The document centre: who has the current version of the policy?
Version tracking, read records, category structure and access permissions in corporate document management — and why search matters more than folders.
A company's document problem is rarely "we're out of space" — it is "which one is valid". Three versions of the leave procedure sit in three folders; only the person who wrote it knows which one applies, and they are on holiday.
Why isn't a file server enough?
A file server stores documents but carries no context. There are three things it doesn't know: which version is in force, who has read it, and who it applies to. Folder structures try to compensate — filenames like "2024_v3_FINAL_approved.docx" are really an attempt to squeeze into the name what the system doesn't carry.
A document centre keeps that information in the data itself: version number, effective date, owner and read record travel with the document.
What does version tracking solve?
Two things. Trust: the employee is sure the document they opened is current. And evidence: in an audit you can answer "when did this policy take effect, and who could access it then?"
Not deleting old versions matters too. Archiving rather than deleting answers retrospective questions and makes it possible to roll back an accidental update.
Read records
Not every document needs a read record; but for safety instructions, ethics policies and privacy notices that record becomes a legal requirement. The critical distinction: the read record must be tied to the document's version. When the text is updated, earlier acknowledgements should lapse — otherwise the claim that someone was informed cannot be evidenced.
Categories or search?
Most teams that build a folder hierarchy discover six months later that nobody uses it. People don't browse; they search.
So keep the category structure plain — a single level such as HR, Finance, IT, Quality is enough for most companies — and spend the energy on search. Search has to do more than match words: someone looking for "travel expenses" should find the "business trip reimbursement procedure". Semantic search closes exactly that gap.
Access permissions
Not every document has to be open to everyone. Payroll templates, board presentations and contract drafts in progress should stay restricted. The single rule here: the restriction belongs in the data layer, not in interface hiding. If a hidden document can still be downloaded with a direct request, there is no restriction.
Time-limited download links follow the same logic: a shared link that stays valid forever makes the permission meaningless.
AI and documents
The document centre is the most valuable source for an assistant: policies are already written and their current versions are known. But a deliberate boundary is needed — we only index plain-text documents. Extracting text from PDF and DOCX produces silent errors: a figure in a table is misread and the assistant presents it as fact. A wrong answer costs more than no answer.
Templates and forms
The most used part of a document centre is usually not the policies but the forms: leave requests, expense claims, asset handover records. Putting these in their own category, above search results, ends most of the "where do I find this form" questions reaching HR.
One step further is turning the form from a document into a process: instead of a leave form to download and fill in, a request opened directly. We discuss which forms are worth digitising in our requests article.
Freshness reviews
The most insidious enemy of a document centre is decay. A regular pass that lists documents untouched for a defined period, reminders to owners, and a visible "last reviewed" date on each document keep that risk manageable.
Ownership is critical: every document needs a named owner. Collective ownership like "the HR team" turns into no ownership in practice — the reasoning is in our operations article.
What to do when migrating
Moving the old archive as-is is the most common mistake. Sort into three: documents genuinely used in the last year are migrated, documents whose content is valid but whose format is dated are rewritten, and the rest stays in the archive.
The quickest way to make that split is the access logs of the old system — migrating a document nobody ever opened only pollutes search. We detail this in the 90-day plan; where access restrictions belong is on our security page.