Narawe

Situation

Professional-services firms hold years of client history in scanned paper. Finding one document meant people searching by hand for hours — so firms staffed for the searching rather than the work.

Task

• Get paper in with no friction, including straight from a phone camera.
• Make it findable by the words people half-remember and the ideas they can't name.

Result

A firm's scanned history becomes searchable by phrase and by meaning. Hours of hunting collapse into a query, and the people left are verifying results rather than looking for them.

My Contribution

• Reframed the cost from time to headcount: firms weren't only losing hours, they were staffing around a search problem. That reframing is what made the business case.
• Identified capture as the real adoption risk, not accuracy. A system nobody feeds is a system nobody uses, so scanning had to work from the phone already in someone's hand.
• Chose to fuse keyword and semantic search rather than pick one: the exact phrase on an invoice and the vague idea of a letter are different questions, and firms ask both.
• Consolidated three AI vendors into one — fewer contracts, fewer failure modes, one bill to defend.
• Made the AI summary fire-and-forget. An assistant that can fail someone's upload is worse than no assistant.