Worked Example: Build an Error-Analysis Table

Twenty-eight support-copilot failures are real but unstructured, and the team is arguing prompt versus retrieval without a common pattern view. Error-analysis table: task -> domain -> failure type -> likely layer -> recommended next test The common trap is to tag failures with loose labels like bad answer or confusing response. Those labels describe frustration but do not help the analyst distinguish unsupported citations from stale retrieval, wrong workflow steps, or missing escalation cues. Define useful columns Create columns for workflow task, knowledge domain, failure type, likely layer, and what evidence would test that layer. These columns force the analysis…

Sign up free — one personalized lesson every day, matched to your role and goals.

Already have an account? Sign in