The AI Shift: The long-predicted decline in call centre jobs has finally begun

金融时报（科技），金融时报科技版块 - 来自金融时报的最新科技新闻与分析。关注

This article is an on-site version of our The AI Shift newsletter. Premium subscribers can sign up to get the newsletter delivered every Thursday. Standard subscribers can upgrade to Premium here, or explore all FT newsletters

Welcome back to the AI Shift, our weekly exploration of how artificial intelligence is changing jobs and the labour market. This week we’re taking a look at AI’s early impact on call centre employment, one of the jobs thought to be most exposed to LLMs, and an example of an AI-exposed occupation with significant representation in lower and middle-income countries, making it a particularly useful bellwether for how AI might disrupt employment patterns not only within but between countries.

John writes

The trigger for this week’s edition was a new study carried out by Caelan Wilkie-Rogers, an economist at the employment research firm Revelio Labs, whose fine-grained data on job-to-job moves has become a key underpinning of a lot of the most important work being done on AI’s impacts on the labour market.

Wilkie-Rogers crunched millions of records from firm headcounts and individual workers’ professional profiles in dozens of countries to track what has been happening to employment in call centre roles since LLMs took off, finding that employment is down five per cent globally on its late 2023 peak. Most notably, this is true not only in aggregate and in high-income countries, but in most lower- and middle-income countries too. In the Philippines there are now ten per cent fewer people employed in call centres than when ChatGPT launched, while Kenya has seen a six per cent drop in the past 18 months.

The timings of the declines in higher and lower-income countries are notable. In some richer countries these jobs were on the wane well before LLMs arrived precisely because call centre jobs were being outsourced to countries with lower labour costs. But the more recent downturn in both the global total and LMIC figures hints more strongly at a role for AI.

Call centre jobs were always earmarked as particularly vulnerable to disruption from AI due to their strong emphasis on the kind of routine and clearly specified tasks AI excels at, and the weaker links between employer and employee in outsourced roles overseas make these especially vulnerable. But it’s important to note that, as has consistently been the case in AI job displacement findings to date, Wilkie-Rogers found that the declines in employment are a story of reduced hiring, not increased firing (call centre lay-offs are actually down, just much less steeply than hires).

But equally if not more interesting than the decline in call centre headcount was the finding that if we shift our gaze to higher-paying white-collar roles in LMICs, even those theoretically heavily exposed to AI, employment here is still growing, often strongly. This looks to me like a possible sign that AI may be lowering the specialist skills threshold required for these roles, or another demonstration of the Jevons paradox in action as AI-induced efficiency gains unlock new demand.

Even within the broader customer service industry we see evidence that AI may be helping workers in LMICs move up the value chain: while employment in the most basic call centre jobs is down, it has risen for higher-paying jobs in areas like client relations. But Revelio’s job transitions data add an important nuance: the people moving into higher-value white-collar jobs in LMICs are generally not those leaving bottom-rung call centre jobs. Most people who leave call centre jobs move into similar or lower-paying roles.

To my mind, this slots into several recurring patterns. First, what seems to matter most for job displacement is not just task overlap with AI capabilities but also the status or security of an occupation. Lower paid and outsourced roles are especially exposed, just as we’ve seen with gig work. Second, even in the most vulnerable roles where we do see declining employment, we’re mainly seeing job displacement manifest as foregone hires, not lay-offs. And third, while we are seeing signs of AI displacing hiring in the most-exposed roles, the net impacts so far seem neutral to positive due to increased demand and possible benefits from lowered skill-barriers.

Sarah writes

Thanks for this, John. The fact that the margin of displacement so far seems to be via entry, rather than exit, reminds me of the case of the Victorian bootmakers we explored a few weeks ago (in sum: rather than bootmakers being turfed out of their jobs en masse when mechanisation arrived, the artisanal profession shrank because young people stopped entering it). That 19th century transition unfolded over a generation, but because staff turnover rates tend to be high in 21st century call centres, it makes sense that employers can downsize quite swiftly and quietly through attrition.

It’s another useful reminder that combing through lay-off statistics isn’t necessarily the best way to scan for signs of AI disruption (though I should note that there have been some reports of actual AI-linked job losses in contact centres recently, such as in South Africa.)

But the really big question is what fewer call centre jobs might mean for those lower and middle-income economies which have come to rely on them (in the Philippines, for example, the wider business process outsourcing sector accounted for about 7.4 per cent of GDP in 2023, similar in magnitude to remittances). A recent paper from the World Bank and the International Labour Organization outlined the worst-case scenario. Its authors pointed out that call centre jobs and other entry-level office-based roles have traditionally acted as stepping stones to better things, especially for young people. White-collar automation, therefore, risks “closing off career pathways to upward mobility before they emerge, mirroring the logic of premature deindustrialization.”

But as you say, it’s encouraging to see in the Revelio data that other white-collar roles in these economies — indeed, better-paying ones — are holding up so far. When we spoke to Krishn Kaushik, our Mumbai correspondent, he was quite optimistic about the outlook for higher-skilled job creation in India’s “global capability centres”, for example.

What we don’t know, of course, is whether those jobs can become alternative entry points for those young people no longer going into call centre work. Nor do we know whether AI might eventually come for these roles, too.

John writes, plus a note on the latest AI maths breakthroughs

Thanks Sarah, I agree with all of that. One important thing that I think comes through from this research (and echoes the US analysis by the Brookings Institution and Opportunity@Work that we covered a few months back) is that although graduate jobs and high-end knowledge work like law and software often dominate the headlines about AI job displacement, the evidence to date suggests it’s what we might call light-blue-collar jobs — highly routine office support roles requiring little to no specialist knowledge, with call centre workers being perhaps the clearest example — where we’re seeing the strongest evidence of AI-accelerated contractions in employment.

But another big AI story this week demonstrates how even extremely complex and specialist knowledge work roles can be heavily exposed to the same technology. I’m talking, of course, about OpenAI’s announcement on Tuesday that an unreleased model has solved hundreds more famous mathematical problems. After the ructions caused by its rushed and chaotic announcement of its solution to the Navier-Stokes problem last month, this time OpenAI consulted with professional maths groups, published detailed results and documentation alongside the new discoveries, and I’ve not heard many murmurings about snoop-and-scoop plagiarism of human researchers’ ongoing work on the same problems.

I strongly recommend browsing the stunned responses from career mathematicians as they pore over the solutions to problems they had been wrestling with for years — sometimes decades — now solved by AI in a few hours. I was struck by the similarities to what you heard from translators when reporting your book, Sarah: even if one’s job remains intact, it can be disorienting and demoralising to witness the shift from doing the deep, stimulating work of pursuing problems in a way you choose, to being the person who checks AI outputs. It also raises the question of who will direct and assess AI mathematics in the future if a shock like this leads skills to atrophy and talent pipelines to narrow? Rest assured — this is a topic we’ll be returning to in the weeks ahead.

The AI Shift: The long-predicted decline in call centre jobs has finally begun

John writes

Sarah writes

John writes, plus a note on the latest AI maths breakthroughs

Recommended reading:

Recommended newsletters for you