The big shift to autonomous agents and generative AI looks efficient on the surface, but scrutiny is sharpening around how much value they deliver. Beyond spending on software rollouts and training staff to use tools such as Anthropic’s Claude, organisations now pay for tokens, the tiny data units that meter every interaction with these models.
Each token reflects a slice of text processed, so higher usage means higher bills even when the tools feel frictionless to employees. The gap between seamless user experience and rising token consumption is rapidly becoming a boardroom topic.
Financial groups are already under pressure to explain how this plays out in practice. At Macquarie’s recent investment banking conference, Westpac’s leadership pointed to “tangible outcomes” from early AI deployments, saying the tools are not just experiments.
The bank is still working through how to measure and disclose token consumption, and how that relates to productivity gains, which shows internal dashboards and metrics are still evolving. Other large enterprises face the same challenge, they must tie abstract token usage to concrete business output.
That means understanding not only total spending on tokens but which workflows consume them and which ones actually move the needle.
