The best framework for evaluating AI investments isn’t ROI on AI. It’s Return on Manual Work.
Here’s why the reframe matters. ROI on AI puts the vendor’s technology at the center of the calculation. You’re comparing what AI produces against what it costs, which is how the vendor wants you to think — because their pricing is what they control, and the “productivity improvements” are marketing.
Return on Manual Work puts your business at the center. You calculate the fully-loaded cost of the manual work AI would replace. That’s your ceiling. If AI can be deployed for materially less than that ceiling and produce equivalent-or-better output, it wins. If not, it doesn’t, no matter how impressive the demo was.
The calculation is straightforward. Pick a process. Multiply the hours it consumes per week by the fully-loaded cost of the person doing it. Multiply by 52. That’s the annual cost of the manual work.
Example: An operations analyst spends four hours per week on a recurring reconciliation. At a fully-loaded cost of $50 per hour (that’s salary plus benefits, roughly, for a mid-market analyst), that reconciliation costs $10,400 per year. Any AI solution that costs less than $10,400 per year AND produces equivalent output is winning against the manual baseline.
Now the honest complications. The AI solution has to actually produce equivalent output, which requires you to have a way to check. Some processes have hidden requirements that AI can’t fully meet — the analyst was also spotting anomalies during the reconciliation that a naive AI wouldn’t catch. Some processes are low enough judgment that AI clears the bar easily; some aren’t. The Return on Manual Work framework doesn’t skip these complications; it makes them explicit by requiring you to compare against the baseline honestly.
The framework produces three useful outputs. First, a prioritization: your highest-cost manual work is where AI investment should start. Second, a decision rule: if AI costs more than the manual work, don’t do it, no matter how novel the technology. Third, a measurement standard: if AI is deployed and the manual work isn’t actually reduced, the AI didn’t succeed, whatever the vendor is claiming.
I’ve watched businesses skip this framework and end up with an AI tool bill that exceeds the cost of the labor it was supposed to save. That’s not an AI failure; that’s a math failure. The math is the framework’s whole purpose.
If your business hasn’t calculated its top ten manual-work costs, that’s the first exercise. Everything downstream — vendor selection, pilot prioritization, ROI review — depends on having those numbers.