AI's First Measurable Return Is a Smaller Payroll

For the companies spending on AI, the most specific number of the year is a savings target, not a sales figure. Three of them, side by side: $1.5 billion a year at HSBC, £500 million at WPP, $1.5 billion at Omnicom and IPG together ^[1][2]. Every dollar on that list is a payroll dollar — money saved by paying fewer people, not money earned by selling something new. HSBC's version is the cleanest. In March the bank was weighing up to 20,000 job cuts, roughly a tenth of its workforce, to chase $1.5 billion in annualized savings from automating back-office work ^[1]. That same month it appointed its first chief AI officer, hired specifically to drive those generative-AI cost cuts toward a return-on-equity target above 17 percent ^[3]. By October the plan had reached UK wealth management, where roughly half of management and specialist roles and around 70 percent of financial advisers are going — reversing a hiring drive the bank began only two years earlier ^[4]. HSBC's chief executive, Georges Elhedery, has said on the record what generative AI means for certain jobs at his own bank.

generative AI will destroy certain jobs — Georges Elhedery

The savings line is itemized and public. The revenue line is where the search gets hard. Go looking for what these companies have earned from the spending, and you find a different number: enterprise AI spending rose 110 percent this year while a maturity score measuring whether firms can actually use what they bought climbed to just 51 out of 100 ^[5]. Operating costs for AI agents now routinely outrun what companies budgeted to build them, because more usage means more spent on model calls, data retrieval, and the people reviewing the output ^[6]. Only 23 percent of enterprises have scaled agentic AI in even one business function ^[6]. The dollars are going in; the work is not yet coming out. The cuts are not waiting for the technology. More than 99 percent of executives surveyed by Mercer expect AI to reduce headcount within two years, and 98 percent are already planning organizational changes around automation ^[7]. Gartner's finding sits awkwardly beside that: the reductions so far have not reliably improved financial results, and companies including Klarna and IBM have walked back automation decisions or rehired staff after quality slipped ^[7]. HubSpot's chief executive, cutting 660 jobs, insisted the move had nothing to do with cost pressure or AI efficiency ^[8]. GitLab's said the same about its own layoffs and its roughly 30 percent cut in country footprint ^[9]. Then September's labor number, which changes the picture. US job cuts fell 20 percent from a year earlier, and year-to-date layoffs are down 39 percent ^[10]. Yet hiring plans sit at their lowest since 2011, with the usual holiday seasonal surge simply absent ^[10]. The displacement is running through jobs never posted, not through people fired.

Hiring plans are up over the year, but we're not seeing the surge of hiring plans that come with the holiday season, which suggests a very cautious approach. — Andy Challenger

Some of what gets labeled AI is not AI. Sam Altman, OpenAI's chief executive, has put a name on the wave of AI-linked job cuts: AI washing — companies using the technology as a cover for ordinary cost-cutting and for unwinding pandemic-era over-hiring ^[11]. The economist Adam Triggs ran the arithmetic from the other side: the job losses firms attribute to AI exceed the number of roles research actually categorizes as disrupted, against revenue-per-worker figures that have barely moved ^[12]. Two independent voices, one point. The counter-case is short but real. At the top of the market the revenue exists: Google grew 24 percent year over year on enterprise AI demand ^[13], and Salesforce raised guidance on roughly $3.4 billion of AI annual recurring revenue ^[14]. Then there is Publicis, which expanded revenue 5.6 percent by training 85 percent of its client-facing staff on AI platforms while cutting only 200 jobs ^[2]. Set beside WPP's £500 million of cuts, that makes payroll-first look like a choice rather than a law. Temasek's chief investment officer put the risk plainly today: the AI trade unwinds, he said, if tighter safety regulation arrives or if customers fail to generate returns from their spending ^[15]. The second trigger is already measurable and self-published — a maturity score of 51 out of 100, and a run of savings headlines standing in for the missing revenue line ^[5]. Meanwhile UBS still projects AI capital spending approaching $1.45 trillion in 2027 ^[16]. The evidence for the trigger he named arrives quarterly, in the customers' own announcements — the next savings target, the next 51.