Why AI Economics Needs To Be Included In The Equation
Alec Scott is a Principal at CDW for Intelligent Platforms.
gettyThe FinOps Foundation’s 2026 “State of FinOps” report shows that 98% of FinOps practitioners now include managing AI spend among their responsibilities. But cost visibility, allocation and determining value are common challenges when it comes to AI usage.
The operating model I use for enterprise AI starts with assigning the CFO three responsibilities: cost attribution, return expectations and scale economics. That changes the focus from how much the platform cost to what each capability cost, who benefited and whether it’s worth scaling.
Best practice already calls for AI agents to have identities so that roles, permissions, ownership and accountability can be assigned and governed. I believe the same discipline should extend to financial management.
Before an agent or AI service reaches production, require a record that identifies its owner, business workflow, funding source and the business unit or cost center responsible for its use. If one capability supports several functions, define how its shared cost will be allocated before deployment.
Start by establishing an enterprise-wide financial taxonomy for AI spend. Finance and technology should agree on categories before creating tags, including model consumption, agent or application services, infrastructure and compute, data and retrieval services, platforms, software licenses and implementation or operating expenses. The taxonomy should also include attributions such as AI capability, business unit, cost center, workflow, funding owner and provider.
Tags then apply that taxonomy to financial records. Carry them through purchasing, cloud billing, vendor usage, invoices, accounts payable, internal allocations and financial reporting. Include the agent or capability ID so finance can trace spend to what consumed it.
Finance can’t reliably reconstruct attribution after a consolidated invoice arrives. Without a common taxonomy, one business unit may classify the same platform as cloud, another as software and another as professional services. If spend isn’t classified and tagged upstream, the invoice shows what you paid, not what consumed it or who owns the cost.
Tokens, API calls and GPU hours matter to engineers. They don’t tell a business leader what was accomplished.
Define the output the business is buying. It might be an incident resolved, an invoice processed, a claim reviewed, a case closed or a proposal drafted. Then, baseline the existing process using cycle time, labor effort, rework, quality and current cost per outcome.
The FinOps Foundation describes this as “use case economics,” or “the total cost of achieving a specific business outcome.” I use a related measure I call “return on compute” (ROC): How much business value is produced for the compute consumed? ROC doesn’t replace ROI. It gives operating teams a measure they can monitor between investment reviews.
Financial tags and transaction IDs solve two different problems. Financial tags answer: Who consumed the money? Transaction IDs answer: What business work did that consumption produce?
If an AI service resolves an incident, its runtime record should carry the agent or service ID and the incident number. If it processes an invoice, it should carry the invoice or workflow transaction ID. The same principle applies to claims, cases, requests, orders or other units of work.
That transaction ID creates the bridge from AI activity to the business outcome. AI telemetry shows what was consumed. Financial tags connect consumption to spend. The transaction ID connects activity to the work record. The business system shows whether the work succeeded, how long it took and whether it required rework.
Together, those records move finance beyond allocation and into unit economics: cost per resolved incident, processed invoice, closed case or cycle-time improvement per dollar of AI consumption.
AI spend is variable, so a quarterly invoice review isn’t enough. I recommend a monthly operating review that looks at cost, outcome and trend together.
For each material AI capability, review total spend, unit cost, business outcome, volume, forecasted spend and thresholds that require action. If unit cost rises without a gain in quality, speed or output, investigate. If unit cost falls while outcomes hold or improve, there may be a case to scale.
This creates a better management discussion. Leaders can decide whether to expand, change, constrain or retire an AI capability based on evidence rather than enthusiasm.
The goal isn’t better accounting for its own sake. It’s to give leaders the evidence to decide what to scale, what to change and what to stop. That’s the difference between having AI bills and having an AI cost-management capability.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
