Voice AI is only useful if it actually solves your problem

A natural-sounding voice is not what makes voice AI worth buying. That is the main argument in a new piece on Unite.AI, written from the point of view of the IT services company DXC. The value, it says, is whether the AI agent removes friction, keeps context and actually resolves the problem.

What changed

The piece describes three phases in how companies buy AI. First came a rush to adopt it for its own sake. Then leaders were asked what they got in return, and often could not say. Now buyers want proof of results, and proof that getting those results did not put anything critical at risk.

Gartner, a research firm, expects agentic AI to resolve 80% of common customer service issues on its own by 2029. Agentic AI means software that takes steps toward a goal instead of only answering questions. The author says the exact number matters less than the word resolve.

A test on an earnings call

DXC tried voice AI in a very public place. Over three quarters in a row, CEO Raul Fernandez wrote his prepared remarks for the company's earnings call. A custom AI voice model, built with ElevenLabs, delivered them in his voice.

• First call: showed the basic capability.
• Second call: remarks were also shared in Spanish, French, German, Portuguese and Arabic.
• Third call: the script went out in six languages at once.

Each time, Fernandez told listeners that he wrote the words and that an AI model was delivering them.

Where voice fits and where it does not

The author is clear that voice does not replace everything. Screens and written text are still better for comparing dense information and checking precise details. People are still needed when a situation is unclear, sensitive or high stakes.

The suggested approach is to use voice where talking is the natural way to get something done, and to design a clear handoff to a person when it is not. That takes connected data, clear limits and someone accountable for what happens next.

What is not known yet

The piece says four lessons came out of the earnings call test, but the text provided does not list them. It also mentions three practical areas in a partnership with ElevenLabs without naming them. Examples include insurers handling routine claims intake and manufacturers giving field technicians hands-free access to procedures. No results data is given.