The information needed to improve a well procedure usually already exists, it is just spread across daily drilling reports, NPT logs and lessons learned files that nobody has time to read against the procedure itself. Most of it gets looked at during a scheduled revision, if at all, which means the same problem can repeat on the next few wells.
The better approach is to tie that data back to the specific steps it relates to. If NPT keeps showing up around one operation, that step gets flagged for review with the reports attached as evidence, instead of waiting for the next revision cycle. AI can do the matching across a large volume of reports quickly, and an engineer still decides what actually changes in the procedure.