Darren Ford, our CEO, is just back from Vegas, and here he shares his reflections on CCW and how contact centre AI is “getting serious’.

 

I recently attended CCW Las Vegas, one of the major events for the contact centre and customer experience industry. As you would expect from an event in Vegas, it was big, busy, hot, and full of energy.

But once you looked past the scale of the event, one thing stood out: the contact centre AI market is becoming increasingly noisy. There were plenty of vendors, platforms, BPOs, consultancies, startups, and technology providers doing interesting work, but from the exhibition floor it was often difficult to tell them apart. Similar strap-lines, similar introductions, and similar promises around automation, agent assist, voice, analytics, and customer experience.

Why were AI vendors hard to tell apart at CCW Las Vegas?

As you would expect, AI was everywhere. There were BPOs talking about how AI can support human agents and create more hybrid service models, conversational AI platforms showing how businesses can automate customer interactions, smaller AI startups focused on automation, and broader technology vendors positioning AI as part of a wider contact centre transformation story.

One thing that stood out was the focus on voice. Over the last few years, a lot of attention in conversational AI has been on web chatbots, messaging, and digital self-service. At CCW, voice felt much more prominent than I expected.

That makes sense. The contact centre is still heavily voice-led for many organisations, and voice remains one of the highest-cost and highest-friction parts of customer service. But voice also raises the stakes. A poor chatbot interaction can be frustrating, while a poor voice experience can feel much more personal, immediate, and damaging. When customers are speaking to an automated system, often because they need help quickly, the design, reliability, testing, and governance of that experience really matter.

Why does vendor fit matter as much as functionality?

One of the most interesting comments I heard came at the end of a presentation from an organisation that had been through its own technology selection process. The speaker was asked what advice he would give to other organisations looking for contact centre technology, and his answer was refreshingly practical.

He said that when you look around the exhibition floor, you can see a large number of companies that, at least on the surface, appear to be doing similar things. His advice was not to spend too much time early on trying to compare every feature in isolation, because that can quickly become confusing.

The better questions were not just about features, but about fit:

  • Does the vendor understand your business?
  • Are they the right fit for your organisation?
  • Will you matter to them as a customer?
  • Do you want a SaaS product, a managed service, a services-led partner, or something in between?
  • Do you want to build internal capability, or outsource more of the operation?

That point really resonated with me. Technology matters, of course, and so do features, architecture, security, integrations, scalability, and cost. But for many organisations, success in contact centre AI also depends on the working relationship, operating model, level of support, and whether the solution fits the organisation’s culture, capability, and ambition.

Some organisations want a large transformation partner. Some want a specialist team with deep subject matter expertise. Some want to build their own capability, while others want more hands-on support.

This is where the idea of a “white glove service” becomes important. Not services in the narrow sense of professional services days or implementation hours, but a tailored approach where the provider understands the customer’s goals, operating model, constraints, and internal capability, then shapes the technology, delivery model, and ongoing support around what will maximise business outcomes. That point became even clearer when comparing the conversations being had by organisations at different stages of their AI journey.

From exploration to experience

One clear pattern at CCW was the difference between organisations still exploring AI and those that already had experience from previous implementations.

For those still exploring, the event must have been difficult to navigate. When every vendor uses similar language, it can be hard to know what is genuinely different.

The more experienced organisations were having very different conversations. They were asking:

  • How do we measure whether the system is actually performing well?
  • How do we prioritise the right use cases?
  • How do we test changes before they go live?
  • How do we regression test probabilistic models?
  • How do we keep automated conversations aligned with brand, policy, and governance standards?

Those are much better questions, and they are also a sign that the market is maturing. The first wave of AI conversation is often about possibility: what can we automate, improve, reduce, or launch? The next wave is about operational control: what is working, what is failing, where customers are getting stuck, what needs to change, and how do we know whether the change made things better?

The irony of human conversation

There was also something quite ironic about the event. At The CAI Company, much of our work is focused on helping organisations automate conversations between humans and machines. We spend a lot of time thinking about how AI can improve customer conversations, reduce effort, and make service more efficient.

And yet, one of the best parts of CCW was the quality of the human conversations.

The conversations that really stood out were not always the scheduled presentations or polished demos. They were the informal conversations over coffee, at lunch, walking between sessions, or at events afterwards. Many were not purely business-focused either. A 30-minute conversation might include family, travel, previous roles, how people got into the industry, current challenges, and lessons learned the hard way.

That was refreshing, and it built a different kind of connection. It was also a useful reminder that not every conversation should be automated, shortened, or optimised away. Some conversations are valuable precisely because they are human, unstructured, and allowed to go somewhere unexpected.

How do you really measure contact centre AI performance?

One of my biggest takeaways was how much the more mature conversations focused on measurement. Not just reporting, dashboards, or headline metrics, but a real understanding of conversational performance.

It is useful to know how many conversations were handled by an AI system, but that does not tell you whether they were handled well. It is useful to know your containment rate, but that does not tell you whether customers received the right outcome. It is useful to track fallback rates, but that does not always explain the root cause of failure.

Traditional contact centre metrics can tell you volume, duration, escalation, abandonment, CSAT, or cost. But they do not always tell you where the customer became confused, whether the answer was appropriate, whether the conversation followed policy, or whether a specific piece of automation is creating avoidable demand.

As AI becomes more deeply embedded in customer service, organisations will need better ways to understand conversational performance continuously, not just through occasional reviews, small samples, or customer complaints.

Why are testing and governance now essential?

Another important theme was testing, especially as more organisations start using probabilistic models in customer-facing environments.

In traditional software, regression testing is already a critical discipline. With conversational AI, especially systems that use generative AI, this becomes more complex. The same user question may not always produce the same response, small prompt changes can have wider behavioural effects, new content can affect answers in unexpected ways, and a model update can alter tone, accuracy, or reliability.

That does not mean organisations should avoid these technologies. But it does mean they need stronger testing practices. They need to know whether the system still behaves correctly after changes, whether it has been tested against real customer language, whether improvements in one area create problems in another, and whether teams can move quickly without increasing risk.

That is where governance becomes so important. Automated conversations are now part of the customer experience. They represent the brand, influence trust, create or reduce operational cost, and can either support compliance or introduce risk.

Governance cannot be an afterthought. It needs to cover what the system is allowed to say, how it should behave, when it should escalate, how it uses data, how performance is measured, and how issues are identified and corrected.

Final thoughts

CCW Las Vegas reinforced something I have believed for a long time: successful contact centre AI is not just about choosing the most impressive technology. It is about asking the right questions, selecting the right use cases, designing the right experience, measuring the right things, and creating the right operating model to improve continuously.

The market is full of possibility, but also full of noise. The organisations that succeed will be the ones that move beyond broad AI claims and focus on practical delivery, measurable outcomes, testing, governance, and customer experience.

The next phase of contact centre AI will not be defined by the loudest AI message, but by the organisations that can make AI work reliably, responsibly, and measurably in the real world.