An organisation that does not write things down runs on a small number of people who remember them. That works, quietly and expensively, until one of them is on holiday, changes team, or leaves.
The cost is not the day the person leaves. It is the eighteen months of decisions made without the reasoning behind the previous ones.
For an applied reference, see Monitask’s workforce analytics software overview.
What is actually worth writing down
Most documentation projects fail because they attempt to document everything, produce a large volume nobody reads, and are abandoned. The useful subset is narrower.
Decisions and their reasoning. The highest-value category by a wide margin. What was decided, when, by whom, what alternatives were considered and why they were rejected.
Without this, organisations re-litigate settled questions and — worse — reverse decisions without knowing what the original constraint was. A decision record is what lets someone in two years ask "is that constraint still true?" instead of assuming it never existed.
How things work, at the level of intent. Not step-by-step instructions, which go stale in weeks. The shape of the system and why it is that shape.
Conventions. Where things live, what is named how, which tool is used for what. Cheap to write, and it removes most of the friction new people experience. See tool sprawl.
Anything with an external dependency. Contracts, licences, renewals, contacts at suppliers, credentials arrangements. This category is small and its absence is acute.
Things learned the hard way. The incident, the failed approach, the thing that looked obvious and was not.
What is not worth writing down
Step-by-step instructions for interfaces that change. They go stale, and stale instructions are worse than none because people follow them.
Anything the system can tell you. Documentation that duplicates a queryable source will diverge from it.
Comprehensive process manuals nobody asked for. The classic failure: months of effort producing a document that is out of date on publication and that nobody reads.
Meeting minutes as a substitute for decision records. Minutes record what was said. A decision record records what was decided and why. They are not the same, and minutes are almost never re-read.
Why documentation projects fail
They are projects. A one-off effort produces a snapshot that decays. Documentation is maintenance, not a deliverable.
Nobody owns it. Shared ownership means no ownership.
It is written for the wrong reader. Most internal documentation is written by someone who understands the system for a reader who is imagined to also understand it. The actual reader is someone who joined three weeks ago at 6pm on a Friday.
It is not where people look. Documentation in a system nobody searches is not documentation.
Writing it is unrewarded. If nobody's performance conversation ever mentions it, it will be the first thing dropped.
Nothing is ever deleted. Confidence in documentation is destroyed faster by one wrong page than it is built by fifty correct ones. An unmaintained page that is subtly wrong is a liability.
What works instead
Write it as a by-product of the work, not as a separate task. A decision record written at the moment of the decision takes ten minutes. Reconstructed six months later it takes an afternoon and is less accurate.
Make the format small. One page. Context, decision, alternatives, consequences. If the format is heavy, it will not be used.
Date everything and put an owner on it. A page with a name and a date can be trusted or questioned. An anonymous, undated page can only be guessed at.
Review on a schedule and delete aggressively. A smaller set of pages that are correct beats a large set of uncertain vintage. Deleting is the most underrated documentation activity.
Write for the newest person. If a page requires context the reader does not have, it is a reminder for the author rather than documentation.
Put it where people already look. Fighting an existing habit loses.
Ask new joiners what was missing. They can see the gaps clearly, and only for about a month.
The test that identifies the real gaps
Pick a person on the team and ask: if they were unavailable for a month starting tomorrow, what would we not be able to do?
The answers are your documentation backlog, ordered by risk. Nothing else needs prioritising first.
Repeat for two or three people and the overlap tells you where the organisation is genuinely fragile — which is almost never where anyone assumed.
The connection to everything else
Undocumented reasoning is why reorganisations undo work that was done deliberately, why the same argument recurs quarterly, and why decisions get remade. See decision rights and reorganisations.
It is also the mechanism by which knowledge leaves an organisation. Someone resigning takes their memory with them, and the notice period is spent on handover conversations that attempt in two weeks what a written record would have done continuously. See internal search and the knowledge that leaves with people.
The minimum worth doing
If you do one thing: a one-page decision record for every decision that would be expensive to reverse. Context, decision, alternatives considered, consequences, date, author.
Ten minutes each, no tooling required, and it addresses the category with the highest cost of absence. Everything else in documentation is optional by comparison.
For broader public guidance and background, consult the National Institute of Standards and Technology.