Close policy Q&A in emails; SharePoint?Confluence search with cited AI and controller review standardized answers

Overview

A controller team fielded repetitive policy questions during close because guidance was buried in old emails and PDFs across multiple repositories. Different teams answered the same questions differently, new hires relied on shadowing, and escalations spiked when deadlines neared. Intelligex deployed a permission?aware knowledge search across policies, past memos, and ERP help content, and added an AI copilot that cited sources and flagged uncertain answers for finance review. Responses became consistent and traceable, fewer questions escalated, and new team members came up to speed with less ad hoc mentoring—while SharePoint, Confluence, ERP help, and existing collaboration tools remained unchanged.

Client Profile

• Industry: Multinational corporate services
• Company size (range): Enterprise finance organization with regional controllership teams
• Stage: Policies stored in SharePoint and Confluence; ERP help and past memos scattered across shared drives and inboxes; Q&A handled in chat and email
• Department owner: Finance & Accounting (Controllership)
• Other stakeholders: Record?to?Report, Procure?to?Pay, Order?to?Cash, Payroll, Internal Audit, IT/Knowledge Management, Legal/Compliance

The Challenge

During close, the same questions returned in different forms: capitalization thresholds, cutoff rules, intercompany eliminations, three?way match tolerances, revenue policy interpretations, and payroll accrual practices. Answers lived in a mix of policy PDFs, archived emails, and past audit memos. Some references were outdated or lacked effective dates, and links to ERP help were broken or buried in bookmarks. Analysts pinged controllers in chat, who searched their inboxes or memory, then forwarded attachments without context. New hires collected personal note files that didn’t match formal guidance.

Inconsistency crept into decisions. One team followed the latest policy revision, another relied on a prior memo, and both could point to “official” documents. The same journal treatment might be defended differently across regions. Internal Audit flagged that citations were missing in many close checklists, and leadership observed that escalations clustered around topics with unclear ownership or messy document trails. Time spent answering questions replaced time needed to analyze flux and risk.

Why It Was Happening

Root causes were content fragmentation and a lack of governance around how policy knowledge was stored, discovered, and cited. SharePoint libraries, Confluence spaces, ERP help pages, and email archives each had their own structure. Version history and effective dates were inconsistent, and obsolete documents remained searchable alongside current policy. There was no canonical taxonomy for topics, processes, or entities, so people used different terms to describe the same concept.

Permission boundaries complicated search. Some audit memos and legal interpretations were restricted, while general policy needed broad access. Generic enterprise search could not reconcile these needs with finance context or produce auditable citations. Without a human?in?the?loop review step, confidence in machine?generated answers would have been low, so questions continued to route to controllers by default.

The Solution

Intelligex implemented a permission?aware knowledge pipeline that indexed policy sources, normalized metadata, and provided a finance?context search with an AI copilot. The copilot retrieved relevant passages across repositories, drafted answers with page?anchored citations, and flagged low?confidence or conflicting sources for controller review. A governed content registry tracked document lineage and effective dates. Approved answers and clarifications were published back to the knowledge layer, improving future responses. Integrations leveraged Microsoft SharePoint for document governance, Confluence for SOPs and FAQs (see Confluence Cloud), ERP help content such as Oracle Financials Cloud guides, and search capabilities built on Azure Cognitive Search.

• Integrations: SharePoint libraries and metadata; Confluence spaces; ERP help and user guides; past audit and policy memos from secured archives; collaboration tools for Q&A intake and notifications.
• Canonical knowledge schema: Standard fields for topic, process (R2R, P2P, O2C, Payroll), entity/region, effective date, owner, approval status, and sensitivity; document lineage and superseded links.
• Permission?aware indexing: Connectors respected repository permissions and sensitivity labels; search results and citations were filtered to the requestor’s access.
• AI copilot with citations: Retrieval?augmented generation that cited page anchors and document versions; confidence scoring and conflict detection; low?confidence answers routed to controller review.
• Review and publishing workflow: Maker?checker approvals for new guidance, clarifications, and deprecations; effective?dated changes; comment threads with rationale and attachments.
• Ask?to?publish loop: Frequently asked questions promoted to Confluence pages with canonical language and links to source policy and ERP help.
• Dashboards and alerts: Trending questions by close day and process, content gaps, aging policies without owners, and review backlog; alerts for expiring or superseded content.
• Audit trail: Immutable logs of searches, answers, citations, approvals, and content changes; exportable evidence packs for Internal Audit.

