←  Case Studies

Designing Human-in-the-Loop AI Systems for Anti-Financial Crime

Applying systems strategy, workflow architecture, and governance-aware operational design to support enterprise AI adoption in a regulated Anti-Financial Crime environment.

Overview

An enterprise AI initiative within a highly regulated Anti-Financial Crime (AFC) environment was being actively developed to support workflows such as regulatory analysis, policy review, and compliance documentation support.

As I analyzed the system’s workflows and rollout model, I identified a growing gap between technical capability and operational usability.

While the underlying AI platform functioned technically, it lacked the workflow structure, governance visibility, continuity controls, and operational support required for scalable enterprise adoption.

As the initiative matured, many of these concerns would later emerge during rollout, reinforcing a key finding of the project: successful AI adoption depends as much on operational design and workflow integration as it does on model capability.

What began as workflow analysis evolved into a broader systems initiative focused on operationalizing enterprise AI adoption through:

  • Structured workflow architecture.

  • Governance-aware interaction design.

  • Operational enablement strategy.

  • Continuity and escalation modeling.

  • Enterprise integration concepts.

The result was a systems-level framework, designed to help transform a transient AI chat tool into a more governable and operationally resilient enterprise environment.

Impact At-A-Glance

✓ Created a governance-aware AI operational framework

✓ Designed a structured analyst-focused agent workspace

✓ Mapped workflow architecture for regulated AFC tasks

✓ Devised an adoption and operationalization strategy

✓ Integrated Agentic AI governance and escalation reporting

Context

The Problem Space

The division utilized an AI agent intended to support high-stakes workflows including:

  • Regulatory monitoring

  • Policy analysis

  • Multi-document comparison

  • Gap analysis

  • Compliance documentation support

While technically functional, the system introduced several emerging operational risks during active development.

All workflows were handled through a single generalized conversational interface, while governance visibility, continuity behavior, escalation pathways, and operational transparency remained limited.

Most critically, the system treated operational compliance work as transient conversation rather than structured enterprise workflow.

This introduced growing operational risks around:

  • Workflow ambiguity

  • Validation behavior

  • Continuity

  • Operational trust

  • Long-term scalability

The challenge was no longer purely technical. It had become operational, requiring the alignment of workflows, governance, user behavior, and organizational adoption.

The core problem was not whether the AI system technically worked.

The problem was whether enterprise users could reliably operationalize it inside real compliance workflows.

Service Strategy

One Front Door, Differentiated Workflows

One of the most significant operational risks was treating all AFC workflows through a single generalized conversational interface.

Different operational tasks carried different:

  • validation requirements

  • source expectations

  • escalation thresholds

  • reasoning patterns

  • risk tolerances

To address this, I developed a lightweight workflow architecture model built around a unified entry point with differentiated operational pathways:

Regulatory Intelligence
Focused on regulatory monitoring and emerging guidance analysis.

Analysis & Alignment
Focused on policy analysis, multi-document comparison, and gap detection workflows.

Knowledge & Drafting
Focused on internal documentation support and structured synthesis tasks.

This structure preserved a unified user experience while introducing clearer workflow boundaries and more defensible operational behavior.

The enterprise integration strategy was designed to support:

  • Governance separation

  • Operational continuity

  • Enterprise support integration

  • Secure access control

  • Future scalability

Enterprise AI Agent Access Model

Flow documenting the integration of AI tools within the enterprise and the secure access pathways.

Flow map illustrating the proposed enterprise AI tools access pathway. From the top down: Enterprise Intranet, Self-Service Portal, AI Portal, then the enterprise agents, on the same line.

My Role

Lead Product & Service Design

Systems Strategy · Operational UX · Workflow Architecture · Interaction Design

I Independently:

  • Identified operational adoption risks

  • Mapped workflow and continuity gaps

  • Developed governance-aware workflow concepts

  • Designed the interaction architecture

  • Created high-fidelity prototypes

  • Produced rollout and operational maturity recommendations

  • Validated concepts with project stakeholders and power users

This work was developed alongside an active enterprise AI initiative, operating under real organizational, security, and regulatory constraints.

Core Areas

  • Enterprise AI Operationalization

  • Governance-Aware UX

  • Workflow Architecture

  • Human-in-the-Loop Systems

  • Cross-Functional Translation

SYSTEM Snapshot

A high-level view of how the system operates across structured workflows, validation layers, and governance checkpoints.

The models below translate regulatory expectations into enforceable system behavior.

System Overview: End-to-End Workflow

End-to-end workflow: how AFC tasks move through structured, governed workflows–from intake to validated output.

End-to-End AFC AI Workflow End-to-end workflow diagram showing AI task flow from user input through structured routing, differentiated processing pathways, validation layers, risk evaluation, and required human review before final output. Agent “Front Door” (Shared entry point: one interface for all AFC work) Task Routing/Intent Layer (Identifies task type and routes to the appropriate workflow) Knowledge & Drafting (Synthesis + SME-Reviewed Outputs) Analysis & Alignment (Structured Comparison + Evidence Mapping) Regulatory Intelligence (Freshness + Source Trust) Differentiated Pathways Consolidated Analysis Output Verification Engine (Source validation + hierarchy enforcement) Cross-Check Layer (Contradiction detection + secondary validation) Risk & Confidence Evaluation HITL/SME Review (Required for output validation and approval) Final Outputs (Insights, Analyses, Draft Reports, Training Materials)

SELECT SYSTEM COMPONENTS

Verification + Cross Check

Detects system drift, degraded performance, and inconsistencies over time. Cross-references outputs against multiple signals to ensure accuracy and reliability.

