AI Has a Return-on-Investment Problem

AI Changes the Economics of Technology

Traditional IT financial management was designed for predictable systems. You bought or built software, its costs were largely fixed, and its behaviour was stable. You could budget for it with reasonable confidence.

AI is fundamentally different. The Gartner report argues that rapidly deploying AI produces unpredictable cost, risk, and value — and that relying on legacy financial models exposes organizations to operating costs that can quietly exceed the value the AI generates.

This matches our own experience closely. We have written before about building increasingly sophisticated AI systems that consumed large amounts of compute and tokens without necessarily creating much economic value. The lesson was not to stop using AI. It was to treat cost, value, and outcome as things that must be measured continuously, not assumed.

The Hidden Risk: Value Drift

The most useful idea in the report is what Gartner calls value drift.

AI systems, the analysts note, do not simply fail. They quietly lose value over time. A model can continue to perform well technically while becoming financially unviable — because its costs rise, the conditions around it change, or the outcomes it produces slowly stop justifying the spend.

Without continuous monitoring, an organization can end up scaling systems that erode value instead of creating it — and not realize it until the bill is large and the benefit is gone.

Gartner projects that by 2028, 70% of AI initiatives will be decommissioned due to unmanaged cost explosions and value drift in organizations that lack a dedicated AI financial management practice.

Measure What Matters

The report’s prescription is a discipline it calls AI-specific financial management: continuously tracking the cost, value, and risk of non-deterministic AI systems across their full lifecycle, rather than judging them once with a static return-on-investment model.

In practice that means:

Tracking AI cost, risk, and value across the entire lifecycle rather than at a single point in time. Replacing static ROI models with dynamic financial management. Preventing value drift and controlling rising costs before they compound. And measuring success with metrics that reflect real business outcomes, not just technical performance.

Why This Matters to How We Invest

For an investor, this is not an abstract technology debate. It is a valuation and diligence question.

As AI spreads through the companies we build, acquire, and back, the winners will not simply be the ones that adopt AI the fastest. They will be the ones that can show, with discipline, that their AI spend produces more value than it consumes — and that can catch value drift before it becomes a write-off.

This is the same principle behind our own operating approach: improve the process first, measure the economics, and then use AI to amplify what works. The Gartner data simply puts a number on the cost of ignoring it.