A scorecard for the AI age

2026-06-28 · Source: OpenAI News · Field: Business & Management — Corporate Strategy & Leadership, Operations & Process Management, Project & Product Management · Depth: Intermediate, medium

Summary

A new "scorecard for the AI age" proposes "Useful Intelligence per Dollar" as the primary metric for CFOs to evaluate AI investments, moving beyond traditional software adoption metrics. This framework, introduced on July 17, 2026, addresses four key questions: the quantity of useful work completed, the cost per successful task, the dependability of AI results, and the value growth per AI dollar as usage scales. The article highlights that a lower cost per token does not always equate to a lower cost per outcome, emphasizing the full cost of a successful task including human review and retries. It cites GPT-5.6, released last week with Sol, Terra, and Luna tiers, noting GPT-5.6 Sol's 54% reduction in output tokens on the Artificial Analysis Coding Agent Index. The framework also stresses the importance of dependability, tracking "Ready to use," "Needs correction," and "Needs escalation" outcomes, and the compounding gains from improved compute and models.

Key takeaway

For CFOs and AI/ML Directors evaluating AI investments, shift your focus from cost per token to "Useful Intelligence per Dollar." This means assessing the full cost of successful outcomes, including human review and retries, against the value created. Implement metrics tracking useful work, cost per successful task, and dependability to ensure your AI spend genuinely reduces effort and scales value, rather than just optimizing raw compute.

Key insights

The true value of AI is measured by "Useful Intelligence per Dollar," focusing on successful outcomes over raw compute cost.

Principles

• AI value is determined by work accomplished, not just adoption or token cost.
• Dependability (accuracy, consistency) directly reduces rework and increases confidence.
• Economic gains compound as better models and infrastructure drive adoption and investment.

Method

The proposed method for calculating "Useful Intelligence per Dollar" involves adding the full cost of completing work, counting tasks meeting quality, and dividing total cost by successful tasks.

In practice

• Define "done" for specific workflows (e.g., customer issue resolved).
• Track outcomes: "Ready to use," "Needs correction," "Needs escalation."
• Establish clear boundaries for AI access, systems, and human review.

Topics

• AI Value Measurement
• Useful Intelligence per Dollar
• AI Cost Optimization
• AI Model Dependability
• Workflow Automation
• GPT-5.6

Best for: CTO, VP of Engineering/Data, AI Product Manager, Executive, Director of AI/ML, Consultant

Related on AIssential

• Intelligence Per Dollar — AI model evaluation is shifting from raw performance to cost-efficiency, prioritizing "intelligence per dollar."
• The Outcome Density Scorecard: Measuring AI Value Beyond Hours Saved — AI value is best measured by "outcome density," delivering more valuable results with less organizational friction.
• How to Measure AI Value — AI value extends beyond efficiency, encompassing augmentation and innovation to create new capabilities and improve human performance.
• AI to ROI Metric of the Week: Cost Reduction — AI-driven cost reduction directly impacts profitability and competitiveness, freeing resources for strategic growth.
• AI economics reshape FinOps as enterprises seek greater visibility and control — AI's rapid adoption necessitates FinOps evolution, integrating cost accountability into code-level decisions and development workflows for optimal value.
• Expensive AI? Really? — AI's perceived expense often reflects an organization's failure to align technology with business problems and design for value, not the tool's inherent cost.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by OpenAI News.