Contact centres have spent the past three years putting AI on their side of the call. The other side of the call is now doing the same.
Business Insider reported last week that TTEC, one of the largest customer service outsourcers, first spotted AI callers about eighteen months ago, when its quality team flagged voices that "sounded weird". Its advice to clients since has been to take the calls, but carefully. Much of the traffic is unglamorous: a small medical practice sending an agent to obtain insurance preauthorisation, bill-reduction services that have replaced human negotiators with software.
The same report describes an insurance claims handler who fields repeated calls from an agent working for an accident-attorney firm, and who is marked down by her employer's empathy-tracking system for failing to accommodate it. That detail deserves attention. A quality framework was scoring a person on her warmth towards a machine.
The demand is not speculative
Accenture's Consumer Pulse research, published in June and based on a survey of 25,590 consumers in 16 countries, found that 74 per cent would trust a personal AI agent with routine tasks such as negotiating a deal or resolving a complaint. Only 9 per cent would let one make purchases entirely on its own. Read the two numbers together. Customers are far readier to delegate the service conversation than the purchase, and the service conversation is where the contact centre lives.
The tools are already on sale. Google Search now calls local businesses on a user's behalf in categories such as home repair, beauty and pet care, a feature that reached every user in the United States this summer; the caller announces itself as an automated system from Google. Business Insider cites Meta's Muse among the consumer agent products making this easier.
The forecasters point the same way. In its strategic predictions for 2027 and beyond, published this month, Gartner expects more than 10 billion autonomous agents, created by people, companies and governments, to be clogging public services by the end of 2030, as they identify entitlements, check eligibility and submit requests with minimal human effort. The same list predicts that by 2030, 80 per cent of organisations with public-facing AI will suffer cost exhaustion attacks, in which bad actors inflate operating costs simply by generating usage.
Every part of the operation assumes a human on the line
Walk through a standard inbound voice journey and count the assumptions.
Authentication leans on knowledge-based questions, voice biometrics or a one-time code sent to a registered phone. An honest agent cannot answer the first, should fail the second and does not hold the third. A dishonest one may get through all three. Business Insider quotes a credit card retention specialist who believes agents are mapping her company's authentication steps, and whose employer's policy forbids her from asking whether the caller is AI.
Routing and capacity planning are built on human behaviour: arrival curves, handle times, abandonment. An agent does not abandon. It will sit on hold for as long as it takes and call back as often as its instructions allow. One customer can become ten contacts, and a workforce plan built on last year's patterns will not see it coming.
Quality and empathy scoring measures conduct towards a person. It has nothing useful to say about a caller that cannot be reassured and does not need to be, and as the claims handler found, it can actively punish staff for handling one sensibly.
Disclosure rules were written for firms deploying AI on customers. Very few say anything about customers deploying AI on firms.
Why the Gulf should care first
The UAE has committed to moving half of its government sectors, services and operations onto agentic AI within two years, with citizen-facing services as one of the programme's two tracks, as Computer Weekly reported when it was announced in April. AI News reported in August that federal officials had begun building frameworks to classify which government tasks may be handed to an agent at all. Nothing had been published at the time of that report, and we can find nothing since.
That work concerns the state's own agents. Gartner's prediction concerns everyone else's. A government that automates its service desk will also receive agent-driven traffic from citizens, businesses and intermediaries, and the private sector around it (banks, telecoms operators, utilities, airlines) will receive the same. Gulf consumers are early adopters of digital channels, so there is little reason to expect this traffic to arrive later here than in the United States.
The UAE does hold an advantage many markets lack. UAE Pass gives it a national digital identity that could, in principle, anchor delegated authority. But an identity layer for people is not a mandate layer for their agents. Nobody has yet specified how a caller proves that it acts for a named person, for which task and within what limits. Alorica's co-chief executive told Business Insider the industry may need a registration scheme along the lines of ID.me to tie a bot to a real person. Until one exists, each organisation has to write its own answer.
What we would do now
None of this requires a new platform. It requires decisions that most operations have not yet made.
- Find out whether it is already happening. TTEC noticed through quality monitoring. Most contact centres and their outsourcers have the same recordings and are not looking. Ask the question and start tagging suspected agent calls.
- Set policy by task, not by caller. Order status, appointment availability and published prices can be given to a declared agent with little risk. Changing contact details, moving money or disclosing account data should require the customer, or a verifiable mandate, back in the loop.
- Write the disclosure rule. Decide whether agents must identify themselves, whether staff may ask, and what happens when a caller will not say. Leaving front-line staff unable to ask is the worst of the options.
- Repair the scorecards. Separate agent contacts from human ones in empathy, CSAT and handle-time measures. Otherwise your quality data degrades and your people are penalised for sound judgement.
- Give agents their own door. A declared agent is better served through a structured channel, an API or self-service endpoint, than a voice queue staffed by people. It costs less per contact, and it makes abuse visible, which is the first defence against the cost exhaustion Gartner describes.
The bottom line
Contact centres have spent three years deciding how much of the conversation to hand to AI. Their customers are now making the same decision, independently and without asking permission.
The organisations that cope will treat the AI caller as a channel to design for, not an anomaly to tolerate. The ones that do not will find out from their fraud team, or from their workforce plan.