Document Management Best Practices

These are ordered by how much difference they make, not by how obvious they are. The first four account for most of the gap between an implementation that works and one that is quietly worked around.

1. Name one person accountable for the taxonomy#

Not a committee. A named individual with authority to say no to a new document type, a new metadata field or a new folder branch. Every other practice here degrades without that person, and the most common cause of a failed implementation is not a wrong decision but an unmaintained one.

The symptom to watch for is value lists accumulating near-duplicates. That is what an unowned taxonomy looks like six months in.

2. Enforce a small number of metadata fields#

Optional fields decay to blanks. That is not a prediction, it is an observation of every metadata model that has been in place for a few years. The only reliable defence is to make a few fields mandatory at upload and accept that the rest will be sparsely populated.

A field earns mandatory status if a report, a retention rule or a routing decision depends on it. "Department" usually does not. "Renewal date" does.

3. State the retention trigger event, always#

A schedule that says "seven years" without saying seven years from what cannot be applied consistently, and the errors are large rather than marginal. A contract with a five-year term and a six-year retention period, triggered at signature, is destroyed five years early.

Every rule needs a trigger: creation, closure, contract termination, end of employment, batch expiry, end of the financial year, or the subject reaching the age of majority.

4. Mark the current version unambiguously#

A version history is necessary and insufficient. What prevents harm is a reader being able to tell at a glance which version applies — and, where documents are printed, a watermark that carries the status onto paper.

The failure this prevents is expensive in every sector: building to a superseded drawing, working to a retired instruction, signing an agreement with a clause that was withdrawn.

5. Put approval thresholds in the workflow, not the policy#

Encoding the rule in configuration means it applies to every document rather than to the ones someone remembered to escalate. This is the difference between a control that operates and a control that is documented.

6. Log reads, not just writes#

A change log satisfies change control and answers none of the questions that arise when a document appears somewhere it should not have. Logging access produces a lot of entries; that is the cost of being able to answer.

7. Raise a disposition review rather than deleting automatically#

Retention periods are judgements made years before the expiry date. An automated deletion cannot evidence who authorised it, and "the system deleted it" is not an answer to a regulator.

8. Make the correct route the easiest route#

People use whichever intake path is least effort. If filing a document requires leaving the application they are working in, it will not happen reliably. An Outlook add-in or a watched mailbox does more for adoption than any amount of training.

9. Migrate less than you think#

Bring across what is still referenced, index the rest physically, and apply retention on the way in. Moving everything postpones every decision and produces a repository with the same problems as the one you left.

10. Keep authoring where it already happens#

Collaborative drafting in Word, Excel or Google Docs is better than anything a control-first system offers. Integrate rather than compete: the document opens in the tool people know and saves back as a new version.

11. Report on what is missing, not what is present#

The valuable reports are negative: records with no retention class, consignments without a proof of delivery, certificates lapsing next month, policies past their review date. Those find problems while they are still cheap.

12. Review shares and overrides on a schedule#

Per-document permission overrides and shares accumulate silently, and they are the mechanism by which a permission model stops describing reality. A quarterly review of a shared-access report catches it; nothing else does.

Practices that sound sensible and are not#

A few conventions recur in document management projects and make things worse. They are worth naming because they are usually proposed with good intentions.

Deleting old versions to save storage. It looks like housekeeping and it is the removal of evidence. Storage is cheap relative to the cost of explaining an absence, and a retained-version limit set once and never reviewed will discard exactly the revision someone eventually needs.

Automatic deletion at the end of a retention period. The most requested feature in records management and the one worth resisting. Periods are judgements made years earlier, and an automated destruction cannot evidence who authorised it.

A deep folder hierarchy. Beyond about four levels you are encoding metadata as folders, and each additional level multiplies the number of plausible places a document could be filed. Depth feels like organisation and produces ambiguity.

One metadata schema across every document type. The compromise that makes nothing reportable. A contract needs counterparty and renewal date; an invoice needs supplier and amount. Sharing one field set means most fields are irrelevant to most documents, which returns you to blanks.

Requiring everyone to acknowledge every policy. Produces noise and trains people to click through without reading, which is worse than not collecting the attestation at all because it creates a false record.

Sequencing matters as much as the practices#

The order in which these are introduced affects whether they stick. Enforcing mandatory metadata before anyone has a reason to search is experienced as bureaucracy; introducing it alongside a report people wanted is experienced as the price of the report.

The same is true of approval workflows. A workflow imposed on a process that was working is resented. A workflow introduced after an audit finding is welcomed, because it visibly solves a problem the team just had.

That is an argument for starting where a failure is currently visible, and against rolling out document control uniformly because it is good practice.

The two reports worth building first#

Records with no retention class. The honest measure of how complete the programme is, and the number that identifies the archive that will grow without limit.

Documents past their review date. For anything with a periodic review — policies, procedures, risk assessments, certificates — this is the report that turns an audit finding into a task.

Both are negative reports, which is the general pattern: the useful information in a document system is almost always what is missing rather than what is present.