OpenAI Proposes “Useful Intelligence per Dollar” as an AI Value Scorecard

OpenAI published a business framework on July 17, 2026 that asks companies to judge AI investment by “useful intelligence per dollar.” Its central argument is that adoption metrics and token prices do not capture the outcome that matters: how much valuable work reaches an acceptable quality bar.

The proposed scorecard has four parts: useful work completed, the full cost per successful task, dependability, and whether each AI dollar produces more value as use grows. It is a company-proposed framework rather than a neutral industry standard, but it gives teams a practical way to move an AI discussion beyond activity counts.

Measure the workflow, not the prompt

The framework recommends starting with one workflow and defining what “done” means in the system where work happens. For support, that might be a customer issue resolved. For engineering, it may be a change that passes tests. For legal work, it could be an accurate, timely review.

That approach changes the cost calculation too. The price of a token is only one input. A task’s real cost includes compute, employee time, retries, review and rework. A cheaper model can be more expensive in practice if it requires several attempts; a more capable model can be better value if it gets the job right with fewer interventions.

Dependability is part of the economics

OpenAI suggests tracking whether a result is ready to use, needs correction or needs escalation to a person. The categories are simple, but they make a useful distinction between a fluent answer and a workflow result that someone can rely on.

Teams also need explicit boundaries before AI moves from drafting to action: which data it may access, which systems it may change and when a person must review or approve a step. Safety, privacy and control are not separate from return on investment; they affect how much review, remediation and confidence a workflow requires.

What teams should do next

• Choose a high-volume, well-defined workflow instead of trying to measure every AI use at once.
• Define success before comparing models or vendors.
• Track all meaningful human and system costs, not only model price.
• Separate results that are ready to use from results that needed edits or escalation.
• Re-measure over time to see whether quality holds while cost per successful task falls.

Goodiebase take

“Useful intelligence per dollar” is best understood as a decision framework, not a universal KPI. Its strongest idea is simple: teams should measure completed, acceptable work in context. If an AI tool makes activity look impressive but adds review loops, uncertainty or hidden operational cost, it has not yet proved its value.