The fastest way to waste money on AI is to automate a process that is already broken.
Automate exception handling before asking why the exceptions exist, and this is what happens: the bot processes exceptions three times faster, exception volume triples within a quarter, because the root cause is still there, now running at machine speed. Nobody is surprised except the budget owner.
This is the FDE rule I work by: fix the process first, then decide what the agent should own. Walk the workflow. Find the step that exists because of a decision made in 2019. Kill it or fix it. Then automate what is left.
AI multiplies whatever you feed it. Feed it a clean process and you get leverage. Feed it a mess and you get a faster mess.
Most AI projects skip this step because it is not exciting. It is interviews and flowcharts. But it is the step that decides whether the project pays.
Re-engineer first. Automate second. In that order, every time.