Call Center Outsourced evidence brief · Desk review · Published
How should a contact center measure knowledge article staleness?
A practical measure of knowledge staleness that starts from the decision an article supports, checks the current policy source, and weights review by customer impact.

On this page
- Key stats
- Key takeaways
- Question and scope
- What the evidence supports
- Define a stale article
- Start from the decision the article supports
- Record the policy source and its date
- Separate age from use
- Sample the articles agents actually open
- Check the answer against the current source
- Weight the review by customer impact
- Set review dates by risk
- Detect an article that is right but late
- Handle a client-owned article
- Review the searches that return nothing
- Compare stale rates across queues
- Use the finding in coaching
- Recheck after a policy change
- Limits of the evidence
- Manager review questions
- Conclusion
- Put this into a support lane
- Related operating guides
- FAQs
- Sources
Key stats
- 2.67 million U.S. customer service representative jobs in 2025 (BLS)
- $21.53 median hourly wage in May 2025 (BLS)
- 289,500 projected annual openings, 2025 to 2035 (BLS)
Key takeaways
- Tie every article to the decision it supports and the source that confirms it.
- Report last edit, last confirmed, and use count together rather than age alone.
- Review high-impact articles first and on a shorter cycle than routine ones.
- Recheck affected articles automatically after any client policy change.
Question and scope
This brief asks how an outsourced contact center should measure whether a knowledge article is stale. The population is the article set used by one queue during a defined period, and the unit is an article an agent opened to answer a customer. The question is whether the article still matches the policy or product it describes, not whether it was written recently. The analysis covers the measure and the review process, not the writing style of the article. Record the queue, the period, and the source systems before sampling.
What the evidence supports
Federal data-quality guidance calls for documented sources, tested reliability, and stated limitations, and contact-center standards call for defined processes and outcomes. Neither source defines a universal staleness threshold or a required review cycle. The operating inference is to tie each article to a named source and a review date, then test a sample against the live policy. A measure that counts age alone will mislead when a stable policy has not changed. The method should show what the article was checked against, not only when it was edited.
Define a stale article
Treat an article as stale when it no longer supports the decision it was written for, or when its source cannot be confirmed. A typo or an old screenshot may be a quality issue without changing the decision, while a wrong fee amount is stale in the sense that matters. Separate a factual conflict, an unconfirmed source, and a broken link so the review can act on each. Record the definition with the sample so the number is reproducible. The definition should be agreed with the client before the first review.
Start from the decision the article supports
Every article exists to help an agent make a decision or take an action. Write that decision at the top of the review record, such as whether a return is allowed or which document proves eligibility. An article with no clear decision is a candidate for removal or conversion into training. When the decision changes, the article must be rechecked even if the text still looks current. This step prevents a grammar review from being mistaken for a content review.
Record the policy source and its date
Name the document, system, or owner that the article depends on, and record the date that source was last confirmed. An article that cites a client policy page should carry the page version or the confirmation date. Without a source, a reviewer can only judge whether the text sounds plausible. Give the source field the same attention as the article body. A missing source is itself a finding, not a neutral blank.
Separate age from use
An old article that nobody opens may be harmless, while a new article used on every call can cause immediate harm if it is wrong. Report the last edit date, the last confirmed date, and the number of opens in the period. Age alone ranks a rarely used article above a heavily used one, which is the wrong order. Use both figures when choosing the review queue. The report should show the two measures side by side.
Sample the articles agents actually open
Pull the articles opened from the case system rather than the full library. The used set is usually much smaller and more relevant to the customer experience. Sample across call types, shifts, and agent tenure so the review is not limited to one team. Include the articles opened on escalated cases, because a wrong answer there costs more. Keep the sample size and the selection rule with the result.
Check the answer against the current source
For each sampled article, compare the key sentence with the current source rather than the previous version of the article. Read the sentence as the agent would use it on a call, including any condition that changes the answer. Check the examples and the linked forms as well as the main paragraph. A current rule with an outdated example is still a stale article. Record the exact conflict so the writer can correct it.
