What: AI token costs have become a CEO-level financial discipline, requiring companies to measure return on AI by workflow and outcome.

Why it is important: Without workflow-level cost attribution, AI spending scales faster than its returns, compressing margins without accountability.

BCG argues that as AI moves from pilots to large-scale deployment, companies need a new way to measure the cost and value of intelligence. Token-based pricing is replacing traditional software subscription models, making AI spending harder to track as usage expands across workflows, products, and customer interactions.
The winners will not be companies with the smallest or largest AI budgets, but those with the highest return on AI. CEOs must understand where token costs sit in the P&L: as capital expenditure when building AI capabilities, operating expenditure when running internal work, and cost of goods sold when AI is embedded in customer-facing products. BCG estimates that AI inference compresses gross margins directly, with AI-enabled software margins resetting in the range of 65% to 80%, compared with 50% to 65% for AI-native products. BCG proposes measuring AI through cost per outcome, combining the cost of human intelligence and token consumption. It warns against both excessive token use and overly restrictive caps. Instead, leaders should stop unnecessary AI use, route tasks to appropriate models, cache reusable context, govern workflows with clear ownership, and train employees to distinguish high-value AI use from token waste. BCG cites Google’s DORA research, which found that AI acts as an amplifier of whatever organisational system already exists, delivering stronger gains where practices are mature and compounding problems where they are not.

IADS Notes: AI cost discipline has moved up the agenda because spending is compounding faster than most organisations can measure or justify. In June 2026, BCG and The Consumer Goods Forum found that CPG and retail companies pulling ahead on AI are those that link each use case to a measurable financial outcome. In April 2026, BCG connected operating margin expansion to AI-enabled workflow redesign, reinforcing the need to treat AI spending as a margin lever rather than a generic technology cost. In January 2026, Retail Touchpoints showed that smaller, domain-specific AI models can improve accuracy, efficiency, and customer satisfaction when matched to specific tasks. In December 2025, the Wall Street Journal reported that CEOs continue to invest in AI despite uneven returns, while Forbes reported in October 2025 that AI agents require governance, workforce adaptation, and clear oversight to deliver durable performance gains.

What CEOs need to know about the true cost of artificial intelligence