The conversation repeats almost word for word. Someone wants to use AI in the business, looks at the state of the documents, and concludes that the tidying has to come first. The tidying never happens, because it is dull work with no deadline and no owner, and the project stays where it is.
Why the mess is not the problem
A classic system needs clean data because it works with columns and fixed rules: one missing field and it stops. A model works with text. A table with headings split over two rows, a quote written in the body of an email and a scanned contract are all, to it, the same thing: text it can read and pull facts out of.
An old quote
from email
A price spreadsheet
in three versions
A procedure
written two years ago
A long email
with everything you agreed
One folder, then a summary written by AI
- What we sell
- What our prices are
- How we work
What actually happens when you gather it all
The summary the model writes is not interesting because it reads well. It is interesting because the first time you read it, you will find three or four things that do not line up: a price that differs between two quotes, a procedure nobody follows any more, a client logged against two different people. That is the tidying you needed, obtained as a side effect.
You do not gather the documents to make them tidy. You gather them to find out where they disagree.
Terms used here
- Unstructured data
- Emails, documents, notes, scans — anything not in columns. It is exactly the kind of input a model handles well.
- Operating summary
- The short page stating what you sell, at what price and how you work, written from the documents you already have.
Where to start
- 1Pick one area — quoting, say — and put everything related to it in one folder, unsorted.
- 2Ask for a summary in three sections: what we sell, what our prices are, how we work.
- 3Mark what is wrong in that summary. The list of corrections is your first tidy document.
