1.5 million contacts a year: Cyta's service chief shows the unglamorous machinery around the AI

Cyta has published a detailed account of how it actually handles 1.5 million customer contacts a year, and the interesting thing about it is how little of it is artificial intelligence. Spyros Christofides, the company's Customer Experience Director, framed the philosophy as using “technology to make service simpler for the customer and empower our people to serve them better” — and then, unusually for a telecom talking about AI, produced the operational numbers underneath it.

The figures are worth reading in full, because each one is a design decision rather than a metric:

• 30,000 customers a year use the callback feature during peak periods — i.e. 30,000 people who were not made to hold.
• 8,000 remote diagnostic sessions a year run through customers sharing their phone camera with an agent, so a fault gets seen instead of described.
• 50 support videos have been watched 485,000 times.
• A 1,250-page internal knowledge base is consulted 750,000 times a year by staff.
• 130,000 calls were analysed in the quality-assessment pilot built with CYENS Centre of Excellence.

That last line connects to the Greek-language model CYENS built for Cyta's contact centre. The distinction Christofides draws is the one that matters: the AI is pointed at operator assistance and call-quality analysis — helping and grading the humans — rather than deployed as a customer-facing chatbot standing between the caller and a person. For a company whose customer base includes the exact population that CYENS and Cyta are separately teaching to use a smartphone through Digital Kin, that is a coherent choice rather than a timid one.

Why a hackathon audience should care. Read the list again as a ranking of what solved the most problems per euro. A 1,250-page wiki and 50 how-to videos absorbed well over a million interactions between them. The camera-sharing diagnostic — effectively a WhatsApp video call with a workflow around it — removed 8,000 site visits or misdiagnoses. None of those would win a demo day. The AI component, the only part that would, is the one confined to an internal quality pilot.

This is the single most common failure mode in fintech and service-tech hackathon projects: teams build a conversational agent for a problem whose real fix is a searchable knowledge base, a callback queue and a way for the customer to show you the broken thing. Cyta has published, with numbers, the order in which those interventions actually pay. Anyone heading to BoC Fintech Hackathon 7.0 in November with a customer-service concept could do worse than treat this article as the baseline to beat.

The scepticism to keep. These are the company's own figures, presented by the executive responsible for them, in an interview rather than an audited report. There is no published first-contact resolution rate, no average wait time, and no complaint volume — the three numbers that would tell you whether 1.5 million contacts a year is good service at scale or a lot of people having to call twice. “30,000 used callback” is also a figure that grows when queues are long. The architecture described is sound; the outcome data is not in evidence.

Source