EY Launching Unit to Keep a Lid on Artificial Intelligence Costs

EY is hiring a head of “agent economics” to oversee the growing ranks of its AI-powered workforce, in a sign that white-collar companies are increasing scrutiny on whether their investments in AI technology are justifying the huge cost.

The executive will lead the firm’s new AI value realization office to give leadership “end-to-end visibility of where and how our investments in AI are driving material and tangible impact in our business,” said Dan Diasio, EY global consulting AI leader.

The move comes as many companies struggle to extract value from their AI investments. A June survey by consultancy firm Bain & Co. found most companies had saved less money than expected from AI, while last year an MIT report said 95% of corporate AI pilots fail.

Meanwhile, a growing number of tech firms have moved away from a flat fee for AI tools to a usage model, meaning that companies can rack up large bills for using their computing power.

Uber Technologies Inc. and Walmart Inc. have both introduced caps on employee AI use to keep a lid on expenses, Bloomberg previously reported.

EY’s global vice chair for consulting, Errol Gardner, said the new unit was needed to help manage AI agents and to try to ensure the firm’s AI rollout was not siloed by business function or division.

“You can’t expect the HR or the talent function to manage an agentic workforce, and you can’t expect tech to manage an agentic workforce, and you probably can’t expect your finance function to manage an agentic workforce,” Gardner said, adding the firm was also encouraging its clients to adopt similar models of oversight.

The new head of agent economics will report to the firm’s global managing partner for business enablement, Anthony Caterino, Diasio added.

In April, EY set AI token budgets for staff in areas of heavy use such as software development, Diasio said. The budgets are intended to make sure workers are using the right level of AI model for the job at hand. If staff exceed their budget for tokens — as the basic unit of measurement for AI computing is known — they have to seek approval for more.

The firm has also built AI routers into some of its internal tools to direct the user to the most appropriate model for the job to help save on token costs, Diasio said.

“We are thoughtful in our business about where and how we need frontier capability,” Diasio added, referring to the most advanced AI models available.