Project Overview

RockHaven Children’s Services Inc. is a licensed child welfare organization with 25 years of operational history, currently serving approximately 40 children in care through a network of foster parents, supervisors, and administrative staff. RockHaven holds a pure license, meaning it satisfies every applicable sector standard, and the organization treats that distinction as a core marker of its credibility with families and referral partners.

Maintaining that standing depends on documentation. RockHaven manages three primary file categories, child files, foster parent files, and employee files, and each carries a complex taxonomy of records dictated by its licensing authority: intake packages, plans of care at 7-day, 30-day, 90-day, and 180-day milestones, case notes, serious incident and serious occurrence reports, medical and dental records, hearing tests, medication consents, safety assessments, safety plans, and rights forms. Every document carries its own timing requirement, and every gap represents exposure. The licensing authority had recently introduced administrative monetary penalties for missing documentation and signalled that it may conduct unannounced reviews, shifting the compliance burden from periodic catch-up to continuous point-in-time readiness.

At the time of engagement, all filing, naming, and completeness checking was performed manually. The Financial and Administrative Director maintained physical tracking books with paper checklists, and a self-audit required opening every individual file by hand. Bronson was engaged to design, develop, and deploy a Document Management Automation and Compliance Audit Solution built on RockHaven’s existing Microsoft 365 and SharePoint infrastructure, replacing manual filing with an auditable automated workflow and giving supervisors and directors real-time visibility into file completeness.

The Challenge

• Automating records management inside a vulnerable-sector organization is not a straightforward document classification problem. The records concern children in care, the consequences of a misfile are material, and the appetite for opaque automated decisions is close to zero.
• Manual, paper-dependent audit process. Completeness checking required a person to open each file individually and compare it against a handwritten checklist, a process staff described as archaic and extremely time consuming.
• No compliance visibility at any level. There was no mechanism to see what percentage of a supervisor’s caseload was in compliance, where the gaps sat, or which deadlines were approaching. Directors had no organization-wide view at all.
• Regulatory penalty exposure. Administrative monetary penalties combined with the prospect of unannounced review meant the organization needed defensible evidence of compliance on any given day, not evidence assembled after notice.
• Fragmented intake across channels and senders. Documents arrived from foster parents, caregivers, and medical providers through multiple channels. A shared administrative inbox had been created to centralize intake, but routing and filing remained entirely manual.
• A migration in flight. RockHaven was actively migrating files into SharePoint to give supervisors remote access, so the target environment was still moving while the solution was being designed.
• Hard automation boundaries. Care decisions, serious incident escalation, and outbound correspondence to external agencies could not be automated under any circumstance. Those boundaries had to be documented and enforced in the design, not left to convention.
• Sustainability after handover. RockHaven needed to operate and extend the system independently, including adding new children and foster parents and updating compliance rules, without ongoing dependency on external developers.
• RockHaven needed automation that reduced administrative burden while making every decision the system took visible, reversible, and defensible.

Our Solution

Bronson structured the engagement as a sequence of workstreams that established the compliance logic before any automation was built, so that every automated decision traced back to a documented rule.

1. Document Ecosystem Discovery and Data Modelling Bronson catalogued every document type across the three file categories with its required timeline, and mapped current intake workflows by sender, channel, and filing destination. That inventory became a formal data model defining entities (children, foster parents, employees, supervisors, and placements), their relationships, and a document taxonomy mapped to compliance milestones, serving as the single source of truth for the classification engine, dashboard, audit log, and reporting.

2. Compliance Rules Definition
Working directly with the Program Director and supervisory team, Bronson defined the per-child compliance checklist: which documents each entity type requires, at which milestones, how deadlines calculate, and what constitutes an overdue or missing record.

3. Automation Boundary Register
Bronson documented the processes that would deliberately remain manual, including care decision-making, serious incident and serious occurrence escalation, external correspondence, new child and foster parent onboarding, and system configuration changes, setting explicit limits on automation before development began.

4. SharePoint Structure and Naming Configuration Bronson configured folder hierarchy, metadata schemas, and naming convention enforcement, with fallback handling that quarantines non-conforming documents in a review queue under a clear error description rather than silently misfiling them.

