Offsite prep stalled by file hunts — permissioned search for SharePoint & Box

Overview

Strategy offsite preparation at a global media conglomerate stalled because analysts spent cycles hunting for files across SharePoint, Box, and personal drives. Version drift and permission mismatches created uncertainty about which slides, memos, and market scans were current. Intelligex deployed a permissions-aware enterprise search layer using Azure Cognitive Search and Okta groups, then layered retrieval-augmented generation to surface citations from prior offsite decisions and analyst notes in Confluence. The team assembled materials with less rework, had clearer visibility into historical context, and avoided duplicate analyses because everyone pulled from the same governed index with reliable access controls.

Client Profile

• Industry: Media and entertainment conglomerate
• Company size (range): Multinational enterprise with multiple business units and brands
• Stage: Established public company
• Department owner: Strategy, Analytics & Executive Leadership (Office of the CEO and Corporate Strategy)
• Other stakeholders: Corporate Communications, FP&A, Business Unit Strategy, Legal & Compliance, IT/Security, Knowledge Management

The Challenge

Each offsite required a curated set of pre-reads, strategy memos, market scans, and performance snapshots. In reality, content lived in many places. Analysts stitched together decks from SharePoint folders, Box workspaces, and emailed attachments. Some documents were locked to certain groups; others were copied into personal drives and fell out of date. When leaders asked for historical context or prior decisions, teams searched chat logs or old agendas because there was no authoritative way to locate the right source.

Deadlines loomed while access requests and version checks piled up. A senior leader’s question—“Have we answered this before?”—triggered a scavenger hunt that slowed the agenda and recreated work. IT did not want to rip out existing repositories or reshape identity. Strategy leadership asked for a way to surface relevant material from the tools in use, respect existing permissions, and provide citations to prior decisions so that context was at hand when it mattered.

Why It Was Happening

Content and identity were fragmented. SharePoint sites, Box folders, and Confluence spaces used different naming conventions and permissions. Search worked within each system but not across them, and it did not blend document relevance with meeting outcomes or decision logs. Analysts saved local copies to move faster, which created more versions and diluted ownership.

There was no shared memory for past offsites. Decisions lived in decks and notes without a simple way to surface them. Even when a prior analysis existed, teams could not find it quickly or confirm whether it was the final version. Without a cross-repository, permissions-aware index tied to corporate groups, the organization defaulted to email threads and ad hoc curation at the worst time—right before executive reviews.

The Solution

We implemented a permissions-aware enterprise search layer that indexed existing repositories and respected access via Okta groups. Azure Cognitive Search handled cross-repository indexing and enrichment, while a retrieval-augmented generation (RAG) service summarized results with citations back to the original sources. Confluence decision pages and analyst notes were first-class inputs. Nothing was replatformed: SharePoint remained the primary intranet and file store, Box continued to host partner workspaces, and Confluence stayed the home for team notes. The strategy team received a single search experience with authoritative results, link-backs, and a light governance layer for curated “canonical” answers to recurring questions.

• Cross-repository indexing using Azure Cognitive Search, ingesting SharePoint sites, Box folders, and Confluence spaces with metadata and permissions
• Identity and authorization alignment via Okta groups so search results reflected each user’s entitlements (Okta Groups)
• Retrieval-augmented generation to summarize and link to prior offsite decisions and analyst notes, using Azure’s “use your data” pattern with citations (Azure OpenAI: Use your data)
• Entity extraction and tagging for topics, brands, regions, and business units to refine relevance and facet search
• Freshness and version signals that prefer approved or most recent documents and flag obvious duplicates
• Confluence decision logs surfaced as a distinct result type for quick access to past rationales
• Permissions-aware search UI embedded in SharePoint with role-based views for executives, Strategy, and analysts
• Human-in-the-loop curation for “canonical answers” to recurring strategic questions, with Corporate Strategy and Legal review before publication
• Privacy and compliance controls to mask restricted content while showing safe summaries and links for access requests
• Operational dashboards tracking query trends, zero-result topics, stale content, and curation queue

