The people who can run AI customer contact at enterprise scale cannot be hired off the shelf. The work combines customer service with telephony, AI, heavy data back-ends and the compliance demands of banks, and no labour market produces people who understand all of it. So we grew them ourselves, project by project, and we would rather have ten real experts than a hundred beginners.

This is a note about how we think about our team, not a recruitment advert. It matters to clients because the people behind a managed service decide what happens on the day something goes wrong, and on all the ordinary days when results could be better.

Five disciplines in one phone call#

A single automated call to a customer in early delinquency looks simple from the outside: the phone rings, a voice introduces itself as AI, verifies who it is speaking to, talks about an overdue instalment and records a promise to pay. Behind that call sit five disciplines that normally live in different departments, or different companies.

What one call depends on
DisciplineWhat it coversWhat goes wrong without it
Customer service and experienceTone, pace, wording, timing, what the customer already agreed last timePeople hang up, or agree to things they will not do
TelephonyNumbers, SIP trunks, call attempts per second, routing, transfers, voicemail detectionCalls never connect, or numbers get flagged as spam
AI and dialogueSpeech recognition, intent understanding, dialogue states, voicesThe system mishears, misunderstands or says the wrong thing
Data back-endsFile and API exchange with the client, hundreds of millions of logs to store, rotate, analyse and deleteWrong records called, results lost, reports that cannot be trusted
ComplianceGDPR roles, AI Act transparency, DORA on the client’s side, audits, information securityThe client cannot use the service at all

Simplified; based on Vee’s managed-service scope.

The difficulty is not that each discipline is hard, although each is. It is that they interact. A change in calling schedule affects telephony capacity. A new verification step changes the dialogue, the data exchange and the compliance documentation at once. A drop in results might be a recognition problem, a number-reputation problem, a database problem or a change in the client’s own systems. Someone has to be able to look at all of it together.

Why the market does not supply these people#

Look at who works in this field today. The outsourced call-centre model is built on customer service and operations, and its core expertise is managing people. The voice-agent platform model is built on models, voices and latency, and its focus is the technology rather than the business effect of a specific process. Integrators know systems. Banks know compliance. Each holds part of the picture.

We are, in a real sense, building this profession as we go. When we started in 2012 there was no such thing as an experienced designer of automated collections dialogues, or a telecoms engineer who also understood why a bank’s compliance team cares about which number a customer sees. Those people had to be made. We made them on shared experience: the same projects, the same calls, the same mistakes and the same fixes, across every phase of automation since then.

The same applies to advice. Clients increasingly ask us not only to run a process but to help them think about it: what the AI Act actually requires of a given configuration, what it means for a component to be auditable, where AI can be used safely and where a deterministic design is the wiser choice. Answering those questions well requires people who have built and operated such processes, not people who have read about them. Nobody trains for that role in advance. It comes from years inside real projects.

How we grew them#

We have specialisations, but not silos. Every project has a tandem on our side: a business owner responsible for the profitability of the process and the wider partnership, and an operations owner responsible for stages, schedule and the team. Around them work dialogue designers, the voice studio, telecoms and integration engineers, a quality team, analysts and an incident team working to agreed service levels. Each person is expected to be deep in one or two areas and to understand the others well enough to see how their work affects them.

Above all, everyone is expected to understand the client’s business: what success and failure mean in this process, why a bank cares about a debtor drifting into a later bucket, why a clinic cares about an empty slot more than a confirmed one. An engineer who knows only that a call connected is not finished. An engineer who knows why the call matters will notice when something looks technically fine and is commercially wrong.

New colleagues learn the way the rest of us did: inside a live project, next to people who have run similar ones. They follow a process through the whole delivery path. They sit in the first diagnostic conversations, where we review the client’s goals, database and current results and listen to recordings of the client’s own calls. They see the three-part specification of dialogue, integration and telecoms take shape, the tests run first on single elements and then end to end, and the launch on a sample rather than the whole database. By the time a process reaches steady production, they have seen why each decision was made. Listening is part of that training.

Every day, in every project, our people listen to selected calls, especially unusual ones, to hear with their own ears what technology still cannot. – Dawid Wójcicki, Founder & CEO of Vee

That habit does more for a new colleague than any manual. Within weeks they hear what a customer sounds like when the verification question is badly timed, when a pause feels like a dropped line, or when a recognition error was understandable given the noise. They start to connect what they build with what the customer experiences.

Why ten experts beat a hundred beginners#

In a traditional contact centre, growth means hiring. More volume needs more agents, more team leaders, more trainers. In our model the conversations are executed by Vee’s AI; Vee employs no call agents. Volume is carried by infrastructure. What the team does is design, listen, analyse and improve, and that work does not get better with more beginners. It gets better with more judgement.

We don’t manage people, so we control the service far better: one change and every live conversation already reflects it. – Dawid Wójcicki, Founder & CEO of Vee

That leverage cuts both ways. One good decision improves every conversation at once; one careless decision damages every conversation at once. The cost of inexperience is therefore much higher than in a team of agents, where a weak agent affects only their own calls. This is why we would rather have a small number of experienced people than a large number of new ones, and why we look at a person’s contribution in terms of what they add that tools cannot.

We use AI heavily in our own daily work: for analysis, for drafting, for summarising unusual situations. Those tools make an expert faster. They do not turn a beginner into an expert, because knowing which question to ask the tool, and recognising when its answer is wrong for a regulated process, is the expertise. Tools amplify judgement; they do not supply it.

48,626,959Cases handled by Vee’s AIA case is a record assigned for handling, not a person or a call.Cumulative to 31 October 2026, Vee operating data. The experience behind the team’s judgement.

What this means for clients#

For a client, the practical consequence is that the people at the weekly meeting can answer the question in the room. When a bank’s collections lead asks why reach dropped last Thursday, the answer might involve a number being flagged by a mobile network, a change in the bank’s file export and a public holiday in one region. Someone on our side has to understand all three and say what we will change, and who must approve it.

  • Fewer hand-offs. The person who designed the dialogue knows how the data arrives and how the telephony behaves.
  • Faster diagnosis. Problems that cross disciplines are recognised as one problem, not three tickets in three queues.
  • Advice, not only execution. Experienced people can tell a client honestly when a process will not pay back, or when a different bucket is the better place to start.
  • Compliance conversations in the client’s language. Banks audit us on information security, GDPR, the AI Act and DORA; the people answering understand both the regulation and the call.

What to ask about a supplier’s team#

When you evaluate a provider of automated customer contact, the team matters as much as the technology. These questions usually reveal the difference quickly:

  1. Who will be at our weekly meeting, and can they answer telephony, data and compliance questions without taking them away?
  2. Who listens to our calls after launch, how often, and what happens with what they hear?
  3. Which of your people have run a process like ours in a regulated sector, and for how long?
  4. If results fall, who diagnoses the cause across dialogue, telephony and data, and how quickly?
  5. Who on your side understands our business outcome, not only the call?

Good technology is now widely available. The people who can make it serve customers, inside the rules of a bank or an insurer, are not. That is the part of our service we have spent the longest building, and the part clients notice most once a process is live.

Follow The First Second

Occasional issues on the mechanics and economics of AI-powered customer contact. No noise, no vendor news.

Follow via RSS