Five AI Investment Types Leaders Must Understand to Capture Value

The AI ROI puzzle

Corporate leaders face a troubling disconnect: AI adoption is widespread, but measurable returns remain elusive. Recent surveys show 88% of organizations now use AI in at least one function, yet only 39% report any EBIT impact—and even then, typically less than 5%. Meanwhile, 60% of companies investing in AI generate no material value, according to BCG analysis.

The problem isn't AI itself. The problem is that executives are applying commodity-style ROI calculations to investments that create value in fundamentally different ways. Brown University professor Baba Prasad argues in Harvard Business Review that AI investments fall into five distinct categories, each requiring its own financial logic and measurement framework.

Two tactical investments for survival

The first category is competitive parity—matching capabilities competitors have already deployed. When Bank of America's virtual assistant Erica handles 58 million monthly conversations, that's impressive operationally but creates zero competitive advantage because JPMorgan Chase, Wells Fargo, and others offer similar tools. The right question isn't "What's the ROI?" but "What does falling behind cost?" Prasad recommends capping these investments at industry median spending.

The second is option value—building institutional fluency that opens future opportunities. Moderna deployed ChatGPT Enterprise to 3,000 employees in 2023, and by 2025 had created 750 custom GPTs across functions. No single tool produced breakthrough returns, but collectively they built the absorptive capacity that now enables the company to target 15 new products in five years with just 6,000 staff—work that would traditionally require 100,000 employees. These investments should be measured like R&D: by adoption velocity and organizational learning, not immediate payback.

Three strategic investments for durable advantage

The remaining three categories create competitive moats that competitors cannot easily replicate.

Unique integration embeds AI into distinctive workflows and institutional processes. Amazon's AI-driven supply chain coordinates over a million robots across 300 fulfillment centers, improving regional forecasts by 20% and enabling 9 billion same-day or next-day deliveries in 2024. Competitors with identical AI technology cannot replicate this advantage because it depends on decades of supplier relationships and operational culture. Measure these by process-level performance deltas—cycle time, defect rates, customer retention—not enterprise ROI.

Data flywheels generate proprietary data that continuously improves AI performance while creating switching costs. John Deere's See & Spray technology uses 36 cameras to identify individual weeds, reducing herbicide use by 67%. Each spraying session generates millions of data points that make the system smarter about that specific farm's conditions, making it extraordinarily costly for farmers to switch to competitors. The key metric is flywheel velocity—how fast the AI improves per operational cycle.

Organizational capability building is the most valuable and most overlooked category. Walmart equipped 1.5 million store associates with AI tools while reskilling 50,000 frontline employees into entirely new roles like drone technicians and AI agent developers. CEO Doug McMillon frames the goal as building the capacity "to change all the time, not just once." These investments create strategic agility—the ability to exploit whatever technology comes after AI.

Why it matters

Most companies allocate 70% or more of AI spending to tactical investments at the bottom of the spectrum, then measure everything with standard ROI tools that make the entire portfolio look disappointing. Strategic investments remain both underfunded and assessed with inappropriate metrics. Organizations that learn to allocate across all five types—and measure each correctly—can justify spending to stakeholders while building advantages competitors cannot replicate. The framework also explains why predictions that AI will become a commodity utility are mistaken: the technology may commoditize, but the integration, data ecosystems, and organizational capabilities it enables never will.

These findings were detailed by Baba Prasad, professor of leadership at Brown University's School of Professional Studies, in Harvard Business Review.