Most AI demos use a clean PDF. Real businesses run on carbon copies, faded thermal receipts, and delivery notes filled in by hand on a dusty counter. So let us test the messy version on purpose.
The test#
A delivery note. Handwritten. A carbon copy, so the ink is grey and light. Photo taken quickly, in bad lighting, at an angle - exactly how it happens when the lorry is waiting. The kind of image you would assume no computer could read.
The prompt:
This is a photo of a handwritten delivery note. Read it and give me the
items as a clean table: item name, quantity, and any notes. If a word or
number is unclear, mark it [UNCLEAR] instead of guessing.What comes back#
In a few seconds: a clean, typed table. Item names spelled out, quantities
lined up, the supplier and date pulled from the top. A couple of smudged figures
marked [UNCLEAR] rather than guessed - which is exactly what you want it to do.
The thing that used to mean squinting and re-typing at the end of the day is done before the driver has left.
Why this matters#
This is stage 2 doing something that surprises people: reading your actual paper, not just neat digital files. It means the delivery notes, receipts, and handwritten records your business already generates can become searchable data without anyone re-typing them.
It is also, quietly, a bridge back down to stage 0. If your records were stuck on paper, this is one of the cheapest ways to start getting them digital - one photo at a time.
For where this fits on the climb, see the five stages of AI adoption.
