercial environment with the customer's consent for the call and the recording, on infrastructure that meets UK GDPR.
What we didn't expect
The AI set the temperature of the call, and the human agent met it there. The agents matched its calm, even pace: no rising emotion, no defensive posture. The calls were also more efficient than a comparable human-to-human call, with every piece of information given on the first ask.
What it means for your operation
The capability is here: a competent human agent works the call as a normal call, which means most operations can't tell when a customer's AI is on the line. The everyday version is already routine too: customers drafting chats and complaints with ChatGPT, Claude and Gemini before they reach you. None of it shows up in standard CX analytics.
A call where neither side has declared what the AI is authorised to do works fine until it doesn't. The customer-side AI in this test was built on the Service Handshake; the contact centre had no equivalent.
Common questions
Did the contact centre agent realise it was an AI?
No. The agent ran standard identity verification and case handling, and the interaction proceeded exactly as a routine customer service call would.
Is one test enough to draw conclusions?
One recording is an anecdote, not a benchmark. What it shows is narrower: the capability is here, the interaction can be calmer rather than worse, and the interaction layer needs a standard.
How do I prepare my contact centre?
Give your team a protocol for when a customer's AI calls, check delegated authority before acting on it, and publish what AI agents may do with you. Dual CX sets out the steps.