Main content

Generative AI accents are coming to call centers - is this a good thing?

George Lawton Profile picture for user George Lawton June 27, 2024
Generative AI tools that soften accents in call centers may improve customer experience. But what are the long-term implications?

call centre

Sanas has introduced new tools that use generative AI to soften accents for call center employees. The new tool builds on Sanas’ previous work on AI-powered noise cancellation tech that reduces the need for alternative soundproofing measures. The company argues this can reduce linguistic bias in the call center industry. It has built models for 436 English variants in India and another 37 in the Philippines.

Preliminary research suggests it can improve customer experience and raise the rankings of knowledgeable call center agents by making conversations more intelligible. The company estimates it increases average customer satisfaction by 22%, reduces average call handling time by 18%, and opens opportunities for individuals in more rural areas.

The new tool also promises to reduce the pressure on agents sometimes hired, fired, or promoted based on their accents. It can also reduce the need for accent training to be better understood by customers.

Expanding the labor pool

The company was founded by Stanford students inspired by the troubles a fellow student from Nicaragua ran into while working as a contact center employee during a break from school. Their friend recanted tales of how customers had been rude to him. Some had given him two-star ratings even after he had quickly fixed their technical issues.

The co-founders were surprised because their friend was at the top of his class in systems engineering. But he kept getting called in by his supervisor to get coached on his accent. The co-founders were surprised because they could all understand him perfectly clearly.

That’s when they realized that this was actually a global problem and opportunity. Anant Singh, Sanas VP of Global Go-to-Market, explains:

We learned that there's an industry that, on average, rejects about 35 to 50% of applicants, not because they don't have the talent but because of their accent mismatch, because they have a lot of mother tongue influences in their accent. This prevents them from getting a really good paying job.

Singh stresses that it was important to soften the accent while keeping the agent's identity and uniqueness intact, which is what makes the conversation feel human. This is different from simply generating speech. He explains:

No conversation that you and I will ever have will be perfect. But a machine will always be perfect when speaking from speech to text and text to speech. However, the reason why human conversations are great is because they are imperfect. There are a lot of fillers. There's a lot of emotions. There's a lot of intent in what you're saying. How you say ‘hello’ literally could change the way you talk to the person or the person perceives you.

My take

The tech also raises troubling questions about the potential to increase systemic bias. My aspirational side imagines that the occasional encounter with a new accent helps broaden my horizons. Perhaps losing occasional exposure to new accents might gloss over an experience of living in a larger and more diverse world by glossing over cultural differences.

However, Singh argues that accent-softening technology has the potential to create a much clearer separation between work and home culture. He spent many years working in a call center environment and lost his native accent as a result. While it made it easier to get hired and promoted, his mother now jokes about how it’s harder to understand his accent. Speaking in a flawless American accent, Singh explains:

I never used to speak like that. Do you think I was born speaking like Americans? No. I had very good English but with an Indian accent. I worked in a contact center environment for four years. My accent is bleached. You asked me to speak in an Indian accent today. But I can't do it. What happens is we try to fit ourselves into a specific mold and lose that part of our culture.  What we wanted to do was to ensure that people were able to preserve that part of the culture and not lose it.

Another realization is that most of us are already frustrated when calling for customer support. The process of understanding a different accent may just add a bit more frustration to the experience. Perhaps there is some value in reducing this frustration, even if it comes at the cost of appreciating linguistic diversity.

Singh said the service is being offered to customer service agents on an opt-in basis. Aside from raising their customer service ratings, it also reduces the rate of abuse from customers. This also helps minimize call center churn. Singh recounts a story from one town hall meeting Sanas held with agents:

One lady, I remember she said, ‘I know nobody talked about this, but they should be. It's been 76 days without abuse.’ The first image I had in my mind was like a construction site with a sign: ‘There has been ten days since an incident or an accident.’ And agents track that because, at the end of the day, they're humans. The agents know they are yelling at the company and use that as a forcefield to protect themselves, but they are the ones listening to the calls. And at some point, everyone breaks. That is why this industry has one of the highest turnover rates of any industry in the world. By some estimates, 36% churns every quarter.

Shielding agents from customer frustration may turn out to be another use for generative AI tech. Softbank is working on technology to make angry customers sound calmer on the phone. Only time will tell if this makes it more challenging to appreciate the severity of the customer’s issue as well.

A grey colored placeholder image