CFO Slice

McKinsey has highlighted a growing paradox in enterprise AI spending: models are becoming significantly cheaper for a given level of capability, but companies’ overall AI bills continue to rise as they use the technology for increasingly complex and intensive tasks. The firm said agentic AI is particularly difficult to budget because agents can take different approaches to the same task, with execution costs varying by as much as 30 times between runs. McKinsey recommends that companies assess AI economics at the task level, considering execution costs, success rates, and the human time required to verify outputs, rather than focusing solely on token prices. To manage spending, businesses should improve visibility into AI usage, optimize workflows by matching models to tasks and reducing unnecessary activity, and strengthen sourcing discipline. However, McKinsey cautioned against indiscriminate cost-cutting, arguing that companies should instead direct AI investment toward areas generating the strongest returns.