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 before anyone connects the pattern back to a specific step.
The difficulty is mostly one of volume and format, since daily reports are written in free text by different people on different rigs, NPT codes are often broad and lessons learned files tend to describe what happened without saying which part of the procedure it relates to. An engineer preparing a revision would need to read months of reports and match each relevant entry to a step by hand, which is rarely realistic alongside the rest of their work.
The better approach is to tie that data back to the specific steps it relates to, so the evidence sits where the decision gets made. If NPT keeps showing up around one operation, such as tripping through a particular formation or running casing past a known tight spot, that step gets flagged for review with the reports attached as evidence, instead of waiting for the next revision cycle. The engineer sees the step, the wells where the issue occurred and what the reports said, which makes it much easier to judge whether the procedure is at fault.
Not every repeated issue means the procedure is wrong, and part of the review is separating problems the procedure could prevent from ones it could not. A stuck pipe event caused by an unexpected formation is different from one caused by a step that does not call for enough circulation time, and only the second is a clear reason to change the procedure, although the first might still justify adding a note or a check.
AI can do the matching across a large volume of reports quickly, and an engineer still decides what actually changes in the procedure. Over time this turns the reports a team already writes into a steady source of procedure improvements, so a lesson from one well has a real chance of reaching the next one.
