Does an AI knowledge base replace our intranet?
Not necessarily. The intranet or wiki can remain a source. The knowledge base sits over approved sources and helps people retrieve and apply the right information.
A knowledge base should help people apply the answer, not just find a page.
Use an intranet or wiki when staff mainly need to publish and browse maintained pages. Use Company Knowledge Base when they need permission-aware, source-cited answers across approved documents and systems, with a clear next action. The wiki or intranet can remain a source instead of being replaced.
See how Company Knowledge Base works
Option A
Company Knowledge Base
A governed AI knowledge layer over approved sources, permissions, citations, and workflow context.
Option B
Intranet or wiki
A human-maintained content repository where staff browse, search, and read pages or documents.
Compare where each path fits.
| Criterion | Company Knowledge Base | Intranet or wiki | Kipanga Verdict |
|---|---|---|---|
| Primary use | Ask a question across approved documents and systems, then get a sourced answer or next step. | Publish maintained pages that staff search, browse, and interpret themselves. | Use the knowledge base when the job is answering and applying, not only publishing. |
| Source authority | Should retrieve from approved repositories, with excluded sources and named content owners. | Authority depends on which pages staff maintain and which version they find. | Both need source owners. AI retrieval also needs an explicit source allowlist. |
| Access controls | Should map source and answer access to user roles and groups so restricted content stays restricted. | Controls access at the page, space, or document level, depending on the platform. | Do not deploy an AI layer until its permissions match the underlying systems. |
| Provenance and freshness | Should cite the source, expose answer traceability, and flag material that is stale or due for review. | Shows page history and update dates for users to inspect themselves. | A citation is useful only when the source is current and authoritative. |
| Uncertainty and correction | Should abstain, show a source gap, escalate to a person, and route corrections to an owner. | Relies on users to notice unclear content and contact the page owner. | Define the failure path before staff depend on the answers. |
| Audit and data security | Needs query logs, source-version history, retention rules, encryption, and clear model-provider boundaries. | Needs access reviews, content history, retention rules, and platform security controls. | The AI layer adds answer-level logging and provider decisions to normal content governance. |
| Workflow support | Can connect answers to process guidance, drafts, or escalation paths. | Primarily publishes maintained guidance; forms and workflow support depend on the platform and its integrations. | Add retrieval when staff need an answer assembled from several approved sources before taking the next step. |
| Evaluation and ownership | Needs a representative test set, answer-quality thresholds, monitoring, fallback behavior, and a post-launch owner. | Needs editorial standards, search administration, review dates, and accountable page owners. | Neither option stays useful without an operating owner after launch. |
Choose Company Knowledge Base when
- Staff ask repeated questions that cut across documents or systems.
- Answers need source references and permission-aware access.
- The business wants knowledge to support workflows, onboarding, sales, or support.
- Owners can maintain approved sources, test answer quality, and correct failures after launch.
Choose Intranet or wiki when
- The need is mostly publishing static pages.
- Staff are comfortable browsing and interpreting content themselves.
- There is no clear high-value workflow yet.
- Content ownership and hygiene need to be fixed first.
Before deployment, name the source owners, match permissions to the underlying systems, test representative questions, and decide how staff report wrong or stale answers. If those controls are not ready, improve the intranet or wiki first.
Related implementation: SignsSA email qualification
Kipanga built a company-knowledge layer from SignsSA's website and internal documentation, then used it inside an email qualification agent. The published $100k+ annual savings, 160+ hours saved, and 50% of inquiries disqualified apply to the full email workflow, not the knowledge layer alone.
Read the SignsSA case studyQuestions teams ask before choosing.
How do we control what it can answer and who can see it?
Define approved and excluded sources, map source permissions to user roles and groups, and test that restricted material never appears in an answer for an unauthorized user.
What happens when a source is outdated or it does not know?
The system should abstain, cite the gap or stale source, and route the question or correction to a named owner rather than inventing an answer.
What security decisions are needed before rollout?
Confirm data classification, access controls, retention, encryption, model-provider boundaries, residency requirements, query logging, and who can review answer history.
How should we evaluate it before staff rely on it?
Build a test set from real staff questions, include permission and stale-source failures, agree on an answer-quality threshold, and name the owner for monitoring, incidents, and fallback behavior.