How to measure the ROI of AI before you invest in it
Investors and boards are asking harder questions about AI spend. A simple framework separates real value from expensive experiments.

In this article
Insight prepared with AI assistance and reviewed by the Winter Kpital team.
The question has changed
In the first years of generative AI, boards asked what the company was doing with AI. Now they ask what it is getting back. Gartner has described this as the move past the peak of inflated expectations.
A three-part test
Start with a baseline: the cost, time and error rate of the process today. Define the target: what changes if AI handles part of it, and who checks the output. Price the full cost: models, integration, data work, human review and change management. Deloitte and PwC surveys repeatedly find that integration and people costs are underestimated far more than model costs.
Leading indicators
Before revenue moves, watch cycle time, throughput per employee, quality scores and adoption by the people who actually do the work. Adoption is the indicator most companies skip and the one that predicts everything else.
What investors look for
Clear unit economics per task, gross margins that survive model price changes, and evidence that customers use the product weekly, not just during the pilot.
Sources
Gartner, Hype Cycle for AI. Deloitte, State of Generative AI in the Enterprise. PwC, AI Jobs Barometer and CEO Survey.
