The AI Value Realization Gap: the gap your dashboard doesn't show

Four numbers from 2025 tell the story:

• 88% of organizations already use AI in at least one business function (McKinsey, State of AI).
• Only ~6% are high performers: they attribute ≥5% of their EBIT to AI (McKinsey).
• 95% of GenAI pilots show no measurable P&L impact (MIT NANDA, The GenAI Divide).
• 42% abandoned most of their AI initiatives in 2025, up from 17% in 2024 (S&P Global).

Massive adoption. Scarce value. That distance has a name.

What it is

AI doesn't create value directly. It creates intelligence potential.

The AI Value Realization Gap is the distance between the value AI makes technically possible and the value the organization actually materializes. In the working paper it has a precise form:

V = α · P · Γ · Φ

Where V is realized value; P, AI's technical potential; Γ, how close the firm is to its industry's frontier (what Dean Barr's AITG measures); and Φ, organizational conversion efficiency, which is what AVRT models. The gap is the fraction that doesn't convert: (1 − Φ).

Two firms with the same P and the same Γ diverge because of Φ. And Φ isn't technology: it's capabilities, governance, and leadership. Complements, not substitutes.

How it's measured

Φ is observed in the conversion chain. An illustrative example:

A bank's AI generates 200 risk alerts per month.

• 80 change a real decision → DCR = 40%
• 56 are executed as actions → ACR = 70%
• 28 produce measurable value → VCR = 50%

End-to-end conversion: 0.40 × 0.70 × 0.50 ≈ 14%. 86% of the potential evaporated, not because of the model, but because of decisions, execution, and measurement.

That is the gap. And that's why the dashboard doesn't show it: we measure users, prompts, and pilots. Activity, not conversion.

What's next

Each metric in this series instruments one component of the equation: ALDI = 100 · (1 − Φ), DCR, ACR, VCR, and AIRR. Part 3 covers the AI Value Chain, the theory's central framework.

If tomorrow you were asked to quantify what fraction of your AI's potential turned into results, would you have an answer?

Sources

McKinsey, The State of AI (Nov 2025; survey Jun–Jul 2025, n>1,900) · MIT NANDA, The GenAI Divide: State of AI in Business 2025 · S&P Global Market Intelligence (Mar 2025). Methodologies differ and MIT's 95% has been debated; what is robust is the convergent pattern: high adoption, low value capture. The bank example is illustrative.