Skip to content
anti sandbox.
Knowledge quality

Audit a support knowledge base before connecting AI

Build a source inventory that separates approved answers, outdated policies and missing information before an AI support pilot.

Start with the questions customers actually ask

Collect a small, de-identified sample of recent support questions and group them by the decision the customer needs. A shipping question, an exception request and an account-access problem require different evidence. For each group, identify the document that should control the answer. A large document library is not proof that the common questions are covered.

Give every source a disposition

Use four outcomes: keep, revise, retire or missing. Record the source owner, approval date, affected product or market, and replacement document. Quarantine uncertain material until an owner decides. For a US team serving Saudi customers, check whether an English policy and its Arabic version describe the same conditions; translation quality cannot repair a disagreement in policy.

Fictional worked example

Suppose 12 delivery questions point to an old FAQ promising two days, while the approved operations policy says three to five working days after dispatch. Mark the FAQ for revision, keep the approved policy as the reference, and retest those questions after replacement. Do not silently average the two promises. This is a fictional audit example, not an Anti Sandbox performance result.

Action checklist

  1. Create a row for each question group: source, owner, market, approval date and disposition.
  2. Resolve conflicting promises before making the material available for answers.
  3. Record the percentage of sampled questions with an approved source; retain the sample size.
  4. Assign an owner and review date to every missing answer.
Related pages
See it yourself

A workspace worth exploring.

Open the demo