A CFO's quarterly review of a custom agent is a cost/ROI exercise, and the Copilot Studio / Foundry usage analytics center that review on how the agent behaves and how much it is used. The agent orchestration method (C) drives how many model calls, tool invocations, and knowledge-source retrievals occur, which directly shapes consumption and cost, while average session time per agent (D) indicates engagement and whether sessions (and the credits/tokens they consume) are being used efficiently. The specific GPT model list (A) and average characters per message (B) are low-level implementation details rather than the components a CFO assesses for the quarterly cost analysis.
Community Discussion