01What happens.
An external repository contracted to hold the company's published documents accumulates unmanaged material over time. Draft versions of annual reports and manuals are uploaded beside the final versions, and nothing in the store identifies which version was approved.
A machine reading the store may find 10 versions of the same document and select among them, producing confused and inconsistent results.
02Why it matters.
The company secretary approved the final report, and the head of product approved the final manual.
The store holds those, and it also holds the 3 drafts each went through, uploaded by people who needed to share them and never removed. The landscape AI reads extends to the store, and the store is outside the company's own sites. A draft served as final is a position the company never took, in its own name, from a place it pays to keep its records.
03The case.
A consumer brand contracts a public-facing document repository to hold its published material / assets. The store is reliable and well indexed, and over time it is used by more teams than it was set up for, because it is the easiest place to put a file that needs sharing.
Product teams upload manuals as they are drafted, so that agencies and translators can work on them. Legal uploads warranty terms in draft for review. Investor relations uploads the annual report in stages before the board signs it. Once the finals are uploaded, the drafts stay, because nobody owns removing them and it is not part of anyone's process and the store never asks.
After years the store holds several versions of most documents, with names that differ by a suffix or a date, and nothing in the store's own metadata marks which version is final. The store's search returns them side by side, and so does a machine reading it.
A customer asks AI about a product's warranty term, and the answer comes from a draft that legal rejected, which stated a longer term. An investor asks about a figure in the annual report and receives the number from the second draft, before the audit adjustment. A retailer asks about a safety instruction and receives the translator's working copy. Each answer is from the company's own store, and each presents material the company never approved as final.
The company had paid for the store to be its record. What it had not done was manage the record, because the store was a service and the service worked, and a store that holds everything looks like good practice until something in it is read
04Exposure.
A draft made publicly accessible in a repository the company contracts can be served as the company's position even though it was never approved as final. Where the draft is a warranty term, it can prompt a contractual dispute; where it is an unaudited figure from the annual report, it conflicts with the company's disclosure record and can mislead investors.
The repository extends the company's landscape beyond its own sites and outside the version controls applied to its website publishing. The cost of the confusion, in complaints, corrections and investor questions, is the cost of a record kept without governance over what went into it.
05What AI Readiness changes.
The store was outside the company's sites and inside the landscape AI reads, and no one had looked at it as a machine would. The outside-in view lists what the repository holds under the company's name, which versions AI selects for each question, and where a draft is being served as final. We provides that view and evidences it.
With the list in hand, the company removes the drafts and superseded versions it controls and marks the final versions in a form the store and a machine both read. It then limits the published repository to approved final versions.
06Related use cases.
The corrected leaflet never reached the customers, because the old copies kept answering
Duplicates and untracked copies across the estate.
AI put a claim the company had withdrawn back under its name, and the dispute found it first
A draft in a repository supplying an unapproved figure.
Investors acted on results the company had already superseded
The unaudited draft reaching the disclosure record.
07References.
- Dawiso, 2025: companies connecting AI to document repositories holding 10 versions of the same document, with AI selecting at random and returning confused, inconsistent responses. Attributed and paraphrased. https://dawiso.com/blog/enterprise-ai-data-governance-guide/dawiso.comWe do not enable external links - please copy and paste, you are then certain as to the link being opened
Next step
Talk it through, in your terms.
The briefing sets this case in your organisation’s terms, and the discussion follows your questions.
Illustrative scenario, drawn from a documented pattern. No company is identified.