American Accents (and AI) Could Save India’s Outsourcing

(Bloomberg Opinion) — For Tomato.ai and other California-based startups trying to crack the code of real-time accent conversion, the old “you like tomayto, I like tomahto” is no reason to call the whole customer-service thing off. If done right, accent neutralization could have significant potential in India — a country in the grip of an employment crisis.

The outlook for computer software, which has led the country’s world-beating growth in services exports for two decades, is highly uncertain. Global retailers and manufacturers could pause large IT projects to weigh the impact of US tariffs. Banks and asset managers may also cut back on technology spending as they assess the impact on financial flows. Artificial intelligence is a further threat to the code-writing skills of a young workforce. Yet, in this bleak environment for white-collar jobs, AI could be a boon for the rest of the outsourcing industry.

France’s Teleperformance SE, the world’s largest call-center operator, is rolling out an AI system that softens English-speaking non-native accents to make them more understandable. TP, as it is now known, has 90,000 employees in India. Its strategic partnership with Palo Alto-based Sanas is serious enough for TP to play two audio clips on its earnings call. In them, Indian and Filipino agents can be heard speaking in their own voices. The AI layer neutralizes their accents for American listeners.

Promoting an obvious, and ubiquitous, form of social discrimination is loaded with ethical questions. But what’s a practical alternative? Maybe President Donald Trump’s inward-looking policies will one day end America’s exceptionalism. Until then, it may pay to please — or at least not annoy — customers in the world’s biggest market. A softening of accents could help many young people who may not be otherwise considered for international call-center jobs. That’s true not only for India and the Philippines, but also across Africa, where TP has 50,000 people.

AI models developed by the likes of Tomato.ai, Sanas and Krisp can be trained to handle French and Spanish, too. (All languages have their prejudiced native speakers.)

In India, though, the need is more immediate. Some 2.5 million jobseekers, including PhDs, MBAs and lawyers, applied recently for fewer than 54,000 government positions — as general orderlies. Even trained engineers are struggling because software companies have all but stopped hiring. However, for the country’s back-office industry, generative AI has uses beyond replicating the American twang.

Contrast the muted share-price performance over the past year of Bengaluru-based Infosys Ltd., a large software exporter, with the near-50% jump in the shares of New York-listed Genpact Ltd., which helps multinationals streamline backend operations, such as accounting, finance, human resources, and customer care. A big chunk of Genpact’s 140,000-plus employees are in India; not all of them are tele-callers.

Infosys expects a slowdown after revenue grew 4.2% in constant-currency terms last year. Genpact’s sales expanded by 6.7%, and it’s predicting them to grow as much as 8.2% in 2025, led by initiatives like its GenAI tool for clients to handle account-payable inquiries from suppliers.

Ultimately, every automated business process may need fewer human beings. Yet in this case technology may not be an enemy of employment. Outsourcing companies are discovering that when it comes to training GenAI models, the best results come not from hiring more machine-language specialists, but exploiting the insights of the agents who have experience of fielding queries.

When a 500-seat call center can reliably train and run a process with 100 professionals and AI, it can pitch for more — and more complex — assignments, helping absorb surplus talent. Genpact had fewer than 120,000 employees when ChatGPT was released in late 2022. It has nearly 20% more now. In the same period, Infosys has shed 6% of its workforce.

That’s partly because the programming shops grew too fast during the pandemic and have had to adjust headcount to a subdued global economy. But technology is also playing a role. Clients want to deploy AI to cut costs — and not necessarily to create expensive code-writing work.

India’s society continues to place a premium on software, and politicians never seem to tire of wooing these job creators with tens of acres of free land. However, the average starting salary for programmers hasn’t improved in a long time. Yes, the per-employee revenue at software firms is $25,000 higher than at a back office, but the difference in labor productivity has narrowed in the past two decades. And now comes Trump’s trade war and uncertainty around US immigration policies. The sheen of software is wearing off.

Things are less gloomy at the top of the talent pool. Multinationals are increasingly using India for their so-called global capability centers. Or what some analysts like to call “Back Office 3.0,” a progression from harnessing cheap and abundant labor, to using the country as a satellite operation, to finally making it a core part of engineering R&D, like in the case of General Electric Co.’s aviation-engine research in Bengaluru.

But even here, tariff-related uncertainty is weighing on expansion, particularly for centers run by pharmaceutical and healthcare companies, according to a report in the Economic Times.

In the short run, it’s a return to the 1.0 version of the back office. Services have so far flown under Trump’s tariff radar, perhaps because the US has a surplus with the rest of the world. You might ask if it’s possible any more to separate goods from services — what’s an iPhone without the App Store? But as long as Trump makes a distinction between tomayto and tomahto, Indians won’t be complaining.