AI cannot repair a contradictory policy. Quality starts with the source, review process, and escalation path.
Knowledge is part of the operation
A folder of PDFs is not automatically a knowledge base. Customer service must find the right answer for the right question, understand where it applies, and know what to do when the source does not cover an exception. KCS practices recommend capturing knowledge in the context of work, structuring it for reuse, and managing content health.
AI raises the stakes: a fluent answer can hide a missing or outdated source. NIST’s Generative AI Profile treats provenance, measurement, and ongoing risk management as system components rather than optional work after launch.
The minimum record for every article
| Field | Purpose |
|---|---|
| Title and question | Support search and intent matching. |
| Approved answer | Defines what may be communicated. |
| Source and version | Make correction traceable. |
| Scope and audience | Prevent applying a rule to the wrong customer. |
| Exceptions | Show when to stop and hand off. |
| Owner and review | Prevent orphaned content. |
Build in three passes
- Inventory: list common reasons, current sources, and contradictions.
- Normalize: write one direct answer per intent with links to detail.
- Validate: test real questions, language variation, missing data, and out-of-scope cases.
- Every answer has a source and owner
- Duplicate documents have been reconciled
- Dates, prices, and policies have review schedules
- Personal data is excluded unless required
- Exceptions reach a person with context
- Agent corrections return to the editorial process
Keep the base alive
Review unanswered questions, low-use sources, human corrections, and repeat contact. Do not publish everything at once: validate one domain, then expand. Use the guide on what to automate, apply the conversation scorecard, and monitor resolution metrics. A strong base reduces uncertainty; it does not remove supervision.
Turn this guide into an operating routine
Start with a small, observable pilot for “Customer service knowledge base: checklist before adding AI”. Select one team, one channel, and a defined period. Record the baseline: volume, queue age, ownership, exceptions, and complaints connected to the process. Then write a short rule, name the decision owner for ambiguous cases, and document how to return to the previous flow if something fails. This design prevents an unmeasured broad rollout and creates a fair before-and-after comparison.
- An operating objective stated as one verifiable sentence
- An execution owner and a supervisor for the review
- A conversation sample covering normal cases and exceptions
- Start date, duration, and a fallback rule
- A primary metric, a safety indicator, and event sources
- A final decision: keep, adjust, expand, or retire
During the pilot, review real conversations rather than trusting the dashboard alone. Confirm that state, owner, history, and next action match what the customer experienced. Separate process failure, tool limitation, integration failure, and execution error; each cause requires a different correction. At the end, publish the approved rule where the team works, retire old instructions, and schedule another review. The desired result is not only a better number. It is an operation that can explain why performance changed and reproduce the improvement without depending on one particular person.