role-based-training-generator

name: role-based-training-generator description: Use when creating training materials for AI capability deployment. Use after acceptance testing passes and go-live is scheduled. Produces role-specific modules with procedures, anti-patterns, and decision aids.

Role-Based Training Generator

Overview

Generate training materials tailored to each role interacting with an AI capability. The goal is training that enables each role to use the system effectively AND recognize when something is wrong.

Core principle: Training isn't about how the system works - it's about how to do the job with the system.

Training Components Per Role

Every role module must include:

Output Format

training_package:
  capability: "[Name]"
  version: "[Training version]"
  effective_date: "[Go-live date]"
  total_roles: [Number]

role_modules:
  role_name:
    role_description: "[What this role does with the system]"
    time_to_complete: "[X minutes]"
    prerequisite_training: ["[If any]"]

    learning_objectives:
      - "[Observable, measurable objective]"

    key_tasks:
      task_name:
        frequency: "[Daily/Weekly/As needed]"
        steps:
          1: "[Step with specific action]"
          2: "[Step with specific action]"
        sla: "[If applicable]"
        compliance_checkpoint: "[Required documentation]"

    anti_patterns:
      - pattern: "[What NOT to do]"
        why_problematic: "[Consequence]"
        correct_approach: "[What to do instead]"

    decision_aid:
      title: "[Job aid name]"
      format: "[Card/Poster/Desktop reference]"
      content:
        do_when:
          - condition: "[Situation]"
            action: "[What to do]"

    error_recognition:
      signs_of_problem:
        - symptom: "[What you might see]"
          possible_cause: "[What it might mean]"
          action: "[What to do]"
      escalation_contact: "[Who to notify]"

    compliance_reminders:
      - "[Regulation-linked requirement]"

    assessment:
      format: "[Quiz/Practical/Observation]"
      passing_score: "[X%]"
      topics:
        - "[Topic tested]"
      recertification: "[Frequency]"

quick_reference_materials:
  - type: "[Card/Poster/Guide]"
    audience: "[Role]"
    title: "[Title]"
    content_summary: "[What it covers]"

deployment_plan:
  training_schedule:
    - role: "[Role]"
      sessions: "[Number]"
      duration: "[Hours]"
      format: "[In-person/Virtual/Self-paced]"
  go_live_support: "[Floor support plan]"

Task-Based Procedure Format

Don't explain features - explain how to do tasks:

key_tasks:
  review_medium_confidence_match:
    frequency: "Daily - primary task"
    context: "Items in review queue with 80-94% confidence"

    steps:
      1: "Open item from review queue"
      2: "Compare AI-extracted fields to source confirmation document"
      3: "Verify: Counterparty name matches (allow for abbreviations)"
      4: "Verify: Security identifier matches exactly"
      5: "Verify: Quantity matches (check for splits)"
      6: "Verify: Price matches within tolerance (±0.01)"
      7: "Verify: Settlement date matches (check date format)"
      8: "If all match: Click Approve"
      9: "If discrepancy found: Click Reject, select reason"
      10: "If uncertain: Click Escalate, add note"

    sla: "15 minutes per item"

    compliance_checkpoint: |
      Decision automatically logged with timestamp and user ID.
      Override requires reason code selection.

    common_errors:
      - "Approving without checking all fields"
      - "Missing date format differences (MM/DD vs DD/MM)"
      - "Overlooking partial quantity matches"

Anti-Patterns Structure

Every role needs explicit "don't do this":

anti_patterns:
  - pattern: "Bulk approving without review"
    why_problematic: |
      Creates audit trail showing rapid approvals.
      Missed errors cause settlement failures.
      Monthly audit will flag pattern.
    correct_approach: |
      Review each item individually.
      Minimum 2 minutes per item expected.

  - pattern: "Overriding AI to 'speed up' processing"
    why_problematic: |
      Overrides are sampled in monthly audit.
      Repeated pattern triggers investigation.
      Defeats purpose of confidence routing.
    correct_approach: |
      Override only when AI is demonstrably wrong.
      Document specific reason for each override.

