Knowledge Research
A bounded review of caller FAQ coverage as a repeat-contact measure
How FAQ coverage can be studied through repeat contacts, and why a lower question count does not always mean a better answer.

Table of contents
Question and scope
This review asks whether a caller FAQ can be evaluated through the repeat contacts it prevents. The unit is one caller who contacts the team more than once about the same underlying question. The observation window starts with the first contact and ends seven days later. The review covers the question topic, the answer given, the channel and the second contact.
The review excludes contacts about a new issue and contacts from a different person at the same household. It also excludes a repeat contact that follows a service failure unrelated to the question. Those boundaries keep the measure tied to the FAQ rather than to every reason a caller might return. Privacy context for storing caller questions is available from the FTC and the NIST Privacy Framework.
Define FAQ coverage
FAQ coverage is the share of common caller questions that have an approved, current answer. The answer must be findable during the call and short enough to read aloud. A topic with no entry, an entry that is out of date and an entry the handler cannot locate all count as gaps. Coverage is about usable answers, not the number of documents.
The definition should be applied to a fixed list of question topics. Build that list from call notes and handler reports over a defined period. Without a fixed list, coverage can be inflated by adding easy entries. The list should include the questions that generate the most repeat contacts.
Use repeat contacts as one signal
A repeat contact can mean the first answer was incomplete, but it can also mean the caller forgot, misunderstood or needed reassurance. The measure should record the second contact reason and compare it with the first. Only the contacts that repeat the same question belong in the analysis. The rest are useful for other quality work.
Do not assume that a drop in repeat contacts proves the FAQ improved. Callers may have stopped calling because the phone menu changed or because the service is quieter. Pair the count with a read of the calls and a check of the volume. The signal is strongest when the topic list and the call mix stay stable.
Build the topic list from call notes
Pull call notes from a four-week period and group them into question topics. Use the caller's words where possible, because internal labels can hide variation. Have a second person review a sample of the groups and resolve disagreements. A topic list built by one person often reflects that person's habits.
Keep the list to a manageable size, such as the top twenty topics. A long list becomes a document project and loses its operational focus. Add a residual category for questions that do not fit. Review the list each quarter so it follows the actual call mix.
Test findability during a call
An answer that exists but cannot be found in ten seconds does not provide coverage. Ask a handler who did not write the entry to find it during a mock call. Watch where they search and how long it takes. Record the failures as findability gaps, not as training failures.
The test should use the same screen and search terms available during a real call. A separate knowledge base with a better search can hide a problem in the live tool. Ask the handler to read the answer aloud and note whether it sounds natural. A findable answer that is awkward to say still needs work.
Separate answer quality from answer presence
An entry can be present and still fail the caller. It may be too long, too vague, out of date or written in internal language. Read a sample of the answers and score them for directness, accuracy and spoken length. The review should report presence and quality separately.
A high coverage score with low quality is a warning, not a success. It means the team has documents but not usable answers. The corrective action is to rewrite the entries that fail, starting with the topics that generate the most repeat contacts. Quality work should follow the call data rather than a general cleanup.
Account for channel and time
A caller who asks the same question by phone and then by email may be repeating because the channels do not share context. Record the channel of each contact and whether the handler could see the earlier one. A repeat caused by a channel gap is not an FAQ failure. It is a system design problem.
Time also matters. A caller who returns after a week may have forgotten a correct answer, while a caller who returns in ten minutes may have received an incomplete one. Record the interval and read the calls before assigning cause. The timing helps separate memory from a gap.
Compare before and after a change
Choose a small set of entries to rewrite and define what should improve. A reasonable expectation is that handlers find the answer faster and that repeat contacts on the topic become less common. Measure both before and after the change. Do not change the topic list in the same period.
Keep the comparison honest by reporting the call volume and any major service change. If the volume fell sharply, the repeat count will fall for reasons unrelated to the FAQ. State that limitation beside the result. A careful comparison is more useful than a dramatic claim.
State the limits
The measure cannot show whether a caller was satisfied with the answer. It cannot show whether a caller who did not return found the information elsewhere. It also cannot separate a good answer from a caller who simply gave up. Those questions require direct feedback or a different study design.
The measure can show whether common questions have usable answers and whether the same topics keep returning. That is enough to guide a practical improvement cycle. The report should present the result as one part of call quality, not as proof of the FAQ's value. A short note on the next review date keeps the work honest.
A protocol for one team
Select a four-week window and build the topic list from call notes. Match repeat contacts to the first contact and record the interval and channel. Test findability with a handler who did not write the entries. Score a sample of answers for directness and spoken length.
Review the result with the handlers and choose two entries to rewrite. Repeat the measurement after the next four-week window. Keep the method and the topic list stable so the comparison means something. The protocol is deliberately small enough to run alongside normal call work.
Protect caller privacy in the topic list
A topic list built from call notes can expose personal details if the notes are copied without care. Remove names, account numbers and any detail that is not needed to describe the question. Store the list in a place with the same access rules as the call notes. A short example is usually enough to explain a topic.
Do not use a real caller's story in a training slide without permission. Rewrite the example so the question is clear and the person is not identifiable. The privacy rule applies to the review notes as well as the published FAQ. A useful measure does not require exposing the caller.
Report the result
Report the topic list, the repeat-contact count, the findability result and the answer-quality score. State the call volume and the period beside the numbers. Add the main limitation so a reader does not overread the result. A compact table is enough for a single team.
Share the report with the handlers before the review meeting. Ask which topic should be rewritten first and which answer is hardest to say aloud. Choose two changes and assign an owner. The report should end with a next step, not a summary of activity.
Keep the review going
A one-time study becomes stale as soon as the service changes. Put the next review date on the team calendar and keep the topic list with it. Review the entries after a price change, a new location or a new call reason. The FAQ should follow the work instead of becoming a shelf document.
Rotate the person who leads the review so the knowledge does not sit with one handler. A fresh reviewer notices awkward language that the original writer has stopped seeing. Keep the method the same and change the entries. The routine matters more than a perfect first version.