The ROI model I use before recommending any AI tool

A vendor demo is designed to be persuasive, not to answer the only question that actually matters: does this tool's cost, over a realistic adoption timeline, get paid back by a specific, measurable business outcome. Most recommendations skip straight from demo to purchase without ever building that model.

The model I build before recommending anything is deliberately simple: current cost of the process being replaced or augmented, meaning time, error rate, opportunity cost, a realistic adoption curve, since tools rarely reach full utilization in month one, and total cost of the tool including the implementation and training time nobody puts on the invoice but everyone eventually pays.

Run honestly, this model kills a meaningful fraction of tools that looked compelling in a demo: the math often doesn't work at the business's actual scale, or the realistic adoption curve is slow enough that the payback period exceeds what the business is willing to wait.

A recommendation backed by this model is a harder sell in the room than a slick demo, and it's the version that still looks correct eighteen months later, after the demo's excitement has worn off and the invoice has come due several times.