Implementation

• Discovery: Cataloged policy sources across SharePoint and Confluence; inventoried ERP help content and training assets; sampled recurring close questions and escalations; mapped current permissions and sensitivity labels; gathered Internal Audit feedback on citation and versioning gaps.
• Design: Defined the canonical schema and taxonomy; authored indexing scopes and permission rules; specified confidence thresholds and conflict detection logic; designed the review workflow and maker?checker approvals; planned dashboards, alerts, and retention policies.
• Build: Implemented connectors for SharePoint, Confluence, and ERP help; normalized metadata and effective dates; configured Azure Cognitive Search indexes; built the retrieval and answer generation service with citation anchors; created the review queue and publishing flow; assembled dashboards and audit logging.
• Testing/QA: Ran in shadow mode: answered common close questions with draft citations while teams continued email and chat; compared answers to controller guidance; tuned synonyms, filters, and confidence thresholds; piloted permissions with restricted audit content.
• Rollout: Launched for close?related topics first; retained email/chat as a controlled fallback; expanded to procurement, revenue, and payroll policies after stable cycles; enabled mandatory review for low?confidence or conflicting answers.
• Training/hand?off: Delivered sessions for controllers, analysts, and policy owners on searching, reading citations, submitting reviews, and publishing clarifications; updated SOPs for policy updates, effective dating, and deprecations; transferred ownership of taxonomy, indexes, and dashboards to Controllership and Knowledge Management under change control.
• Human?in?the?loop review: Established a recurring forum with Controllership and Internal Audit to review top queries, content gaps, and policy changes; decisions recorded with rationale and effective dates.

Results

Close questions routed through a single, searchable front door. Analysts received answers with page?level citations and versioned sources; when the copilot flagged low confidence, controllers reviewed and approved a response that then became part of the canon. Repetitive questions declined because the same authoritative language appeared wherever the question was asked, and exceptions were documented with rationale rather than in chat threads.

Consistency and onboarding improved. Regional differences were reflected in entity?specific pages with clear effective dates, and outdated documents were marked as superseded. New hires found policies and examples quickly without hunting through mail archives, and controllers focused on edge cases and policy changes rather than on rediscovering prior memos. SharePoint, Confluence, ERP help, and collaboration tools stayed in place; the change was a permission?aware search and review layer with governance and citations.

What Changed for the Team

• Before: Answers came from inboxes and memory. After: A copilot returned cited guidance from governed sources.
• Before: Outdated memos lingered alongside current policy. After: Superseded documents were linked and marked with effective dates.
• Before: Escalations spiked at close. After: Low?confidence answers routed to controllers with context and citations.
• Before: New hires relied on shadowing. After: Search surfaced policy, examples, and ERP help in one place.
• Before: Policy updates drifted. After: Maker?checker approvals and versioning governed clarifications and deprecations.
• Before: Audit citations were rebuilt. After: Exports showed who answered what, with which source and approval.

Key Takeaways

• Start with a taxonomy; normalize topics, processes, and entities so search can retrieve the right guidance.
• Respect permissions; index sources with access controls and sensitivity labels intact.
• Require citations; page?level anchors and version tags make answers defensible.
• Keep humans in the loop; route low?confidence or conflicting results to controller review with maker?checker approvals.
• Close the loop; promote repeated Q&A to canonical pages and deprecate outdated memos.
• Integrate, don’t replace; bind SharePoint, Confluence, and ERP help with a governed search and review layer.

FAQ

What tools did this integrate with? The solution indexed policy libraries in Microsoft SharePoint, SOPs and FAQs in Confluence, and ERP help content such as Oracle Financials Cloud guides. Search was powered by Azure Cognitive Search, and collaboration tools handled Q&A intake and notifications. No changes were required to the underlying repositories.

How did you handle quality control and governance? Content metadata included owners, effective dates, and sensitivity. A maker?checker workflow governed new guidance, clarifications, and deprecations. Every answer carried citations to specific document versions. Low?confidence or conflicting results were routed to controllers for approval, and all searches, answers, and content changes were immutably logged for Internal Audit.

How did you roll this out without disruption? The copilot ran in shadow mode during initial closes, providing draft answers and citations while teams continued normal channels. Differences were reviewed with controllers, index scopes and synonyms were tuned, and permissions were validated. Rollout started with close topics, then expanded to procurement, revenue, and payroll after stable cycles. Email and chat remained as a controlled fallback early on.

How were permissions respected and sensitive materials handled? Indexers read repository permissions and sensitivity labels; the search service returned only content the requestor could already access. Restricted memos remained discoverable only to authorized roles, and sensitive passages were redacted in previews. Access to the review queue and audit logs was role?based.

What happens when the copilot is unsure or policy changes? Low?confidence answers and conflicts triggered a review task with the relevant snippets and sources. Controllers approved or edited the response, which was then published with citations and effective dates. When policy changed, owners updated source documents and marked prior versions as superseded; the indexes refreshed and the new guidance became the default.

How were sources kept current? Scheduled crawls refreshed indexes, and change detection alerted owners to stale content and broken links. Dashboards highlighted aging policies without owners and high?traffic topics needing updates. Approved clarifications and examples were promoted to Confluence with links back to governing policy.