The Voice AI Agents market is growing exponentially. The industry is projected to grow from $2.4 billion in 2024 to $47.5 billion by 2034, while Gartner predicts that by 2028, 70% of customer service journeys will begin with a conversational AI interface.
For global brands, the appeal is obvious. Voice AI offers the potential to deliver consistent, scalable customer support across markets, languages and time zones.
But there is a risk that voice AI will face the same issue faced by outsourced customer service centers, where differences in language and culture create a disconnect and frustration between the brand and customer.
As voice AI advances, language may become less of a barrier. But speaking the same language is not the same as understanding someone.
It’s just one part of the puzzle. Customers bring accents, habits and cultural expectations to every interaction, and they expect support to reflect their reality.
For that reason, localization needs to become a core test of voice AI performance, rather than a translation exercise that happens at the end of development.
What does it mean for voice AI to be ‘local’?
Accents, dialects, terminology and cultural norms can vary considerable even when speaking the same language. In Scotland, for example, a customer might say “aye”, rather than “yes”, refer to something small as “wee”, or talk about “getting the messages” when they mean going shopping. The language is English but understanding the interaction requires familiarity with how that language is used locally.
Cultures also differ in their expectations around directness or politeness. Even something as simple as how a customer is addressed can matter. In some cultures, using titles and surnames is an important sign of respect, while in others, addressing someone by their first name is the norm. A voice agent that is too informal could come across overly familiar, while one that is too formal might feel distant or unnatural.
All this to say that just because an AI might sound like me, it doesn’t necessarily mean it understands me or is equipped to serve someone in my market.
These aren’t cosmetic details. They shape an AI agent’s ability to understand intent and respond appropriately and, ultimately, determine whether customers trust the experience it provides.
It doesn’t sound like a robot anymore
This isn’t to say there haven’t been huge advances in AI. There’s a reason why so many brands are investing in voice AI for CX, and that’s because it’s becoming so convincing. Just look at ElevenLabs, for example.
The organization has built its success on AI-generated voices that can reproduce not just speech, but the pacing, intonation and emotion that make a voice sound human. Its technology not only spans voice cloning, multilingual speech and conversational agents, helping take synthetic voice from a novelty to something businesses can realistically deploy at scale.
The days when synthetic speech was instantly recognizable as robotic are fading. Early digital voices such as ‘Microsoft Sam’ left little doubt that you were listening to a machine. Today, the line is becoming much harder to distinguish. As voice AI becomes more convincing, our interactions with it can start to feel less like using a piece of technology and more like engaging with another person.
The technology’s ability to create a sense of social presence is already apparent beyond customer service. Even actors such as Matthew McConaughey are getting behind AI voice technology, where it can match their performance without requiring them to change the way they sound, demonstrating just how powerful voice and conversational cues can be in shaping our perception.
But the most advanced experiences aren’t universally available, meaning many people may still be exposed to underdeveloped AI voices. What’s more, humans are remarkably good at noticing when something is slightly off. This is what creates the uncanny valley effect.
An unnatural pause, misplaced emphasis or strangely enthusiastic response can quickly break the illusion. Local context adds another layer to that challenge. A voice can sound perfectly human while still sounding culturally out of place. That said, the closer AI gets to human speech, the more conspicuous those moments can become.
Global brands can’t rely on one global answer
This is where global brands need to resist the temptation to treat voice AI as a single product that can be rolled out everywhere.
A frontier model might provide the foundation, but its performance needs to be tested against local data and real customer interactions so it can capture what ‘good’ looks like in terms of:
– The empathy that de-escalates a frustrated customer
– The specific terminology a market actually uses
– The intonation that reads as natural
All of these can’t be inferred by a frontier model; it has to be learned market by market, brand by brand. Successful voice AI models will have an ongoing feedback loop which monitors interactions within individual markets, identifies where misunderstanding or friction occurs, refines the experience and tests again.
Local teams can play an active role in this process as well, bringing market expertise into development rather than simply receiving the “finished product”.
It can’t be treated as a job that is ever truly finished. Language and slang changes, customer expectations shift and new trends introduce expressions and behaviors that may not have appeared in the original training data.
Final summary
Getting this wrong can have an impact on a brand’s reputation. As voice AI becomes more human, customers will naturally expect more from it.
Derby City Council recently experienced this with its AI assistant, Darcie, which struggled to understand a presenter with a strong Derbyshire accent who used local expressions such as “mardy” and “duck,” despite having been upgraded to support nine additional languages.
It is a useful reminder that multilingual does not necessarily mean localized. If a system repeatedly misunderstands an accent or fails to recognize local expressions and expectations, customers will not experience it as a technology problem.
Instead, it will shape their perception of the organization behind it. The brands that get real value out of the voice AI boom won’t be the ones whose technology can speak all 7,170 languages in existence today. They will be those whose customers feel the most understood.
We’ve ranked the best translation software.
This article was produced as part ofTechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.
The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here:https://www.techradar.com/pro/perspectives-how-to-submit
