How do you actually measure ROI on an AI investment before it's too late to course-correct?

We've approved budget for three AI projects this year. Finance wants hard numbers by Q3 and I'm not confident we'll have them. The expected benefits — analyst time savings, faster contract review — are real but diffuse. How do companies build an ROI framework before committing to AI investments, and how do you know early enough that a project is or isn't working?