Where the operation's intelligence lives.
Know is the function with the most documents and the least findability. It's the difference between an operation that knows things and one that can find things. Knowledge base, internal wiki, AI knowledge source, institutional memory. The one source both populations work from, instead of three that don't sync.
v 02 · live
What Know means.
Know is the work of making the operation's intelligence accessible to whoever needs it. The team, and the AI. Not the work of writing documentation. Not the work of maintaining a wiki. The work of designing the knowledge layer so the right answer is findable in seconds, lives in one source the AI is built on and the team trusts, and stays current as the work changes.
Most CX teams have two knowledge bases. The official one: out of date, hard to search, written for compliance. And the real one: Slack messages, agent notes, screenshots in DMs, a shared doc someone made in 2023. The official KB is theatre. The real KB is invisible institutional memory.
And underneath both, a third: whatever the AI vendor ingested when the integration was set up. Six months stale. Out of sync with both the official KB and the real one. Three sources of truth, none of them current, none of them aware of each other.
That's the gap Know names. Knowledge exists; it's just not where the team looks for it. The KB exists; it's just not what the team trusts. The AI runs on a snapshot of what was true once. Senior agents become walking knowledge bases because the actual one isn't usable. When they leave, the knowledge leaves with them. When the AI drifts, nobody notices because nobody is checking what it's grounded in.
Haven's Know module starts with the findability audit across all three sources. What gets searched. What gets found. What gets used. What the team is actually using as knowledge (Slack, docs, notes). What the AI was built on. The gaps surfaced and consolidated into one knowledge layer both populations work from.
The KB becomes the place the team actually goes and the source the AI is grounded in. Articles get retired more than added. Senior agents stop being load-bearing. The AI stops drifting because its source updates with the team's. Knowledge becomes a system, not a person, and not a vendor's six-month-old snapshot.
The progression. Four levels.
Three sources of truth, none current. The team uses Slack, DMs, and senior agents. The official KB is theatre. The AI runs on a vendor snapshot from setup day. Most articles are stale. Search rarely returns the answer. Nobody is checking what the AI is grounded in.
- KB is theatre
- Real knowledge in Slack/DMs
- AI grounded in stale vendor snapshot
- Senior agents are load-bearing
Some articles are trusted by humans. The AI's source is still its own thing. Heavy users have their bookmarks. The rest of the team avoids the KB. The AI's vendor refreshes its source on a quarterly cadence the team doesn't see. New articles get added; nothing gets retired. Volume grows; usefulness doesn't; AI drift doesn't even get measured.
- Trusted bookmarks (humans)
- Avoidance among team
- AI source refreshed by vendor on its own schedule
- No retirement on either side
One source, owned and pruned, feeding both populations. Findability is measured. Stale articles are retired. The AI is grounded in the same source the team uses. Senior agents stop being the knowledge layer. The vendor stops being the AI's separate source of truth.
- One owned KB, two consumers
- Findability metrics across humans and AI
- Routine retirement
- KB beats Slack; KB is what the AI is built on
The KB is a self-improving system, grounding humans and AI alike. Search misses (from the team or surfaced from the AI's recontacts) trigger article creation. Unused articles auto-flag for retirement. The AI's grounding stays in sync. The operation's intelligence compounds rather than scattering across three sources.
- Self-improving KB grounding both populations
- Auto-flagged stale content
- Search-miss + AI recontact → article
- Compounding intelligence, one source
What Know builds.
The findability audit
What the team searches. What the AI was built on. What gets used. The gaps across all three sources are the real KB roadmap.
- Search → found → used funnel (team + AI)
- Top 50 unfound queries surfaced
- AI knowledge source treated as first-class artifact
- Owner per coverage gap
The article lifecycle
Each article has an owner, a review date, and a retirement criterion. Updates propagate to the AI's grounding source. The KB shrinks as it improves, and the AI stays in sync.
- Owner & review date per article
- Retirement criteria named upfront
- Updates propagate to AI grounding source
- Usage telemetry across humans and AI
The knowledge capture ritual
A weekly or fortnightly cadence where senior agent knowledge gets surfaced and turned into shared articles. Same articles ground the AI. No more single points of failure on either side.
- Senior-agent knowledge surfaced live
- Converted into shared articles
- Same articles ground the AI's responses
- Rolling backlog of next topics
See it cascade.
A new contact cluster emerges. Know is where Haven drafts the knowledge article that the team uses and the AI is grounded in. The same signal also drafts the team's handling process (Improve) and the AI's prompt update (Build). One signal, three drafts, both populations stay aligned. See how the cascade starts in Know →