Reviewing Business Memory
How to review, verify, dispute, and prune the facts Muin learns about your organization — so AI answers stay grounded in knowledge you trust.
Business Memory is only as useful as it is accurate. A wrong “fact” in memory can quietly steer AI answers off course. This page covers the review workflow that keeps it clean.
The Three Verdicts
Open Intelligence → Memory and filter by Auto-extracted to see what’s awaiting review. For each entry:
| Verdict | When | Effect |
|---|---|---|
| Verify | The entry is correct | Carries more weight when the AI assembles context |
| Dispute | Wrong, outdated, or misleading | Held back from AI context |
| Delete | Should never have been captured | Removed |
A Working Rhythm
- Filter to Auto-extracted — that’s the unreviewed queue.
- Scan for the consequential ones first — facts about money, terms, deadlines, and commitments matter more than conversational trivia.
- Verify in bulk passes — most extractions are right; a quick weekly pass keeps the queue near zero.
- Dispute, don’t ignore, the wrong ones — an unreviewed wrong fact still looks usable; a disputed one is silenced.
- Watch the patterns — pattern entries (“invoices from X usually arrive late”) summarize recurring behavior. Verifying a true pattern is high-leverage: it informs every future interaction with that entity.
When Facts Change
Memory describes the world at the time it was learned. When reality moves on — a vendor changes terms, a policy is replaced — dispute or delete the stale entry and Add Memory with the current fact. The combination of disputing old + adding new is how you “edit” institutional knowledge.
What Reviewing Buys You
Every verified memory improves the assistant’s answers, agent decisions, and contextual sidebars across the platform. Ten minutes a week of review compounds into an AI that genuinely knows your organization — and can prove where its claims come from.