​

Best Practices

​

Knowledge Gathering

​

Cast a Wide Net First

• Request all documentation, even if it seems tangential
• Interview subject matter experts
• Document tribal knowledge
• Include edge cases and exceptions

​

Create a Knowledge Checklist

□ Training materials (all departments)
□ SOPs and process docs
□ Policy documents
□ Templates and macros
□ FAQ documents
□ System documentation
□ Tribal knowledge documented
□ Edge cases noted
□ Stakeholder sign-off

​

Tag Everything

• source:training, source:wiki, source:interview
• status:verified, status:needs-review
• topic:orders, topic:returns

​

Hierarchy Design

​

Start with the 80/20 Rule

1. Identify the most frequent inquiry types
2. Build hierarchy for those first
3. Add edge cases and exceptions later

​

Use Consistent Naming

Pattern: [Action] [Subject] ([Qualifier])

Examples:
✓ Cancel Order (Before Shipping)
✓ Process Return (Standard)
✓ Reset Password (After Lockout)

Not:
✗ Order Cancellation
✗ When customer wants return
✗ password reset process

​

Design for the Future

• Avoid overly specific scenario names
• Keep workstreams broad enough to expand
• Consider where new scenarios might fit

​

Mapping

​

Map as You Go

• Link documents to scenarios as you create them
• Review mappings when adding new knowledge
• Update mappings when scenarios change

​

Use a Coverage Matrix

​

Link Conservatively

• Would someone handling this scenario actually reference this?
• Does this document provide essential information?
• Is this a primary source or just tangentially related?

​

Writing Step Guides

​

Use Templates

# [Scenario Name]

## When to Use
[Brief description]

## Prerequisites
- [Requirement 1]
- [Requirement 2]

## Steps
1. [First action]
2. [Second action]
3. [Decision point]

## If [Condition]
[Alternative path]

## Escalation
[When and to whom]

## Related
- [Link 1]
- [Link 2]

​

Write for AI, Not Humans

​

Include Decision Trees

IF order has not shipped:
  → Process immediate cancellation
  → Initiate full refund

IF order has shipped but not delivered:
  → Cannot cancel
  → Offer return once delivered
  → Provide tracking info

IF order is delivered:
  → Redirect to return process

​

Test Your Guides

1. Walk through it yourself step by step
2. Have someone unfamiliar test it
3. Run it through the AI agent
4. Note any confusion or failures
5. Refine and repeat

​

Team Collaboration

​

Assign Ownership

• Who is responsible for maintaining it?
• Who reviews changes?
• Who approves new content?

​

Establish Review Processes

• SME review for accuracy
• Peer review for clarity
• Test for AI effectiveness

​

Document Your Decisions

• “Split this into two scenarios because the refund process differs”
• “Combined these because the procedure is identical”
• “Named this way to match the client’s terminology”

​

Maintenance

​

Schedule Regular Reviews

• Weekly: Check for urgent updates
• Monthly: Review high-volume scenarios
• Quarterly: Full audit of hierarchy

​

Track What Changes

1. Update the Step Guide
2. Create a new version with clear comment
3. Notify relevant team members
4. Test the updated guide

​

Watch for Signals

• AI giving outdated information
• Customer complaints about incorrect guidance
• Policy changes announced
• New systems or tools deployed

​

Scaling Tips

​

Parallelize Work

• Different workstreams assigned to different people
• SMEs write guides in their area
• Central team reviews for consistency

​

Batch Similar Tasks

• Write all guides for one contact driver at once
• Map all documents of one type together
• Review all scenarios in one workstream together

​

Automate What You Can

• Template generation
• Consistency checking
• Coverage reporting
• Version tracking

​

Quick Reference