Analyze your inspection data in Claude, ChatGPT, Copilot

The status question you answer by hand every Monday. Pick a period – day, week, month, or your own range and it compares that window against the one before it, then writes the same five parts every time: headline cards, top issues, what changed, what needs attention, and how far to trust the sample. Same shape every run, so after the first read you already know where to look.

Sites, crews, assignees, assets or work types side by side on one metric, measured against the pooled account average pooled, not an unweighted mean, so a site with five records doesn’t drag the number around. Then a second pass compares each group against its own recent history, which is how you tell a long-standing gap from something that only started slipping last month. Name two dimensions and it runs both, then says whether the same pattern spans them.

The checklist items and defect categories that keep coming back, counted across every inspection in the window instead of read one record at a time. For each one it says whether the failure is concentrated on a single unit that’s a repair or spread across many units of the same type, which points at the model or the form. Where your forms carry both a checklist and an issue category, it counts both and reports the gap: how many checklist flags never got logged as an issue at all.

Late, about to be late, and quietly stale, sorted into three tiers so work in reports with no due-date field doesn’t fall through the gap. Those get judged against how long comparable items normally sit, not an invented threshold. Every flagged item comes with its likely blocker: waiting on a signature, parked at an approval step, or plain inactivity. And when the person holding it is also carrying more than their peers, it says so and names someone with room.

Two modes. Point it at one workflow and it computes how long each step actually takes, shows what’s piled up at that step right now, and audits the form itself fields nobody fills, free text that should have been a dropdown, duplicate fields, evidence fields meant to be conditional that fire at the same rate regardless. Point it at “which forms” and it ranks every report by completeness against the weighted account average and names the dominant gap in each. Fixes come back ranked by impact against effort, with the evidence behind both labels.

Give it a date range or a site and back comes the bundle: every record in scope with its key fields, signature status, photo count and step timestamps, in a table that doesn’t truncate. The completeness check runs underneath it, and any gap is stated on the front of the pack rather than left to be found a pack that hides a problem is worse than no pack. Say it’s going to a customer or a regulator and assignee names and internal notes come out.