Scattered deal docs stalled divestiture — Document AI & Cloud DLP centralize buyer Q&A

Overview

A mining company’s divestiture stalled because deal documents were scattered across drives and email, redactions were inconsistent, and buyer Q&A lived in spreadsheets. The deal team spent checkpoints reconciling versions and fixing disclosures instead of evaluating offers. Intelligex automated document classification and redaction with Google Cloud Data Loss Prevention (DLP), enforced role-based access via Okta, and centralized buyer Q&A in a secure portal with audit trails. Leadership saw cleaner diligence packages, fewer errors, and smoother executive checkpoints because every artifact and exchange moved through a governed path.

Client Profile

• Industry: Mining and natural resources
• Company size (range): Multi-site producer with global operations
• Stage: Portfolio rationalization with active divestiture program
• Department owner: Strategy, Analytics & Executive Leadership (Corporate Development / Strategy Office)
• Other stakeholders: Legal & Compliance, Finance, Environmental Health & Safety (EHS), Geology/Exploration, IT/Security, Site Operations, External Advisors

The Challenge

Deal rooms contained technical reports, environmental permits, contracts, historical production data, and geoscience files. These sat in shared drives, personal folders, and email threads with inconsistent naming and metadata. Redaction practices varied by team, so sensitive personal information, proprietary formulas, and counterparty terms were sometimes handled differently across document sets. Advisors submitted revised files without a clear versioning convention. When buyers asked questions, answers were recorded in separate trackers and circulated by email, which made it hard to see what had been disclosed and to whom.

Security and compliance constraints added pressure. Some documents required masking of counterparties or locations under confidentiality agreements, and certain geospatial layers could only be shared in aggregate. The team used Okta single sign-on for internal systems but relied on ad hoc permissions for external users. There was no central place to confirm that a document had been classified, redacted, and approved before release, and no consistent audit trail linking a buyer’s question to the exact source content and who approved the response.

Why It Was Happening

Inputs were unstructured and lacked a common taxonomy. Technical reports, permits, and contracts arrived in mixed formats with inconsistent labels. Redaction was performed manually in PDFs without reliable detection of sensitive entities, and versioned documents were emailed back and forth. Requests for information (RFIs) were tracked in spreadsheets without a durable link to the underlying evidence or a record of what each buyer had received.

Governance arrived late. Document checks occurred just before data room updates, and Q&A went out without centralized review. Access controls were layered per folder rather than per buyer and topic. Without a single intake and approval flow, the team optimized for speed under deadline, which led to rework and avoidable escalations during executive checkpoints.

The Solution

We implemented a secure document workflow and buyer portal that standardized classification, automated redaction, and centralized Q&A with role-based access. Documents were ingested into Google Cloud Storage, classified with Document AI, and run through DLP for policy-based redaction. Items that failed confidence thresholds entered a human-in-the-loop review. A Cloud Run portal presented approved documents by buyer, enforced Okta-backed entitlements, and routed Q&A to owners with citation to governed sources. Nothing was replatformed: existing repositories and advisor deliverables fed the pipeline, and the portal coordinated access, review, and audit logging.

• Secure document storage and lifecycle in Google Cloud Storage with bucket-level isolation and retention tags
• Automated classification of contracts, permits, technical reports, and geoscience files using Document AI
• Policy-based redaction and masking for PII and sensitive terms via Cloud Data Loss Prevention (DLP)
• Buyer portal built on Cloud Run with download controls, watermarking, and per-buyer content sets
• Role-based access and entitlements through Okta groups for internal teams and external buyers (Okta Groups)
• Centralized Q&A with routing, citation to source documents, and approval notes before release
• Audit trail for uploads, redaction events, approvals, and disclosures using Cloud Audit Logs
• Human-in-the-loop review queue for low-confidence redactions, ambiguous classifications, and sensitive geospatial layers
• Version stamping and change-log to tie buyer-visible documents and answers to specific source versions
• Data minimization and masking policies by document type, with export controls enforced at the portal level