Weight the review by customer impact
Not every article deserves the same review effort. Group the used articles by the consequence of a wrong answer, such as a payment error, a privacy exposure, a missed safety step, or a simple address read-back. Review the high-impact group first and on a shorter cycle. A low-impact group can wait for a larger batch. The weighting should be written down so the queue does not drift toward the easiest articles.
Set review dates by risk
Assign a review interval based on the source's rate of change and the impact of an error. A regulated disclosure may need a quarterly check, while a store address may need one only after a client notice. Put the date on the article and in the review calendar. A date that nobody owns is only a wish. Name the person who confirms the article and the person who can approve a change.
Detect an article that is right but late
Some articles contain the correct rule but describe a process the queue no longer follows. The agent reads a step that no longer exists and improvises around it. Compare the article with the current system screen and the approved call flow, not only with the policy document. A stale screenshot or an old system name is a signal to check the whole procedure. Record the mismatch as a process defect rather than a wording problem.
Handle a client-owned article
The client may own the knowledge base and control the publishing rights. In that case the vendor's measure should distinguish articles it can fix from articles it can only report. Track the age of the reported defects and the date the client acknowledged each one. A queue should not be scored as stale for an article the client has not approved for change. Put the ownership field in the report so the accountability is visible.
Review the searches that return nothing
Search logs show where customers and agents could not find an answer. Count the queries with no useful result and read the cases behind them. A repeated empty search may mean the article is missing rather than stale, which is a different fix. Include the empty searches in the monthly review so the library grows where the work demands it. Record the top queries and the cases they touched.
Compare stale rates across queues
A single stale rate is hard to interpret without a reference. Compare queues with a similar call mix and source cadence, and state the differences that could explain the result. A queue with more regulated work may have a higher rate without being less careful. Use the comparison to direct attention, not to rank teams. The report should name the mix difference before the rate.
Use the finding in coaching
Turn the most useful stale article into a short coaching example. Remove customer identifiers and keep the decision point, the wrong answer, and the correct one. Agents remember a real case better than a general reminder. Ask the team which article they would have opened for the same question. Their answer shows whether the library or the training is the weaker link.
Recheck after a policy change
When a client changes a policy, add the affected articles to the next review automatically. Do not wait for the quarterly cycle, because the gap between the change and the update is when the wrong answer reaches customers. Confirm which articles cite the changed rule and which ones only mention the topic. Record the change date and the date each article was corrected. Close the review only when every affected article has a confirmed status.
Limits of the evidence
This brief cannot set a universal review interval or prove that an article is correct for every customer situation. The source may be ambiguous, and the agent may apply the rule to a case it was not written for. A sample can show a pattern without covering the whole library. Use the measure to guide review and state what it cannot show. The client and the policy owner remain responsible for the rule itself.
Manager review questions
Managers should ask which articles were opened most often and which high-impact articles have no confirmed source. They should also ask how many reported defects are waiting on a client decision and whether the last policy change reached every affected article. The remaining question is which article will be reviewed next and who owns it. Record the answers with the sample and the dates.
Conclusion
Knowledge staleness is a measure of whether an article still supports the decision it was written for. Start from the decision, record the source and its confirmation date, and weight the review by customer impact. Report age and use together, and recheck automatically after a policy change. The measure earns its place when it prevents a wrong answer, not when it produces a tidy chart. Review the method with the client at the same time as the article set.
Put this into a support lane
Bring one queue, its most-used articles, and the policy owner who can confirm the source.
Review the knowledge processRelated operating guides
FAQs
Is an old article always stale?
No. A stable rule can remain correct for a long time. Staleness depends on whether the article still matches its source, not on the edit date alone.
Who should own the review date?
One named person should confirm the article and one should approve a change. The client owns the policy source when the article depends on client rules.