Most AI projects stall at the pilot. One reason is that value was never measured: nobody can decide either to stop or to continue.
01. Baseline First
Measure the current state before starting: how long the process takes, how many people work on it, what the error rate is. Without that number you can't later prove improvement.
02. The Right Metric
Usage count is not a measure of value. Time per transaction, first-time-right rate, number of returned requests and workload per person are the real metrics.
03. The Cost Side
Model call cost, infrastructure, development and maintenance. Maintenance is usually forgotten; prompt updates, evaluation and model migrations take continuous effort.
04. The Limit of Time Saved
"It saves two hours a week" doesn't book as value unless those two hours turn into other work. Track where the saved time goes too.
05. Quality Impact
Some projects deliver value in quality rather than speed. A drop in error rate, fewer customer complaints and reduced compliance exposure are all measurable.
06. The Decision to Stop
Stop a project that hasn't crossed the threshold within a set period. Carrying pilots that never deliver consumes the budget and the credibility owed to the ones that do.