Introducing the Outcome Cost Ratio

The question every CFO is asking in 2026 is not "how many tokens did we consume?"

It's "what did those tokens buy us?"

These are different questions. Token consumption is an input metric. Business outcomes are what AI spend is actually for. And right now, the gap between those two questions is where most enterprise AI budgets disappear.

Why token cost is the wrong unit

Token pricing tells you the rate. It doesn't tell you the result.

A Claude Batch session that processes 10,000 invoices at $0.004 per session is dramatically more valuable than a Copilot session that produces a one-paragraph summary at $0.85. The token cost of the first is higher. The outcome value is orders of magnitude greater.

Every AI FinOps platform — including ours — has been measuring the wrong thing. We measure consumption. The CFO needs outcomes.

The Outcome Cost Ratio

OCR = Total agent session cost ÷ Verified outcomes produced

One number. Comparable across vendors. Comparable across use cases. Comparable across the five categories of outcome that enterprise AI actually produces.

Five outcome categories

Not all outcomes are the same. A resolved customer service ticket is not the same thing as a drafted legal brief. Both are valuable. Both have a cost. Neither should be measured with the same unit.

Resolution — a discrete problem is closed without human escalation. The billing model already exists in the market: Zendesk charges $1.50 per automated resolution. Intercom Fin charges $0.99. PromptKing now tracks this across every vendor — not just the ones that bill per-resolution.

Generation — a discrete artifact is created: code written, document drafted, report produced. Verified by artifact creation event. GitHub Copilot tracks accepted suggestions as a proxy. OCR makes this a first-class financial metric.

Transformation — existing data is processed, converted, or extracted: invoices parsed, records updated, forms completed. The highest-volume, lowest-per-unit-cost category. No standard taxonomy has existed until now.

Decision Support — analysis delivered to a human who then acts. The largest category by knowledge worker seat count. A CFO report, a security triage brief, a candidate screening summary. The outcome is the delivery of actionable intelligence. Nobody has defined this as a billable outcome unit. PromptKing does.

Orchestration — a sub-task is spawned and completed agent-to-agent. The newest category and the most strategically significant. When AgentForce spawns a Claude sub-agent to extract contract data and a Bedrock classifier to categorize it, that chain has a cost. Today that cost is invisible. OCR makes it attributable.

What this means for the CFO

OCR answers the board question directly.

"Our Resolution OCR is $1.20 per verified closure. Zendesk's published rate is $1.50. We're 20% more efficient than the market benchmark — and we can show it by vendor, by agent, and by month."

That is a sentence no AI FinOps platform has been able to produce before today.

Available now in PromptKing

The Outcome Cost Ratio is live in the PromptKing governance dashboard. Connect your agents, classify your outcome types, and your OCR populates automatically — rolling 30-day window, per vendor, per agent, per category.

The five outcome categories are PromptKing's contribution to the AI FinOps vocabulary. We're proposing OutcomeType, OutcomeCount, and CostPerOutcome as standard columns in a future FOCUS specification update.

The industry needed a standard. We built one.

---

PromptKing is an AI FinOps platform providing visibility, classification, and policy-definition for enterprise AI subscription spend. The Outcome Cost Ratio and five-category outcome taxonomy are original PromptKing frameworks. Ghost Seat, Underutilized, Normal, Power User, and At-Risk seat archetypes are patent pending (USPTO Application No. 2-01025019).