Companies are pouring money into AI, but many still cannot answer a basic question: Which employees are creating value with it?
Why it matters: CFOs interviewed by The Wall Street Journal say measuring AI ROI requires understanding who is using AI, how often, and whether that usage is translating into significant productivity gains. Without that visibility, leaders struggle to distinguish returns from rising costs.
The focus is shifting from AI adoption to AI economics. As AI pricing moves toward consumption-based models, CFOs need to see if each employee’s usage and behavior are cost efficient or not.
A recent EY study found that:
-
Reasoning models can cost up to 30x more than traditional models.
-
A single task can cost up to $1.20 today compared to what it was at $0.04 in 2023.
-
Agentic AI systems are even more expensive, requiring multiple model interactions, retrieval steps, and reasoning loops to complete a single workflow.
A related KPMG analysis found:
-
Just 26% of companies have a comprehensive understanding of their AI costs. Meanwhile, 50% have only partial, so it is difficult to connect spending with outcomes. 22% have no insight until the bill arrives.
-
Anthropic, OpenAI, Microsoft, and Salesforce now charge companies for tokens, shifting the financial risk onto the customer. Some teams are burning through annual token budgets in months.
-
40% of agentic AI projects will be canceled by the end of 2027, due to escalating costs and unclear value, according to a Gartner forecast.
The cost falls on HR: The total cost of training an agent is front-loaded with each new workflow and applies to each model upgrade. Workforce readiness is now a recurring budget item, and the retraining, role redesign, and human oversight all fall on the CHRO and COO.
Adoption is not value: One company rolled out 12 AI tools for its marketing team, then watched usage fade as employees reverted to old habits. Another company tied agent-written code to measurable engineering gains, monitored usage in near real time, and reported results to leadership weekly. This more disciplined approach proved much better.
What CHROs can do:
-
Treat workforce usage as a budget input, set a metric of success, and build from there.
-
Name a Head of Agent Economics or FinOps lead with visibility across all seven cost categories, and treat total spend as a KPI. This will help CHROs determine what the return on investment is for each employee and how AI is being used.
-
Benchmark work per task or per outcome, then partner with finance and technology on spend ceilings, volume caps, and automatic shutoffs.
-
Price the full cost upfront. Calculate the model fee with orchestration, monitoring, and human-in-the-loop costs, then weigh the total against what each agent should produce.
-
Measure value per role, not logins per team. Give every deployment an outcome target and reward AI proficiency as a performance metric, so adoption sticks.
-
Track as usage spreads. Map which roles boost productivity gains, and which teams consume the tools most. Aim for steady use across the organization rather than a few “tokenmaxxers.”
