Direct answer
There is one route by which a knowledge base rots: wrong answers nobody owns, and out-of-date content nobody updates. A workable routine has three parts — one entry point (front-line staff can flag a wrong answer in one click), one list (fixes reviewed every week), and one owner per topic (who is accountable for which set of wording). Settle those three and one to two hours a week is enough. Miss any one of them and in six months it becomes an ornament nobody dares rely on.
A knowledge base is not finished at launch
Three months after launch there are two common states. One is a base that people use and people correct, getting more accurate with use. The other is a base nobody looks after, where the answers are frozen at the month it was built and the front line stops using it after two bad tries. The difference is not the tooling. It is whether a maintenance routine exists.
The three-part routine: an entry point, a list, an owner
1. A feedback entry point. When front-line staff find an answer wrong or incomplete, they can submit it in one click — for example a “something is wrong with this answer” control under the reply. What matters is that the feedback attaches to that specific question and answer, rather than asking someone to write an email describing it.
2. A weekly list. Aggregate feedback into a “to fix” list and go through it once a week. Four columns are enough: the problem, the document involved, the owner, the status. Out-of-date, wrong and missing are three different categories, handled separately.
3. A named owner per topic. Assign owners by category of material — commercial for pricing, operations for process, delivery for technical. Wording with no owner will dissatisfy somebody whether it changes or not.
How to spot out-of-date material
This is where problems accumulate fastest. Three signals are worth checking deliberately:
- A time signal. Put a review cycle on policy and pricing material (quarterly, half-yearly) and force it through when the date arrives.
- A conflict signal. When two answers to the same question appear, stop and settle which one is authoritative, then retire the other.
- A silence signal. Material that never gets hit may be unused, or it may simply be described in wording that does not match how people ask. The two cases are handled differently.
The time budget for maintenance
| Frequency | Action | Effort |
|---|---|---|
| Weekly | Work through the to-fix list | 1–2 hours |
| Monthly | Spot-check 10–20 real questions | 1 hour |
| Quarterly | Forced review of policy and pricing material + permission re-check | Half a day |
That level of effort is sustainable for a small business. The alternative — letting everything pile up for six months and then running one big clean-up — almost never gets executed: by the time the list runs to dozens of items, nobody wants to start.
One lesson worth keeping
Set up the maintenance routine during the pilot run, not “later, after launch”. Front-line feedback is at its most willing during the pilot; put the feedback entry point to work then and the routine grows by itself. Leave it until the enthusiasm has passed and it has to be pushed through by policy.
One question tells you whether it is being maintained
Ask: has any single answer been changed by a person in the last month? If nobody can answer that, the knowledge base is already losing value.
Key facts
| Three-part routine | A feedback entry point / a weekly fix list / a named owner per topic |
| Time budget | 1–2 hours weekly + 1 hour monthly spot-check + half a day quarterly review |
| Three signals of out-of-date material | Time signal (review cycle) / conflict signal (two answers coexisting) / silence signal (never hit) |
| Test of health | Has any answer been changed by a person in the last month |
Sources
- Maintenance and governance conventions from enterprise knowledge base delivery work
- Three-tier document classification and ownership of authoritative wording
Follow-up questions
What if there is nobody dedicated to maintenance?
You do not need a dedicated person. Go through the to-fix list once a week (1–2 hours), and assign ownership by category of material to existing roles.
What if there is too much feedback to handle?
Sort by impact first: wrong answers (fix immediately) > incomplete answers (fill within the week) > awkward wording (can be scheduled). Most feedback does not need action straight away.
How often should a full review happen?
Policy and pricing material quarterly, as a forced review; everything else can be driven by the monthly spot-check. There is no need to redo the whole base.
Should out-of-date material just be deleted?
Retire it rather than delete it, and keep a record of the retirement so changes in wording can be traced. At the same time, confirm nothing else still references it.