← Case studies · The other side of the audit
Sometimes the answer is: not yet.
Every audit maps where AI is risky or premature, not just where it pays. If the data is messy, automation just makes the mess faster. These are real calls from real audits, names removed.
Automotive & finance
Keep typing the payroll in by hand.
Regulated finance advice, so the AI never gives the advice: a licensed broker stays accountable for every application. And commission payroll gets entered manually after a human reviews the calculation. No automated write, until months of verified accuracy earn it.
Construction & trades
Don't automate on top of a system that melts down.
Weekly reporting ran on a custom backend that was unstable during live projects. More automation would have made the outages worse, not better. The recommendation: retire and rebuild the foundation first. And site-safety documents keep a logged human approval as a design constraint, not an optional extra.
Wholesale & distribution
Some compliance stays human. On purpose.
Ingredient compliance runs through a closed government portal with no API. The tempting build was a scraper. The recommendation was a structured manual upload and a deliberate wait, because a fragile automation on a regulatory process is a liability, not a saving.
Manufacturing
Live money earns automation. Ten runs at a time.
Anything touching live billing, payments or safety sits behind human-in-the-loop review. Invoice automation routes low-confidence lines to a person, and review gates only come off after ten successful live runs. Trust is earned in production, not promised in a proposal.
“AI shortlists and drafts. A person still approves. No tool gets handed the keys.”
The pattern, across every audit
If AI isn’t worth your money right now, the report says that.
That’s the point of paying for a diagnosis instead of a pitch: you get the map of where AI should not touch your operation, alongside where it pays. You keep both, either way.
Free 30 min · we’ll tell you if there’s nothing there