Enterprise knowledge bases
An enterpriseknowledge base, put in order.
An enterprise knowledge base is the shared layer that keeps the company's information sources in order: which documents exist, which are current, who has access to them and who updates them. Several AI assistants and search tools can rely on it, without each one indexing the same sources separately.
Several truths about the same subject
In large companies, the same procedure often exists in several versions, in several places, and nobody knows any more which one is valid. If AI is connected on top of this disorder, it will answer just as inconsistently — only faster.
- Duplicate or outdated documents, with no owner to decide which version is valid
- Different permissions in each system, hard for an AI tool to respect
- Every AI project builds its own index, with its own rules and costs
- No record of what information the AI uses and where it comes from
Who it is for
- Companies with several departments, sites or group companies, with documents scattered across many systems
- Organisations that want several AI assistants on the same sources, without indexing them separately each time
- IT and security teams that need to control which data AI tools can access
- Companies with audit requirements, where it must be known which document a piece of information came from
What we build into the knowledge infrastructure
- 01
Source inventory
A map of the systems, document types, owners and how often each one changes.
- 02
Synchronised permissions
Access rights are taken from the original systems and applied to every search.
- 03
Update connectors
New, modified or deleted documents are picked up automatically, on a schedule set for each source.
- 04
Validity rules
Metadata for version, expiry date and status — approved, in progress, withdrawn — used for filtering.
- 05
Domain owners
Each domain has an owner who approves the sources and receives reports of gaps.
- 06
Shared index and log
A single index, used by several assistants, with a record of what was consulted and by whom.
How it works
How a source gets in and stays current
The steps marked in yellow are approved by a person.
- Input
- Human in the loop
- Action
- AI
- Result
- 01
Input
New source proposed
A department proposes a document space for the AI to use.
- 02
Human in the loop
Owner approval
The domain owner confirms what goes in, what stays out and who has access.
- 03
Action
Connection and indexing
The connector picks up the documents together with their metadata and permissions.
- 04
Action
Regular synchronisation
Edits, deletions and changes to access rights are applied to the index automatically.
- 05
AI
Flagging problems
The system identifies documents that may be duplicate, contradictory or expired and sends them for review.
- 06
Result
Source available, with controls
Assistants and search use only approved, current content.
What needs to be prepared for the knowledge base
- The list of systems where the documents are held and their administrators
- The permissions structure: groups, roles, exceptions
- Document retention and withdrawal policies
- The owners designated for each knowledge domain
- The company's audit and logging requirements
Possible integrations
- Microsoft SharePoint
- Microsoft 365
- Google Workspace
- Confluence
- Notion
- Microsoft Entra ID
- Okta
- Document management systems (DMS)
- SQL databases
- Microsoft SharePoint
- Microsoft 365
- Google Workspace
- Confluence
- Notion
- Microsoft Entra ID
- Okta
- Document management systems (DMS)
- SQL databases
Each integration's compatibility is checked during the analysis phase, for the systems and versions you use.
Control and transparency
What people validate
- Approval of each new source by the domain owner
- Deciding which version is valid when documents contradict each other
- Regular review of permissions, together with the IT team
- Withdrawing expired or replaced documents
- Access to usage logs, limited to designated people
Limitations and how we handle them
- The infrastructure puts the documentation in order but does not write it: incomplete procedures are completed by the domain owners
- Permission synchronisation depends on what the source systems expose via their API; some cases need manual configuration
- Detecting contradictory documents is approximate and also flags cases that are not problems
- Governance takes time from the designated owners; with an agreed review schedule, the knowledge base stays current
What we track in the knowledge base
We agree the indicators together before launch and track them afterwards. The values depend on the process, the data and the volume.
- Source coverage
- How many of the inventoried sources are connected and synchronised — a regular report for each domain.
- Permission accuracy
- Regular tests with accounts for different roles, checking that each one sees only what it should.
- Synchronisation delay
- The time between a document being changed in the source system and its update in the index.
- Expired or duplicate documents
- Reports opened and closed by owners, tracked monthly by domain.
- Unanswered questions by domain
- The domains that generate the most unresolved questions, as a signal of missing documentation.
How we implement an enterprise knowledge base
- 01
Inventory and owners
We map the sources and permissions and designate domain owners together.
- 02
Governance rules
We agree what goes in, how it is approved, how validity is marked and how long logs are kept.
- 03
Gradual connection
We connect the sources one by one, starting with a pilot domain, and test permissions with real role accounts.
- 04
Operation and review
We set the review schedule with the owners and extend to the other domains.
Use cases
These scenarios are conceptual examples showing what can be built. They are not delivered implementations or results achieved for clients.
- Distribution group with several companies
Shared index for several companies
Shared procedures and those specific to each company are indexed with separate permissions, and each team's assistants see only what belongs to them.
Conceptual example - Automotive supplier
Versioned quality documentation
Work instructions have a version and an approved or withdrawn status, so search and assistants use only the version in force.
Conceptual example - Non-bank financial institution
Separate access by department
Risk, compliance and sales documents sit in the same index, but each department gets results only from its own area.
Conceptual example
Frequently asked questions
How is it different from the AI knowledge base assistant?
The assistant is the interface through which employees ask questions. The enterprise knowledge base is the infrastructure underneath: sources, permissions, updates, owners. On a small set of documents the assistant can work without it; at company level, it becomes necessary.
Do we need a new document system?
Usually not. The documents stay where they are, and the infrastructure connects and indexes them according to the agreed rules.
How do you respect existing permissions?
We take the access rights from the source systems where they expose them and apply them to every search. Where that is not possible, we configure them explicitly and test them with accounts for different roles.
What happens to personal data in the documents?
We identify it during the inventory and decide together whether it is excluded, restricted or masked. The legal basis for processing and the retention period are set in line with the GDPR and internal policies.
What happens when a question touches a domain without good sources?
The assistants built on the knowledge base say they did not find the information, and the question goes into the domain report, to the owner who can complete the documentation.
When does an enterprise knowledge base make sense?
When documents are in several systems, several teams or group companies use the same sources and you want several AI assistants without indexing separately for each one. For a single department with few documents, an assistant built on a simple RAG system is often enough.
Can’t find the answer? Write to us and we’ll reply directly.
Want a shared foundation to build AI assistants on?
Tell us how many systems your documents are in and how many teams use them. We will suggest a pilot domain and the governance rules.
Related solutions
- RAG systemsAssistants that search the configured sources before answering and show where each piece of information comes from.
- AI assistant for your internal knowledge baseEmployees ask how something is done in the company and get the answer from internal procedures and policies, with a reference to the source document.
- Semantic searchSearch by meaning across your website, catalogue or documents: relevant results even when users do not type the exact words.