For years, the information about our own company sat exactly where it sits in most companies: quotes in email, prices in a spreadsheet with three versions, procedures in a few people's heads, decisions in chat threads. Nothing was missing. Nothing was in one place either.
Before
Information everywhere, across dozens of places
We gathered the documents
Into one folder, without sorting anything
We asked for the memory in writing
Clients, projects, who does what
Contradictions in our own documents
Different prices, dead procedures
Every agent starts here now
One set of pages, kept current
What we did not expect
The summary written over the documents told us nothing new about what we do. What it showed was where we disagree with ourselves: the same service quoted differently in two quotes from the same month, a procedure nobody had followed in a year, two people listed as the owner of one client. None of that was visible while each document sat on its own.
What changed afterwards
- Every new agent starts from the same pages, not from a prompt written from scratch.
- When a price changes, it changes in one place and every automation sees it.
- The “how do we do this” conversation has a written answer instead of three verbal ones.
We thought we were writing context for the AI. We were writing our procedures down for the first time.
Terms used here
- Company memory
- The short set of pages every automation starts from: clients, projects, roles, prices.
- Single source
- The rule that a fact is written in exactly one place. Without it, versions multiply on their own.
Where to start
- 1Gather everything about one process into a folder, without sorting it.
- 2Ask for a summary and read it looking for contradictions, not for bad phrasing.
- 3Resolve the contradictions in the real documents, then regenerate the summary.