Implementation

• Discovery: Cataloged document types, sources, and current data room structures. Identified sensitive fields and document classes requiring special handling. Reviewed Q&A workflows, buyer segmentation needs, and existing Okta groups. Collected examples of past redaction errors and version sprawl.
• Design: Defined the classification taxonomy and metadata schema; authored DLP policies for PII and sensitive terms; specified routing rules for the human review queue; designed buyer workspaces and role mappings; and planned the Q&A model with citation and approval steps. Documented audit fields and retention rules.
• Build: Established ingestion to Cloud Storage with event triggers; configured Document AI processors and DLP templates; implemented the review queue and approval path; built the Cloud Run portal with buyer-specific libraries, watermarking, and download controls; integrated Okta for authentication and entitlements; and enabled audit logging across components.
• Testing and QA: Ran historical document sets through classification and redaction, validated masking accuracy, and tuned thresholds to reduce false positives and misses. Exercised buyer Q&A routing and citation requirements. Verified role-based access, watermark behavior, and version stamps. Conducted dry runs with Legal and advisors.
• Rollout: Operated the portal in mirror mode alongside the legacy data room while teams validated classifications, redactions, and access controls. Migrated active buyer groups in phases, preserving existing links and adding redirects. Maintained a manual override for urgent disclosures with post-review documentation.
• Training and hand-off: Delivered quick guides for deal coordinators on intake and approvals, for Legal on reviewing redactions and Q&A, and for IT/Security on access and audit reviews. Assigned stewardship for taxonomy updates, DLP policies, and buyer group management with a regular change-control cadence.

Results

Diligence packages became consistent and safer to share. Documents arrived with standardized metadata, redactions were applied according to policy, and low-confidence cases routed to reviewers before publication. Buyers saw a coherent library with clear lineage and watermarks, and Q&A responses carried citations back to the governed source, reducing back-and-forth and misinterpretation.

Executive checkpoints focused on offers and options rather than remediation. The team tracked who received what, when, and why, and could explain any disclosure with an audit trail. Advisors worked within the same flow, which reduced version sprawl and late surprises. Overall, divestiture decisions progressed with fewer fire drills and clearer accountability.

What Changed for the Team

• Before: Data rooms were assembled from shared drives and email. After: Documents flowed through a governed intake with classification and redaction.
• Before: Redaction practices varied by team. After: DLP policies and a review queue enforced consistent masking before release.
• Before: Buyer Q&A lived in spreadsheets. After: A portal routed questions, captured approvals, and cited source documents.
• Before: Access was managed ad hoc. After: Okta-backed roles controlled buyer and internal entitlements with audit logs.
• Before: Checkpoints reopened basic disclosure issues. After: Reviews referenced a single, approved set with version stamps and watermarks.

Key Takeaways

• Automate document classification and redaction, then require human review only where confidence is low or policy is unclear.
• Centralize buyer Q&A with citations; answers should link to governed sources, not screenshots or email snippets.
• Enforce role-based access across internal teams and buyers; identity groups keep entitlements consistent and auditable.
• Preserve lineage with version stamps and audit logs so disclosures and decisions are traceable during and after the deal.
• Meet teams in their tools; integrate shared drives and advisor deliverables into a governance layer rather than rebuilding the data room from scratch.

FAQ

What tools did this integrate with?
Documents were stored in Google Cloud Storage, classified with Document AI, and redacted using Cloud Data Loss Prevention. The buyer portal ran on Cloud Run with authentication and entitlements backed by Okta Groups. Audit logging used Cloud Audit Logs. Existing repositories and advisor deliverables fed the intake; nothing was replatformed.

How did you handle quality control and governance?
A taxonomy guided classification, and DLP applied policy-based masking for sensitive entities. Low-confidence cases entered a human-in-the-loop review with side-by-side originals and proposed redactions. The portal enforced approvals before documents or Q&A became buyer-visible. Every action—upload, classification, redaction, approval, and disclosure—was recorded with timestamps and user IDs in audit logs.

How did you roll this out without disruption?
We ran in mirror mode first: the portal showed the same content as the legacy data room while teams validated classifications, masking, and access. Buyers were migrated in phases, and links were preserved or redirected. Internal teams and advisors kept their tools; the new layer provided governance, access control, and Q&A routing on top.

How were buyer Q&A and access segmentation managed?
Each buyer had a workspace with scoped document sets. Questions were submitted in the portal, routed to content owners, and answered with citations to approved sources. Responses required Legal review for sensitive topics. Okta groups determined which users could see which workspaces and documents, and all exchanges were logged.

How did you ensure sensitive information remained protected?
DLP policies masked PII and predefined sensitive terms at ingestion, and the portal blocked downloads for unapproved or unredacted items. Watermarking indicated buyer and version, and access was time-bound by role. Audit logs and change histories enabled quick reviews of what was shared and when, with the ability to revoke access as needed.