AI Investment and the ROI Gap: A Business Model Problem

The Research Question

Why do enterprises continue to report an AI investment-to-ROI gap even when the technology is performing as designed? Is the gap a measurement problem, a value-capture problem, or a business model problem?

Key Findings

• McKinsey Global Institute research finds that approximately 50% of enterprises have adopted AI in at least one business function, but fewer than 20% can demonstrate measurable productivity gains at the enterprise level — suggesting the ROI problem is upstream of implementation.
• Deloitte Digital research (2025) finds that 62% of CFOs report no reliable method for quantifying AI investment returns, despite increased AI spend year-on-year.
• BCG analysis identifies "business model mismatch" as the primary cause of AI programme underperformance in 58% of cases studied: the AI performs correctly but the revenue model, pricing structure, or cost allocation method cannot capture the value it generates.
• OECD data on enterprise productivity divergence shows that top-performing firms in digital intensity are pulling away from the median, while the median remains flat — consistent with the BCG finding that value capture, not technology performance, is the differentiating variable.

The Analysis

The AI ROI gap is not primarily a technology failure. The pattern in the data, AI performing correctly while enterprise performance stays flat, points to a business model failure. A recommendation engine that improves customer selection accuracy generates no measurable ROI if the pricing model cannot respond to improved selection, or if the cost savings from better selection accrue to a department that is not included in the return calculation. The D1 (Digital Economy) lens frames this precisely: the economic logic of digital assets differs from physical assets because digital value compounds across use cases rather than depreciating with use. Capturing that compound value requires revenue models and governance structures designed for digital economics, not adaptations of physical-asset accounting.

Limitations and Caveats