Contract workflow exception mapping.
What happens when a working piece of software meets the one case nobody wrote down.
Initial condition
We had built a standard software contract generator, not an AI tool, for a local masonry contractor. It worked exactly as designed: it produced a clean contract using the payment terms the business uses on almost every job, part of the money up front, the rest when the work is done. Then a small repair job came through. On jobs like that, this contractor does not take anything up front; the full amount is due when the work is finished. The generator was not broken. It simply had never been told that arrangement existed. It knew the rule. Nobody had ever taught it the word "unless."
Constraints
The system needed to keep handling the normal case automatically, without hard-coding every exception one at a time as it came up, and without ever letting the software guess its way through a situation it had no business deciding on its own.
What was changed
Instead of trying to make the software cleverer, we mapped the business's actual exceptions first, using a simple four-column exercise we call an Unless Map: the normal rule, the "unless" that breaks it, who decides, and what happens next. The system now runs three paths instead of one: the normal path (handled automatically), the known-exception path (handled automatically once someone has spelled out the rule, like the small-repair-job case above), and a stop-and-ask path for anything that does not match a known pattern, which hands off to a named person instead of guessing.
Tools used
The Unless Map exercise itself, applied against the business's existing contract-generation tool. No new AI platform or vendor was introduced to run the mapping exercise.
What was human-reviewed
Every exception, and every condition that should trigger a stop-and-ask handoff, was defined by the business owner and estimator directly, not inferred by the system. The goal was never to make the software exercise judgment it doesn't have. It was to help the system recognize when a person's judgment is needed, and know exactly who to hand that off to.
Result
The contract generator now runs the normal path automatically, applies known exceptions like the small-repair-job case above without anyone re-explaining them, and hands off anything unfamiliar to a named person instead of guessing.
Limitations
This does not remove the need for a person in the loop. It only makes clear, in advance, exactly when that person needs to be involved, instead of finding out after a contract has already gone out wrong.