INSTRUCTIONAL DESIGN · AI-ENABLED CONTENT
File Simulations
A multi-AI workflow for manufacturing realistic practice simulations at scale.
In regulated environments, hands-on practice with real files is prohibited. File simulations solve that — but historically, building one took close to a month: combing through real files, pruning to fit learning objectives, then fabricating new entities and replicating every piece of supporting paperwork from scratch — policies, contracts, medical bills, demands, and complaints. This approach compressed that timeline using Claude projects and an AI-forward LMS, treating prompts and procedures as infrastructure rather than one-off course content.
Development cycle reduction — 30 days → 2 days per simulation
Simulations now scaling from one pilot, across multiple lines of business
Pilot cohort learner satisfaction
The Challenge
A regulated, multi-line enterprise needed cross-line practice for claim professionals. Real files were off-limits by regulation. A peer case-sharing pilot proved the value of cross-line practice — but engagement and effectiveness data showed a clear ceiling in that modality.
Strategy
Build a file simulation using generative AI at a scale and compliance bar that was newly viable. Structure it as a reproducible workflow with prompts and procedures as infrastructure — operable, auditable, and scalable by any team member, not dependent on a single builder.
Architecture
Multi-Tool Orchestration
Five components work as an integrated stack. Governance controls — anonymization verification, prompt-as-code documentation, coach rubrics — are embedded throughout rather than bolted on.
Claude
Anonymization, scenario design, document generation, and the prompt-as-code library documented in Markdown for team handoff
Second Nature
AI roleplay across all scenario types: defense counsel, unit managers, claimants, insureds — interview, negotiation, litigation, investigation
Sana AI
Delivery platform, AI-first learning experience, assessment, and learner data tracking
Claim Software
Organization-specific mock claim environment for realistic system-level practice within actual tools
Human Coaches
Human-in-the-loop by design — calibrating the most developmental conversations, rubric-based evaluation, and structured feedback at regular cadences
Governance
Responsible by Design
Real claims are the starting point — not the content. Every simulation is built so it cannot be traced back to any actual case or individual.
PII Redaction
All names, locations, dates, and private information are identified and removed before scenario work begins.
Scenario Transformation
Redaction is the floor, not the ceiling. Fact patterns are substantively altered, key circumstances changed, and all parties replaced. The real claim is a structural skeleton only — every detail in the finished simulation is purpose-built.
Human in the Loop
A human reviewer verifies redactions, guides the scenario to fit specific learning objectives, and instructs Claude on which facts and circumstances to change — ensuring each simulation is precisely calibrated to the business need.
The Approach
Discover → Design → Build → Pilot → Scale
Discover
Pilot engagement and effectiveness data identified the modality ceiling. Demand for cross-line practice was validated. Mature AI tooling created a viable path to high-quality simulated practice at the necessary compliance bar.
Design
Designed the multi-AI workflow stack and the governance controls it carries — anonymization verification, prompt-as-code documentation, coach feedback rubrics, and the mock environment in the internal claim software.
Build
Authored the Markdown prompt library and project instructions. Built the anonymization scheme with verification output. Produced anchor source documents against real-file structural blueprints — suit and complaint papers, policies, contracts, tender letters, claim correspondence. Configured AI roleplays across scenario types and human roles. Set up the mock claim environment, wrote rubric-aligned coach evaluation tools, and built the simulations on an AI-first learning platform.
Pilot
Small learner cohort. Coaches met learners at regular cadences mirroring real claim review. Captured NPS, qualitative feedback, assessment data, and time-on-task. Pilot returned maximum learner NPS.
Scale
Expanding from one simulation to 16+ across multiple lines of business. Prompt library improvements compound across iterations — ~25% per-cycle improvement anticipated. The unit of production shifted from “a course” to “a workflow.” Reproducibility is the durable asset.
Skills Demonstrated
AI Workflow Design · Prompt Engineering · Instructional Design · Learning Technology Integration · Governance & Compliance · AI-Enabled Content Development · Data-Driven Evaluation