Implementation

• Discovery: Mapped content sources and typical offsite artifacts across SharePoint, Box, and Confluence. Reviewed permission models, Okta group structures, and where personal copies of key files originated. Interviewed Strategy, Legal, and IT to align on sensitive content boundaries and curation standards.
• Design: Defined the indexing scope, metadata schema, and entity tags for business unit, brand, region, topic, and decision date. Designed how Okta groups would flow into search authorization. Planned RAG prompts and citation formats, and documented curation and approval paths for canonical answers.
• Build: Configured Azure Cognitive Search indexers and enrichment, wired identity to Okta groups, and connected SharePoint, Box, and Confluence via APIs. Implemented RAG using the “use your data” pattern to return summaries with citations. Built a SharePoint-embedded search UI with filters and role-aware views. Added dashboards for query health and content freshness.
• Testing and QA: Ran relevance tests with historical offsite topics, validated permission enforcement across sample users, and checked that RAG summaries matched source documents with proper citations. Simulated stale content and duplicate scenarios to tune freshness scoring. Verified that sensitive items respected masking and request flows.
• Rollout: Launched read-only search to Strategy and analyst cohorts, keeping existing workflows. After tuning, enabled executive views and curation of canonical answers with approval gates. Expanded coverage to additional business units, with a standing process to onboard new repositories.
• Training and hand-off: Delivered short guides for analysts on refining queries and interpreting citations; for Strategy leaders on curating and approving canonical answers; and for IT on index operations and permission syncs. Established a human-in-the-loop review for sensitive summaries and a regular audit of entity tags and metadata quality.

Results

Offsite prep stopped revolving around file hunts. Analysts found the latest decks, memos, and decision pages quickly, with summaries that cited the exact sections to review. Strategy leaders saw curated answers to recurring questions, with the underlying evidence one click away. Duplicative efforts tapered as teams recognized prior work and built upon it rather than starting over.

Governance improved without slowing down the work. Search respected existing access, and sensitive content surfaced as safe summaries or guidance on how to request permission. The curation process created a backbone of vetted references that anchored executive discussions. Instead of chasing files, teams spent time on analysis and options, and executive sessions referenced the same sources.

What Changed for the Team

• Before: Analysts hunted across SharePoint, Box, and personal drives. After: A single, permissions-aware search returned the latest, relevant materials.
• Before: Prior decisions were buried in old decks and notes. After: RAG surfaced decision logs and cited sources directly in results.
• Before: Duplicate analyses appeared across business units. After: Freshness and dedupe signals exposed prior work and favored canonical answers.
• Before: Sensitive content required ad hoc checks. After: Okta group–based authorization and masking kept summaries safe and access clear.
• Before: Prep time went to file wrangling. After: Time shifted to synthesis and scenario planning with shared context.

Key Takeaways

• Unify discovery across repositories with permissions-aware search so teams start from the same library of truth.
• Layer retrieval-augmented generation to provide context with citations, not just a list of documents.
• Respect existing identity and access; use group-based authorization so security policy travels with content.
• Curate canonical answers for recurring questions, with human review to balance speed and governance.
• Keep SharePoint, Box, and Confluence; orchestrate them with indexing, enrichment, and lightweight approvals rather than migrating content.

FAQ

What tools did this integrate with?
We used Azure Cognitive Search to index SharePoint, Box, and Confluence content via their APIs. Authorization aligned to Okta groups so search results respected existing permissions. Retrieval-augmented generation followed Azure’s “use your data” pattern to produce summaries with citations. The search experience was embedded in SharePoint to meet users where they work.

How did you handle quality control and governance?
Quality controls included enrichment pipelines that tagged topics and business units, freshness scoring to prefer approved and recent files, and deduplication signals. Canonical answers to recurring questions moved through a human-in-the-loop review owned by Corporate Strategy, with Legal review for sensitive areas. Permission checks occurred at query time using Okta groups so restricted documents never leaked via summaries or previews.

How did you roll this out without disruption?
We launched read-only search to Strategy teams first, tuning relevance and permissions against real queries. Once validated, we enabled executive views and curation workflows, and then expanded to additional business units. No repositories were migrated; the layer indexed and respected content where it already lived, and users accessed it through a familiar SharePoint interface.

How were permissions respected across repositories?
User entitlements flowed from Okta groups into the search layer. At query time, Azure Cognitive Search filtered results based on the user’s groups and the source repository’s ACLs. Summaries and citations were generated only from documents the user could access, and sensitive results displayed masked previews with clear steps for requesting access when needed.

What about Box and Confluence integration details?
Box content was indexed via the Box API, capturing file metadata and respecting folder permissions. Confluence pages and decision logs were indexed via space-level access with page metadata, labels, and history. Decision pages were treated as a distinct result type to make prior rationales easy to find, and citations linked directly to the relevant anchors within those pages.