System diagram illustrating verification and cross-check processes, including source validation, contradiction analysis, secondary validation, and confidence scoring to ensure output accuracy and consistency.

HITL Governance Model

Embeds human review as a required system function, ensuring outputs are validated, contextualized, and approved before use.

Human-in-the-loop governance diagram showing required expert review and approval steps, where AI outputs remain advisory until validated by a subject matter expert.

Risk Degradation + Monitoring

Continuously monitors system behavior over time, identifying emerging risks and triggering intervention before failures escalate.

Risk monitoring diagram illustrating detection of system degradation signals such as repetition, citation issues, shallow reasoning, and confidence anomalies, with feedback loops for correction and refinement.

Core Use Case

Governance-aware, End-User-Focused Prototype

Traditional conversational AI interfaces are optimized for short, transient interactions. This environment required something different: support for longer-running investigations, workflow continuity, governance visibility, reference validation, and structured task execution within a regulated environment.

To address these needs, I designed a structured tri-pane analyst workspace centered on operational usability rather than dashboard-heavy analytics patterns. The experience combined guided workflow entry, validated resources and references, governance visibility, outputs and uploads management, and task-oriented interaction patterns that reinforced continuity, traceability, and evidence-backed decision making.

UX Research-Informed UI Design

Feedback from the initiative's primary stakeholder and power user reinforced the value of a governance-aware, workflow-oriented experience.

While the production environment centered on a generalized conversational interface, the proposed workspace concepts provided clearer task initiation, workflow continuity, operational transparency, and user support, capabilities viewed as critical for broader organizational adoption.

AFC AI Agent tri-pane prototype and the annotated version for engineering.

OPERATIONALIZATION & ADOPTION

Adoption & Enablement Strategy

Beyond the interface itself, I developed a lightweight operational enablement framework designed to support adoption, workflow consistency, governance visibility, and long-term scalability.

The framework focused on helping users understand how the system fit within existing operational processes, reinforcing evidence-backed decision making and human accountability, and treating workflow friction, overrides, and user behavior as signals for ongoing improvement.

This shifted the initiative from AI deployment into operational AI adoption.

Designing Beyond Initial Rollout

In addition to the core experience, I explored forward-looking concepts intended to support enterprise maturity over time, including:

  • Session continuity

  • Structured archives

  • Operational traceability

  • Service transparency

  • Escalation pathways, and

  • User-controlled resources

Several concepts were specifically designed to reduce adoption friction, support less-experienced users, and minimize the operational burden placed on subject matter experts as usage scaled.

OUTCOME

The project produced governance-aware workflow frameworks, systems architecture concepts, structured interaction models, high-fidelity prototypes, and operational maturity recommendations intended to support the responsible adoption of AI within a regulated environment.

More importantly, the work helped shift the conversation beyond model capability and toward the broader operational factors that determine long-term success, including workflow continuity, governance visibility, escalation pathways, organizational trust, and enterprise scalability.

Enterprise AI systems succeed or fail less because of model capability alone, and more because of how effectively they integrate into real operational workflows.

Operational Validation

Several months after the recommendations were delivered, rollout observations provided an opportunity to evaluate many of the risks identified during workflow analysis and systems design.

While the underlying AI platform continued to evolve, user feedback and production observations reinforced a consistent pattern: the most significant challenges were not technical capability issues, but operational ones.

Challenges emerged around workflow continuity, governance visibility, session persistence, operational transparency, and user confidence. Subject matter experts spent valuable time troubleshooting routine tasks, while system behavior became increasingly difficult for users to interpret, validate, and trust.

These observations reinforced the central finding of the project: successful enterprise AI adoption depends on far more than the model itself. Long-term success requires workflows, interfaces, governance structures, and operational support systems that help users confidently integrate new technology into their daily work.

Many of the operational challenges observed after rollout reflected workflow, governance, and adoption risks identified during earlier discovery and systems design.

What This APPROACH Demonstrates

Systems-Level Product Thinking

Identifying operational and organizational risks before they become larger workflow, adoption, or governance challenges.

Workflow Architecture

Structuring AI interactions around real operational tasks and user needs rather than generalized conversational behavior.

Enterprise AI Operationalization

Understanding that successful AI adoption depends on more than model capability. Workflow design, governance, operational readiness, and user trust ultimately determine long-term outcomes.

Strategic + Hands-On Execution

Moving fluidly between systems strategy, operational reasoning, interaction design, and high-fidelity prototyping.

Governance-Aware UX

Designing workflows that support accountability, validation, escalation pathways, and evidence-backed decision making.

Cross-Functional Translation

Bridging gaps between technical systems, governance requirements, stakeholder priorities, and end-user operational realities.

Next

Digital Product Modernization