Learnings from the 2025 Contact Center AI Meetup

Last week on January 14th, 2025, we hosted the Contact Center AI Association (CCAIA) meetup, and I left feeling incredibly optimistic about where voice AI is headed. I also felt grounded in what it actually takes to deliver real value in production today.

One moment that really stuck with me came from the owner of a Texas steakhouse chain. While onboarding AI agents to his call center, he played his platform provider 10 calls from his best human agent and asked a simple question: “Can you make your AI sound like this?” He was skeptical at first, but was blown away by the results. He originally only expected voice AI to handle around 40% of calls, but instead, it ended up handling 87%. 

And this wasn’t by replacing human agents. It was actually by replacing open job reqs he was struggling to fill anyway. With AI taking care of repetitive, basic inquiries, his human agents were freed up to focus on reservations and revenue, which cut human call center agent attrition in half, saving significant time and money on hiring and training. 

As he put it: while 70% of people may say they prefer a live agent to an AI system, 100% of people just want their issue solved now. Focus on that, and you’ll win.

That theme of speed over ideology came up repeatedly. People want choice, but they also want immediacy. One large property manager shared that they always give callers an option: press 1 to talk to an AI agent now, or press 2 to wait for a human. As latency improved, adoption followed fast. They went from just 3% choosing the AI to 25% and growing rapidly, simply because the experience finally felt fast enough to trust. And CSAT increased too.

Latency, in general, was a major throughline. Several founders talked about how expectations have compressed over time, from 8–12 second delays a few years ago to now well under 1 second. Clearing that bar is no longer a “nice to have.” It’s table stakes for voice AI to feel usable, let alone delightful.

There was also a lot of honesty about how hard this is to deploy well with current infrastructure options. The voice AI space is moving quickly, from something that still feels like walkie-talkies toward fully natural, full-duplex conversations, but real-world implementations require deep workflow integration across CRMs, POS systems, and internal tooling. Multiple speakers emphasized that to launch quickly and successfully, you need strong leadership buy-in and very clear goals.

A BPO-focused consultant put it best: the most important question isn’t “How do we add AI?” but “What are we actually trying to accomplish?” AI shouldn’t introduce complexity for its own sake. It should make the customer experience simpler and better. That means looking holistically across the business, beyond a single metric like CSAT or NPS, to things like revenue impact, repeat purchase behavior, and operational efficiency.

We also talked a lot about monitoring and trust. Companies are increasingly moving toward evaluating 100% of calls with AI-powered evals, not just manually spot-checking samples. Getting ahead of bad experiences by quickly surfacing them in monitoring systems before they escalate matters enormously. Negative interactions have an outsized downstream impact. 

Plus with evolving privacy regulations, transparency is critical. It’s no longer sufficient to say a call is “monitored for quality.” Customers are required to know when a third-party AI vendor is involved.

What excited me most is how much the conversation has matured. This is no longer just about cost savings or automation. Voice AI is becoming a lever for better customer experience (CX), lower agent attrition, higher agent effectiveness through real-time guidance, and ultimately, revenue growth.

Huge thank you to John Walter and the Contact Center AI Association for bringing together such a thoughtful, candid group of builders and operators. Grateful to everyone who shared their experiences and excited for what we’re all building next!