20 Feb How to Think About AI ROI in Large Organizations
The conversation around AI return on investment has matured considerably in the past year, and not in the direction most vendors would prefer. Enterprises are no longer asking whether AI delivers value. They are asking why their organization is not capturing more of it.
The data tells a sobering story. According to McKinsey’s State of AI 2025 survey of nearly 2,000 global leaders, 88% of organizations are using AI in at least one business function. Yet only 39% report any EBIT impact at the enterprise level, and most of those say AI accounts for less than 5% of their organization’s EBIT. Adoption is nearly universal. Enterprise-level financial impact is not.
The Measurement Problem
The most revealing number in the current landscape comes from Larridin’s State of Enterprise AI 2025 report: 89% of enterprises have adopted AI tools, but only 23% can accurately measure their return on investment. That gap is not a technology problem. It is a management problem.
Organizations that do measure AI ROI report impressive results: 27% average productivity improvement across measured use cases, 11.4 hours saved per knowledge worker per week, and $8,700 in annual efficiency gains per employee. The operative phrase is “organizations that measure.” The majority are still operating on intuition and anecdote, what Larridin calls “vibe-based AI spending.”
The Wharton Human-AI Research Center’s October 2025 report offers a more encouraging signal: 72% of business leaders now report tracking structured, business-linked ROI metrics tied to profitability, throughput, and workforce productivity. That represents a meaningful shift from the FOMO-driven investment cycles of 2023 and 2024. Accountability is becoming the operating model.
Why Enterprise-Wide ROI Is Hard to Capture
The gap between use-case-level results and enterprise-level financial impact is not an accident. It reflects something structural about how large organizations deploy AI.
Most enterprises are capturing AI value in isolated pockets: a software engineering team using code generation, a customer service team using summarization, a marketing team using content drafting. Each of those use cases may show strong local productivity gains. But those gains do not automatically aggregate into P&L performance unless the organization redesigns the workflows around them.
Deloitte’s 2026 State of AI in the Enterprise report, which surveyed over 3,200 senior leaders, found that only 34% of organizations are using AI to deeply transform their business by creating new products, reinventing core processes, or redesigning business models. The remaining two-thirds are using AI at a surface level, capturing efficiency gains without fundamentally changing how work gets done. Both groups see productivity improvements. Only the first group sees transformative financial impact.
A More Useful Framework
The organizations getting the most out of AI share a few consistent patterns, and they go beyond simply “setting goals upfront.”
The first shift is measuring outcomes, not activity. This sounds obvious but most organizations get it wrong. Tracking how many employees use an AI tool is not ROI measurement. What matters is what changed as a result. A team of five QA engineers armed with AI may not shrink to three people. But if defect escape rates drop by 40% and QA cycle times compress from two weeks to five days, that is a measurable business outcome with direct product quality and time-to-market implications. The value is real. It just requires the discipline to define those metrics before deployment, baseline them, and track them over time. Organizations that implement this kind of structured measurement report 5.2 times higher confidence in their AI investments and 3.8 times higher rates of continued investment, according to Larridin’s 2025 research.
The second shift is redesigning the workflow, not just augmenting it. Deloitte’s research draws a clear line between organizations using AI at a surface level and those using it to genuinely transform how work gets done. The former capture productivity gains. The latter capture competitive advantage. The difference is whether AI changes the structure of the work or just makes the existing structure slightly faster. Faster is good. Different is better.
The third shift is treating the skills gap as the primary constraint. The Deloitte 2026 report identifies the AI skills gap as the number one barrier to scaling, ahead of data quality, infrastructure, and governance. This means the ROI question is increasingly a talent question. Organizations that invest in building internal AI fluency, not just deploying tools and hoping for adoption, are pulling ahead. The technology is no longer the bottleneck.
The honest conclusion is this: AI ROI is not a technology problem. It is a measurement discipline, a workflow redesign problem, and a talent development problem simultaneously. The organizations treating all three seriously are already separating from the field. The window to join them is still open, but it is not as wide as it was two years ago.
Sources: McKinsey State of AI 2025; Larridin State of Enterprise AI 2025; Wharton Human-AI Research Center and GBK Collective, Accountable Acceleration: Gen AI Fast-Tracks Into the Enterprise, October 2025; Deloitte State of AI in the Enterprise 2026; Second Talent AI Adoption in Enterprise 2025
No Comments