Voice AI Adoption: What I’m Seeing From the Contact Center Floor

For years, whenever a customer asked me about AI in their contact center, the conversation followed the same script. Chatbots? Sure, let’s talk. Email triage? Easy. But the moment we got to voice, you could feel the room go quiet.

I understood why. 

Voice is where your most frustrated customers show up. Patience is low, and the stakes are high. Nobody wanted to be the person who put a clumsy bot between an upset caller and the help they phoned in for.

That hesitation is fading, and faster than I expected. Across the thousands of businesses that run their contact centers on our platform, the question has changed from “should we pilot voice AI next year?” to “how quickly can we go live?” 

This week, a new EY India report backed up what I’ve been seeing on the ground with hard numbers, and I want to share a few of them along with my take.

What the EY Report Found

The report is called Demystifying AI in customer service. It draws on more than 100 survey responses and interviews with customer service leaders, collected between late 2025 and early 2026. EY sorted organizations into three groups based on how much value they’re getting from AI in customer service:

• Achievers (23%) report high or very high value.
• Challengers (33%) report moderate value.
• Aspirants (44%) report low value, or haven’t measured it at all.

The gap between the groups is wide. Achievers saw improvements of 25% to 50% or more across several KPIs, and more than 50% on at least three or four core metrics like customer experience, average handle time, and agent productivity. Challengers mostly landed between 10% and 25%. Aspirants saw less than 10% when they measured anything.

Here’s the number that made me sit up. According to EY’s announcement, 91% of Achievers have scaled voice AI deployments. These are live systems handling real calls, not experiments in a sandbox. Meanwhile, more than 80% of Challengers and Aspirants still rely on human agents or basic menu-driven IVRs just to figure out why a customer is calling.

Sit with that for a second. The organizations getting the most out of AI are the ones that took on the hardest channel. I don’t think that’s a coincidence.

Why Voice Took Longer

Chat got AI first because text is forgiving. A customer types, the bot reads, and if the reply takes two seconds, nobody minds.

Voice gives you no such room. People interrupt. They pause, change their mind halfway through a sentence, and switch between English and Tamil or Hindi without thinking twice. A half-second delay on a phone call feels like an awkward silence. And the older speech recognition engines, built on keyword matching, fell apart the moment a caller went off script.

That’s how we ended up with menu trees. “Press 1 for billing” was never a design anyone loved. It was simply the only thing that worked reliably.

What changed over the last couple of years is that the models finally understand intent instead of hunting for keywords. A caller can say “I’ve been charged twice and I’m honestly fed up,” and the system picks up both the problem and the mood. Response times have dropped to the point where conversations flow. And the synthetic voices no longer sound like a 2009 car GPS.

Even so, most organizations are still mid-journey. Among EY’s Achievers, 45% use rules-based voice bots, while 37% have moved on to advanced bots that use AI to classify and prioritize calls. The leaders haven’t finished either. They just started earlier.

What I See Across Our Customers

When I look at the customers who are getting voice AI right, they rarely start with a grand transformation plan. They start with one painful, high-volume problem and solve it well. Usually it’s one of these.

Replacing the menu with a question. Instead of “press 1, press 2,” the caller hears “tell me what you need help with.” This sounds small. It isn’t. When routing is based on what the customer actually said, transfers drop and agents stop starting every call with “sorry, let me move you to the right team.”

Handling the repetitive calls end to end. Order status, account balance, appointment changes, document status. These calls are predictable; they come in by the thousands, and customers don’t want to wait in a queue. A voice agent can close them out in under a minute, at 2 a.m. on a Sunday.

Outbound conversations. Voice AI works well for payment reminders, appointment confirmations, and feedback calls. EY found that 36% of Achievers already use conversational AI for surveys, while 75% to 90% of the other groups still send link-based surveys. Anyone who has looked at their link-survey response rates knows why that matters.

Helping the human agent. This one gets less attention than it deserves. Real-time suggestions during a call, automatic summaries afterwards, and instant access to the right policy document make every agent better. EY found that 45% of Achievers use AI-assisted agents during diagnosis and resolution, and 36% use agentic AI copilots.

And here’s what I tell every customer: voice AI isn’t for every call. A customer disputing a large charge, or calling about a bereavement, needs a person. The best deployments I’ve seen know exactly where that line is.

What the Top Performers Do Differently

The EY report and my own experience agree on one big point. The winners didn’t win because they bought better technology. They won because they did the unglamorous work first.

They aim at customer experience, not just cost. 45% of Achievers named customer experience as their main goal for AI, ahead of revenue and cost. I’ve seen what happens when cost is the only target. The bot gets designed to keep callers away from humans, and customers can feel it within ten seconds.

They fix their knowledge before they launch. A voice agent is only as good as the answers it can reach. 73% of Achievers use AI-powered dynamic knowledge, compared with 20% of Challengers and 8% of Aspirants. If your FAQs are outdated and your policies live in someone’s inbox, no model will save you.

They design the handoff. The bot needs to know when to step aside, and the agent needs the full context when it does. Customers forgive a bot that hands them over quickly. They don’t forgive having to repeat themselves.

They treat governance as a design input. More than 40% of Challengers and Aspirants cite governance, compliance, and change management as their key barriers. Consent, call recording rules and data handling are far easier to build in at the start than to bolt on later.

Where This Goes Next

The next step is voice agents that act on what they hear: reschedule the delivery, process the refund, update the address. That’s already starting, and it will widen the gap between the Achievers and everyone else.

If you’re still running a press-1 IVR, my advice is simple. Pick one high-volume call type. Clean up the knowledge behind it. Decide how you’ll measure success before you launch, then expand from there. The organizations in EY’s top group didn’t get there in one leap, and you won’t either.

I’d love to hear where you are on this. What’s holding your voice AI plans back?