Consultant spend analysis

A contractor's monthly invoices kept exceeding an informal cap, and each time the response was another email restating the cap. Nobody had actually looked at what the hours were for. I wanted to answer that with data before recommending anything.

Extracted every line from twelve monthly work summaries, coded each to a deal or an account, and traced three real deals start to finish to see where the time actually went. The answer was that setup was under two hours per deal and the rest was follow-up with outside underwriters in four different formats. That pointed at a process fix, not a cap.

The recommendation was a flat monthly fee against a defined scope, priced under both the trailing average and the current pace, with a formal re-look six months out once the proposal engine had a full season of data behind it.

This is the kind of thing AI is good for that doesn't look like software. The extraction and deal tracing would have taken a week by hand. It took an afternoon, and the memo went out with actuals instead of impressions.