  - pattern: "Not documenting escalation reason"
    why_problematic: |
      Senior analyst lacks context.
      Resolution takes longer.
      Pattern analysis impossible.
    correct_approach: |
      Always add note explaining why escalated.
      Be specific: "date format unclear" not "looks wrong"

Error Recognition Training

Staff must know when to raise alarms:

error_recognition:
  signs_of_ai_malfunction:
    - symptom: "Confidence scores clustering at unusual levels"
      example: "All items showing exactly 87% confidence"
      possible_cause: "Model calibration issue"
      action: "Flag to supervisor, continue processing, notify Model Ops if persists 30+ min"

    - symptom: "Same extraction error repeating"
      example: "Date field blank on every confirmation"
      possible_cause: "Format change from counterparty"
      action: "Flag first instance, escalate if pattern continues"

    - symptom: "Queue not updating after processing"
      example: "Approved items still in queue"
      possible_cause: "System sync issue"
      action: "Refresh browser, if persists notify IT support"

    - symptom: "Unexpectedly high escalation rate"
      example: "30% escalation vs normal 4%"
      possible_cause: "Model issue or unusual day"
      action: "Notify supervisor immediately"

  escalation_contact:
    first_line: "Supervisor on duty"
    model_issues: "Model Operations: ext 5555"
    system_issues: "IT Help Desk: ext 1111"

Decision Aids

Provide ready-to-use quick reference:

decision_aid:
  title: "Match Review Decision Card"
  format: "Laminated card for desk"
  dimensions: "4x6 inches"

  content:
    approve_when:
      - "All extracted fields match source document"
      - "Proposed trade record matches confirmation"
      - "Any differences are immaterial (trailing spaces, case)"

    reject_when:
      - "Extraction error - AI misread the document"
      - "Wrong trade - details don't match our records"
      - "Can't find matching trade in our system"

    escalate_when:
      - "Format you haven't seen before"
      - "Multiple potential matches"
      - "Discrepancy you can't resolve"
      - "System behaving unexpectedly"

    never_do:
      - "Approve without reviewing all fields"
      - "Override because you're behind"
      - "Ignore repeated errors"

Compliance Integration

Weave compliance into procedures, not separate:

key_tasks:
  override_ai_decision:
    steps:
      1: "Click Override button"
      2: "Select reason code from dropdown (REQUIRED)"  # Compliance
      3: "Add text explanation (recommended)"           # Audit benefit
      4: "Click Confirm Override"
      5: "System logs override with your ID"            # Awareness

    compliance_checkpoint: |
      REGULATORY REQUIREMENT: All overrides logged per FINRA 4511.
      Overrides without reason codes flagged in monthly audit.
      Override patterns reviewed quarterly by Compliance.

Assessment Requirements

Specify how competency is verified:

assessment:
  operations_analyst:
    format: "Practical exercise - simulated queue"
    duration: "30 minutes"
    scenarios: 10
    passing_score: "90% correct decisions"

    scenario_types:
      - "Approve correct match" (4 scenarios)
      - "Reject extraction error" (2 scenarios)
      - "Reject wrong trade" (2 scenarios)
      - "Escalate appropriately" (2 scenarios)

    failure_remediation: |
      Score <90%: Review with trainer, retake in 48 hours
      Second failure: One-on-one coaching before third attempt

    recertification: "Annual or after significant system change"

Time Estimates

Provide realistic training duration:

training_schedule:
  operations_analyst:
    total_time: "2 hours"
    breakdown:
      - module: "System overview"
        duration: "20 minutes"
      - module: "Review procedures"
        duration: "40 minutes"
      - module: "Override and escalation"
        duration: "20 minutes"
      - module: "Practical exercises"
        duration: "30 minutes"
      - module: "Assessment"
        duration: "10 minutes"
    format: "Instructor-led with hands-on practice"

  senior_operations_analyst:
    total_time: "1 hour"
    prerequisite: "Operations Analyst certification"
    format: "Self-paced with supervisor sign-off"

  operations_manager:
    total_time: "45 minutes"
    format: "Self-paced dashboard walkthrough"

Common Mistakes

Red Flags in Your Output

If your training has these, it's not ready:

• Same training for all roles
• Feature explanations instead of procedures
• No "what NOT to do" section
• Compliance as separate module
• No error recognition training
• No decision aids
• No assessment criteria
• No time estimates

Financial Services Context

Financial services training requires:

Compliance Focus

• Audit trail awareness throughout
• Override documentation requirements
• Regulatory context for controls

Error Recognition

• Staff must identify AI malfunctions
• Clear escalation for system issues
• Pattern recognition training

Assessment Rigor

• Practical assessment, not just knowledge
• Certification required before access
• Recertification schedule

Training Package Checklist

Before finalizing:

• Each role has distinct module
• Tasks are step-by-step procedures (not feature descriptions)
• Anti-patterns listed per role
• Decision aids provided (ready to print)
• Error recognition training included
• Compliance integrated into procedures
• Assessment criteria defined
• Time estimates provided
• Quick reference materials specified
• Deployment plan included