What: CFOs and CIOs must measure return on AI by workflow outcomes, not token consumption or AI activity alone.

Why it is important: Token spending is already attracting CEO and board-level attention, putting pressure on finance and technology leaders to explain where AI costs are going and what value they produce.

As AI usage scales across functions and agentic workflows multiply, the token bills CFOs, CIOs, and technology leaders must manage are rising quickly. Boston Consulting Group argues that token costs can no longer remain buried in IT budgets or governed through standard FinOps practices, because AI is spreading across software development, sales, marketing, operations, and customer-facing products. The article introduces return on AI, or RoAI, as a way to compare economic return against the combined cost of human intelligence and tokens. BCG identifies four forces that can drive token spending sharply upward in agentic workflows: broader and deeper adoption, increasing task intensity, the volume of context and loops that agents carry across steps, and default use of frontier models regardless of task complexity. The key metric is not total AI spend, but cost per successful outcome at the required quality, speed, risk level, and level of human review.BCG recommends a workflow-level operating model built around three capabilities: seeing what is happening through budget-owner and outcome-level attribution, shaping costs through model routing, caching, and employee training in token discipline, and governing return by proving value — or stopping and minimising — workflows that generate activity without measurable business outcomes.

IADS Notes: Token consumption must be treated as a financial variable tied to workflow-level outcomes — a finding that runs consistently through recent analysis of how companies scale AI successfully. The companies seeing the strongest returns are those that connect AI use cases to measurable financial impact, supported by stronger data foundations, governance, and workforce upskilling. Token costs behave as an operating variable: they scale with workflow complexity, compound in agentic systems, and can directly affect gross margins in AI-enabled products — which is why finance and technology leadership must own them together. The rise of smaller, domain-specific AI models supports the argument that routing tasks to the right model improves efficiency and avoids unnecessary frontier-model use. The spread of AI agents across business functions makes token discipline, clear ownership, and human oversight the deciding factors in whether AI investment generates return or just cost.

How CFOs and CIOs can manage the token meter