Building dependable AI support
How do you turn a solved support question into AI knowledge?
Turn a solved support question into knowledge by extracting the general rule, adding its prerequisites and exceptions, and removing private case details. Review the draft before approving it. Then test whether the new article answers both the original question and a nearby case where the answer should differ.
What should be generalized?
Capture the fact another customer can use: how a process works, which setting to choose, or when support must help. A successful outcome for one account may have depended on a discretionary exception. Do not publish that exception as a universal promise without approval from the policy owner.
What should be removed?
Remove names, account numbers, private financial details, and internal investigation notes that are not needed to explain the rule. A conversation contains evidence for solving one issue, but not every sentence belongs in general knowledge. Preserve the distinction between confirmed facts and suggestions made during troubleshooting.
How do you avoid contradictory articles?
Check whether an existing article already answers the question. Update or replace the relevant explanation rather than adding a second answer with a slightly different title. Test the conditions that caused confusion and make sure the corrected rule is actually available to the assistant.
Practical checklist
- Extract a general rule rather than a personal outcome.
- Remove private and unnecessary details.
- Check existing knowledge for overlap.
- Review, approve and test the resulting article.
Worked example
An agent confirms that the business serves customers in Australia. The reusable article should explain the actual geographic availability, including relevant limits, rather than save a customer-specific “yes, we can help you” response without its conditions.
How Fire Away fits
Fire Away can propose an article from a conversation and an agent’s teaching instruction. The draft is reviewed and approved before publication; approval records what was taught in an internal note.