Best Practices

Learn how to get the most out of Hay with these proven strategies and expert tips.

Getting Started Right #

Start Small, Scale Smart #

Week 1: Foundation

• Create 1-2 agents maximum
• Upload 10-20 core FAQs
• Connect 1 primary channel
• Test thoroughly before going live

Week 2-3: Expansion

• Monitor and refine based on real conversations
• Add more documentation as gaps appear
• Create 2-3 essential playbooks
• Connect additional channels if needed

Month 2+: Optimization

• Analyze performance data
• Expand to edge cases
• Add specialized agents if needed
• Fine-tune based on patterns

Why this works:

• Learn the system gradually
• See what works before scaling
• Easier to troubleshoot
• Less overwhelming for team

Set Realistic Expectations #

What Hay can do amazingly:

• ✅ Answer FAQs instantly (80-90% resolution)
• ✅ Look up information from documents
• ✅ Follow structured workflows
• ✅ Handle high volumes without fatigue
• ✅ Work 24/7 in multiple languages

What Hay needs help with:

• ⚠️ Highly emotional situations
• ⚠️ Complex negotiations
• ⚠️ Unique edge cases not in documentation
• ⚠️ Situations requiring human judgment

Agent Configuration #

Writing Great Instructions #

Use the "Training a Person" approach:

Bad:

Help customers with orders.

Good:

You are a friendly customer support agent for [Company].

Your main responsibilities:
1. Answer questions about orders, shipping, and products
2. Look up order status when customers provide order numbers
3. Explain our return policy clearly and patiently
4. Escalate to humans when customers are frustrated

Tone: Friendly and professional
Always use the customer's name if you know it
If you're not 100% confident, say so and offer to connect them with a specialist

Agent Specialization Strategy #

Option 1: Single Generalist Agent
Best for:

• Small teams
• Simple product lines
• Under 100 conversations/day
• Similar types of questions

Option 2: Multiple Specialized Agents
Best for:

• Different departments (sales vs support)
• Multiple product lines
• High volume (500+ conversations/day)
• Distinct use cases

Example specialized setup:

First-Line Support Agent

• Handles: FAQs, basic troubleshooting
• Escalates: Technical bugs, billing issues
• Tone: Friendly and helpful

Technical Support Agent

• Handles: Complex technical issues, API questions
• Escalates: Bugs, feature requests
• Tone: Professional and detailed

Sales Agent

• Handles: Product questions, pricing, demos
• Escalates: Negotiation, custom deals
• Tone: Enthusiastic and consultative

Document Management #

The Golden Rules of Documentation #

1. Accuracy First

• One wrong answer destroys trust
• Verify all information before uploading
• Update immediately when things change
• Regular audits (monthly minimum)

2. Clarity Over Brevity
Good:

Q: What is your return policy?
A: We accept returns within 30 days of purchase. Items must be unused and in original packaging. To start a return, email [email protected] with your order number. We'll send a prepaid return label within 24 hours. Refunds are processed within 5-7 business days after we receive the item.

Not great:

Returns: 30 days, unused, email us.

3. Use Customer Language
Write how customers speak:

• ❌ "What is the SKU lookup procedure?"
• ✅ "How do I find my product number?"

4. Include Context

• ❌ "Ships in 3-5 days"
• ✅ "Standard shipping delivers in 3-5 business days after your order is processed. Orders placed before 2 PM EST ship same day. Weekend orders ship Monday."

Document Organization System #

Create a hierarchy:

📁 Policies
  - Return Policy
  - Shipping Policy
  - Privacy Policy

📁 Products
  📁 Product Line A
    - Features and specs
    - Setup guide
    - Common issues
  📁 Product Line B
    - Features and specs
    - Setup guide

📁 FAQs
  📁 Orders
    - Order status
    - Tracking
    - Changes and cancellations
  📁 Shipping
    - Delivery times
    - International shipping
    - Costs and fees

📁 Troubleshooting
  - Common errors
  - Reset procedures
  - When to escalate

Use consistent naming:

• "How to [Action]" - Process guides
• "[Topic] FAQ" - Question collections
• "[Product] Guide" - Product documentation
• "[Process] Policy" - Official policies

Document Maintenance Schedule #

Daily:

• Add new FAQs from yesterday's conversations
• Quick updates for urgent changes

Weekly:

• Review most-used documents
• Check for outdated information
• Add missing details discovered in conversations

Monthly:

• Full content audit
• Remove or archive unused documents
• Consolidate duplicates
• Verify all information is current

Quarterly:

• Deep reorganization if needed
• Major content refresh
• Update screenshots/images
• Rewrite unclear sections

Playbook Design #

Playbook Best Practices #

1. One Purpose Per Playbook

Good:

• "Welcome New Customers"
• "Process Refund Request"
• "Handle Shipping Delays"