5. Intake and Classification Pipeline Development
Bronson built the pipeline that detects email attachments and direct uploads, extracts metadata through OCR, natural language processing, and structured parsing, classifies each document by type, and matches it to the correct file using deterministic rules with AI-assisted fuzzy matching as a fallback. Every classification carries a confidence score, and only documents above the agreed threshold file automatically.

6. Human-in-the-Loop Review Interface
Bronson developed a dedicated review queue for documents that fall below the confidence threshold, cannot be matched to a known record, carry conflicting metadata, arrive unreadable, or belong to a category designated as never auto-file. Reviewers approve, reclassify, or reject in a single action, and corrections feed back to improve subsequent accuracy.

7. Compliance Dashboard
Bronson delivered layered views: organization-level compliance percentage and trending, a supervisor-level view scoped to each supervisor’s own caseload with drill-down to specific children and missing documents, and a child-level view showing what is on file, missing, or overdue against placement date and care milestones. Automated alerts notify the responsible supervisor as deadlines approach.

8. Access Control and Audit Logging
Bronson implemented role-based access across Administrator, Supervisor, and Director roles, and designed an audit logging schema capturing timestamp, actor, document identifier, classification decision, confidence score, and outcome for every action taken.

9. Testing, Acceptance, and Knowledge Transfer
Bronson measured classification precision and recall against a manually verified test set, tuned confidence thresholds, verified that supervisors could not reach files outside their caseload, and ran structured acceptance testing against bounded scenarios before go-live, training, and hypercare.

Key Deliverables

• Document Taxonomy and Data Model – A formal entity and relationship model with a hierarchical classification of every document type across the three file categories, mapped to compliance milestones.
• Solution Design and Technical Architecture Document – The consolidated design record covering requirements, current-state process maps, and the pipeline design for intake, extraction, classification, routing, and filing.
• Compliance Rules Specification – Deadline calculations, milestone logic, and alert thresholds expressed as configurable rules.
• Audit Log Schema – The logging structure ensuring every document action is traceable and exportable for external review.
• RBAC Matrix and Configuration – Role definitions and permissions implemented inside the Microsoft 365 environment, with caseload-level scoping for supervisors.
• Compliance Dashboard – Organization, supervisor, and child-level reporting views refreshed automatically after each processing cycle, delivered in tools staff already use.
• User Guides and Standard Operating Procedures – Role-specific documentation and recorded training covering day-to-day operation, reference data management, and compliance rule updates, supported by a structured acceptance test plan with defined pass or fail criteria.

The Impact

The engagement converted a manual, paper-dependent compliance function into an automated pipeline with continuous visibility.

• Completeness checking that previously required opening every file by hand became an at-a-glance dashboard view refreshed after each processing cycle.
• Supervisors gained caseload-level compliance visibility across approximately 15 children each, with drill-down to the specific missing document rather than an undifferentiated gap count.
• Directors gained organization-wide compliance percentage and trending, supporting point-in-time readiness for unannounced review rather than periodic catch-up.
• Physical tracking books and handwritten checklists were retired in favour of a system of record where every filing action is logged with timestamp, actor, confidence score, and decision rationale.
• Peak intake of approximately 1,000 documents per month is processed automatically, with only genuinely ambiguous or sensitive items routed to human judgment, and automated deadline alerts replaced reactive gap discovery with proactive notification ahead of the 7, 30, 90, and 180-day milestones.

All data remained inside RockHaven’s existing Microsoft 365 tenancy throughout, with no records copied to external infrastructure.

The strategic value of this work sits in the combination of automation and restraint. RockHaven did not need a system that made decisions on its behalf; it needed one that removed clerical burden while leaving every judgment touching a child in care with a qualified person. Encoding that distinction into an automation boundary register, a confidence threshold, and a human review queue is what made the solution acceptable to an organization with good reason to be cautious about opaque technology.

The longer-term benefit is a documented compliance model that outlives the project. The data model, taxonomy, and rules specification give RockHaven an asset it can extend as sector requirements evolve, and role-specific training positioned the organization to operate and adapt the system independently: less time spent proving compliance, and more capacity directed at the children and families the organization exists to serve.