ShapersOfAI.com · From Users to Shapers · AI Leadership · Future of Work · AI Skills

AI Use Case

Intranet Search Agent / AI Enterprise Search

Answers employee questions using approved intranet content and provides direct links to the relevant sources.

Employees often struggle to find reliable information across large intranets. Traditional search returns lists of pages based on keywords, leaving employees to open several results, assess their relevance and assemble the answer themselves.

An Intranet Search Agent allows employees to ask questions in natural language. It searches approved intranet and knowledge sources, identifies the most relevant information and generates a concise answer with links to the original pages.

The agent must respect existing access permissions and clearly distinguish between sourced information, uncertainty and missing content. It should not answer from general AI knowledge when the question concerns company-specific policies, processes or responsibilities.

Beyond helping employees, the agent can provide valuable insights to the intranet team. Unanswered questions, conflicting sources and frequently searched topics reveal where content is missing, outdated or difficult to understand.

Painpoints

Employees do not know where specific information is stored
Traditional intranet search returns too many irrelevant results
Employees must combine information from several pages
Different pages may contain inconsistent or outdated information
Important content is difficult to find without knowing the correct terminology
Communication and support teams repeatedly answer the same questions
Search results do not provide a direct, contextual answer
Employees may rely on unofficial or outdated documents
Existing content gaps are difficult for intranet teams to identify

Goals

Provide direct answers to employee questions
Make approved intranet knowledge easier to access
Reduce time spent searching across pages and documents
Link every answer to its original sources
Respect existing access permissions
Reduce recurring questions to communication and support teams
Identify missing, unclear or contradictory content
Improve the overall quality of the intranet knowledge base
Increase trust in enterprise AI answers

Risks

The agent may produce an unsupported or misleading answer
Outdated pages may lead to outdated answers
Conflicting sources may not be recognised correctly
Employees may trust a generated answer without opening the source
Incorrect permission configuration could expose restricted information
The agent may answer beyond its approved knowledge base
Source links may not support the complete generated answer
Employees may assume the agent replaces an official decision or specialist
Search logs may contain personal or sensitive questions

Implementation Idea

Define a focused initial scope, such as intranet news, workplace information or communication policies
Select the approved SharePoint sites, pages and document libraries the agent may search
Exclude drafts, archives, personal OneDrive files and unapproved content
Review content ownership, permissions, review dates and metadata
Remove or flag outdated, duplicated and contradictory information before launch
Configure a Copilot Studio agent connected to the selected SharePoint knowledge sources
Instruct the agent to answer company-specific questions only from those sources
Require source links for every substantive answer
Ask the agent to state clearly when no reliable answer is available
Prevent it from completing missing information with general model knowledge
Configure authentication so existing Microsoft 365 permissions remain effective
Add suggested questions to help employees understand the agent’s scope
Test the agent with real employee questions, synonyms and common spelling variations
Include difficult tests involving conflicting information and restricted documents
Let users rate answers and report inaccurate or outdated information
Record unanswered questions and unsuccessful searches
Provide the intranet team with a recurring report of content gaps and problematic sources
Route sensitive questions to the responsible team instead of generating an answer
Assign an owner for every connected content area
Review answer quality, sources and permissions regularly
Expand the knowledge base only after the pilot produces reliable results

Practical Microsoft implementation

A first version could combine:

SharePoint Online as the approved knowledge source
Copilot Studio for the conversational agent, instructions and topic boundaries
Microsoft Entra ID for user authentication and permission-aware access
Teams or the intranet as the employee-facing interface
Power Automate for feedback routing and content-owner notifications
Power BI or SharePoint Lists for reporting unanswered questions and content gaps

A pilot should start with one clearly owned content area rather than the entire intranet. Around 50–100 representative employee questions can be used to evaluate answer accuracy, source quality, permission handling and the percentage of questions the agent cannot answer reliably.

Advatera Community Examples

Allianz’s intranet team has been developing a Search Agent in Microsoft Copilot Studio. The aim is to help employees find and access information from the company intranet more effectively.

The Search Agent complements Allianz’s work on editorial content quality: reliable AI answers depend on intranet pages that are current, clearly structured and properly maintained.