Not ideal:

• "Handle All Customer Issues" (too broad)
• "Welcome and Process Refunds" (too much in one)

2. Clear Trigger Conditions

Good triggers:

• "greeting", "hello", "hi" → Welcome
• "refund", "money back" → Refund Request
• "where is my order" → Order Tracking

Vague triggers:

• "help" (too generic)
• "stuff" (unclear)

3. Step-by-Step Flow

Good structure:

Step 1: Greet and acknowledge
Step 2: Collect order number
Step 3: Verify order exists
Step 4: If found → provide status
Step 5: If not found → escalate with collected info

4. Define Escalation Points

In every playbook, clearly state:

Escalate to human if:
- Customer is angry or frustrated
- Information isn't in our system
- Requires policy exception
- After 3 failed attempts to resolve

Testing Playbooks #

Before activating:

1. Test the happy path:

   • Everything goes right
   • Customer provides all info
   • Issue is resolved

2. Test edge cases:

   • Customer doesn't provide required info
   • Information not found in system
   • Customer changes their mind mid-flow

3. Test escalation:

   • Verify escalation triggers work
   • Check handoff is smooth
   • Ensure context is preserved

4. Get team feedback:

   • Have team members test
   • Collect their observations
   • Refine based on feedback

Queue Management #

Response Time Targets #

Set and communicate clear SLAs:

Track and improve:

• Monitor average response times
• Identify bottlenecks
• Staff appropriately
• Celebrate hitting targets

Queue Workflow #

Morning routine (5 minutes):

1. Check queue size
2. Review any urgent (red) items
3. Claim 1-2 you can handle immediately
4. Respond within target time

Throughout day:

• Check queue every 30-60 minutes
• Respond to new escalations quickly
• Update status on longer issues
• Keep customers informed

End of day:

• Review remaining queue items
• Snooze items you're working on
• Note follow-ups needed
• Set reminders for tomorrow

Preventing Queue Buildup #

Proactive prevention:

1. Improve AI training:

   • Analyze why conversations escalate
   • Add documents for common escalation topics
   • Create playbooks for repetitive escalations

2. Set better triggers:

   • Adjust escalation sensitivity
   • Allow AI to handle more edge cases
   • Only escalate what truly needs human help

3. Empower AI:

   • Give clearer instructions
   • Provide more examples
   • Build confidence through documentation

4. Staff appropriately:

   • Know your peak times
   • Have coverage during busy periods
   • Consider time zones for global support

Analytics & Optimization #

Metrics That Matter #

Focus on these key metrics:

1. Resolution Rate

   • Target: 70-85%
   • Higher = AI handling more independently
   • Review: Weekly

2. Customer Satisfaction

   • Target: 85%+
   • Based on thumbs up/down ratings
   • Review: Daily

3. Escalation Rate

   • Target: 15-30%
   • Lower = more efficient AI
   • Review: Weekly

4. Response Time

   • Target: Under 5 seconds for AI
   • Should be consistent
   • Review: Weekly

Weekly Review Ritual #

Every Monday, spend 30 minutes:

1. Review last week's metrics

   • What improved?
   • What declined?
   • Any unusual patterns?

2. Read 10 conversations

   • 5 high-rated (learn what works)
   • 5 low-rated (learn what doesn't)

3. Identify one improvement

   • Add a document
   • Update an agent
   • Create a playbook
   • Fix a recurring issue

4. Track the change

   • Note what you changed
   • Monitor impact over next week
   • Keep if it works, revert if it doesn't

A/B Testing #

Test changes before rolling out everywhere:

Example: Testing agent instructions

1. Current state:

   • Agent A: Current instructions
   • Resolution rate: 75%

2. Create test variant:

   • Agent B: New instructions
   • Same documents, same playbooks

3. Run for 1 week:

   • Split traffic 50/50
   • Track metrics separately

4. Compare results:

   • Which has better resolution?
   • Which has higher satisfaction?
   • Are escalations different?

5. Roll out winner:

   • Apply winning approach
   • Measure continued performance

Integration Strategy #

Start With One Channel #

First integration should be:

• Your highest volume channel
• Where customers prefer to contact you
• Easiest to set up and test

Examples:

• E-commerce → Shopify + Web Chat
• SaaS → In-app chat + Email
• Local business → WhatsApp + Web Chat

Multi-Channel Rollout #

Phase 1: Primary Channel (Week 1-2)

• Set up and test thoroughly
• Train AI on common questions
• Monitor performance daily
• Achieve 70%+ resolution rate

Phase 2: Secondary Channel (Week 3-4)

• Add next channel
• Test with same agent
• Adjust tone if needed for channel
• Monitor cross-channel performance

Phase 3: Expansion (Month 2+)

• Add remaining channels
• Consider specialized agents per channel
• Optimize for channel-specific needs

