BNY has built tools to track how it spends AI tokens and to route tasks to the most suitable models, CFO Dermot McDonogh told Bloomberg Television. The bank runs about 400 AI agents and spends $4B a year on technology. Banks are under pressure as AI bills rise.

BNY has built tools to track how much it spends on AI and to send each task to the most suitable model, chief financial officer Dermot McDonogh told Bloomberg Television. The aim is to avoid overspending as AI costs rise.

“We’ve become quite disciplined and developed some sophisticated telemetry so we can track how we’re spending and how we’re using our tokens,” McDonogh said.

“We don’t use expensive models for cheap queries,” he added. He said sending tasks to the right model will matter more “over time as costs come into play.”

BNY uses models from the big cloud providers and is not tied to any one of them, McDonogh said. That holds even though BNY chief executive Robin Vince joined OpenAI’s board in July. The bank runs about 400 AI agents, and AI is only part of its $4B annual technology budget.

Companies are paying more for AI even as the price per token falls, because AI agents burn through far more tokens. In June, Commonwealth Bank of Australia’s chief executive warned of rising bills, and in July Atlassian put its engineers on monthly AI budgets.

The comments mark a change in tone for BNY. In July, McDonogh told Fortune the bank did not track token use and focused on results instead, saying, “It’s not something we spend any time talking about.”

He still does not want cost control to hold back AI use. “We don’t talk about it in relation to efficiency or job cuts,” he said. “We talk about it in relation to upskilling and trying to do better things.”

Other banks have been more open about cuts. Citi chief executive Jane Fraser said in July that AI job losses have started at the bank.