Channel-Specific Optimization #

Web Chat:

• Quick, concise responses
• Use formatting (bold, lists)
• Proactive engagement
• Business hours focus

WhatsApp:

• Very brief messages
• Use emojis appropriately
• Quick replies preferred
• 24/7 availability

Email:

• More detailed, formal
• Proper formatting
• Complete information first time
• Professional tone

Team Training #

Onboarding New Team Members #

Day 1: Introduction

• Tour of dashboard
• Review organization's agents
• Read key documentation
• Watch AI handle conversations

Week 1: Supervised Practice

• Use Supervision Mode
• Review and approve responses
• Provide feedback
• Share best practices

Week 2: Independent Work

• Handle escalations independently
• Regular check-ins
• Review their conversations
• Answer questions

Ongoing: Continuous Improvement

• Weekly team reviews
• Share interesting cases
• Discuss improvement ideas
• Celebrate wins

Creating a Knowledge Culture #

Make documentation everyone's job:

1. "See something, document something"

   • Found a gap in docs? Fill it.
   • Customer asked new question? Add the answer.
   • Process changed? Update immediately.

2. Weekly knowledge sharing:

   • Team meeting to discuss learnings
   • Share interesting conversations
   • Update docs together
   • Celebrate knowledge contributions

3. Reward good documentation:

   • Recognize team members who improve docs
   • Track documentation contributions
   • Make it part of performance reviews

Security & Privacy #

Protecting Customer Data #

Do:

• ✅ Use strong passwords and 2FA
• ✅ Limit access to what each person needs
• ✅ Review user access quarterly
• ✅ Log out when leaving computer
• ✅ Use secure connections (HTTPS)

Don't:

• ❌ Share login credentials
• ❌ Access Hay from public computers
• ❌ Leave sensitive data visible on screen
• ❌ Email customer data unencrypted
• ❌ Take screenshots with customer PII

API Token Management #

Best practices:

• Create specific tokens for each integration
• Name tokens clearly ("Shopify Prod", "Test API")
• Rotate tokens quarterly
• Delete unused tokens immediately
• Never commit tokens to code
• Store tokens in secure vault

Compliance Maintenance #

Regular compliance checklist:

Monthly:

• ☐ Review data retention settings
• ☐ Process any privacy requests
• ☐ Check for unauthorized access
• ☐ Verify team member access is appropriate

Quarterly:

• ☐ Full security audit
• ☐ Update privacy documentation
• ☐ Review compliance with regulations
• ☐ Test data export/deletion procedures

Scaling Tips #

From 100 to 1,000 Conversations/Day #

What changes:

• Need better organization
• More specialized agents
• Stricter quality control
• Team workflows matter more

Key actions:

1. Better segmentation:

   • Route by topic/complexity
   • Specialized agents
   • Priority queues

2. Automation focus:

   • Every repetitive task → playbook
   • Reduce human involvement
   • Faster escalation resolution

3. Team structure:

   • Dedicated queue managers
   • Specialists for complex issues
   • Clear escalation paths

From 1,000 to 10,000+ Conversations/Day #

Enterprise considerations:

• Multiple teams
• Advanced routing
• Custom integrations
• Dedicated account management

Contact Hay for:

• Enterprise pricing
• Dedicated support
• Custom features
• Training and consulting

Common Mistakes to Avoid #

Mistake 1: Insufficient Documentation #

Problem: AI has nothing to reference
Solution: Start with comprehensive FAQs

Mistake 2: Over-complicating Too Soon #

Problem: Too many agents, playbooks, features at once
Solution: Start simple, add complexity gradually

Mistake 3: Set and Forget #

Problem: No ongoing optimization
Solution: Weekly reviews and continuous improvement

Mistake 4: Ignoring Analytics #

Problem: Flying blind, missing opportunities
Solution: Data-driven decision making

Mistake 5: No Human Backup #

Problem: Escalations go unanswered
Solution: Clear queue management process

Mistake 6: Testing in Production #

Problem: Customers experience issues
Solution: Use playground/test mode extensively

Mistake 7: Generic Agent Instructions #

Problem: Vague, unhelpful responses
Solution: Specific, detailed, example-rich instructions

Success Checklist #

Ready to excel with Hay? Check these boxes:

Foundation:

• ☐ Clear, specific agent instructions
• ☐ 20+ core FAQ documents
• ☐ At least 2-3 essential playbooks
• ☐ Team trained on queue management
• ☐ Regular review schedule established

Optimization:

• ☐ 70%+ resolution rate
• ☐ 85%+ customer satisfaction
• ☐ Sub-30 minute escalation response
• ☐ Weekly analytics reviews
• ☐ Continuous documentation updates

Excellence:

• ☐ 85%+ resolution rate
• ☐ 90%+ customer satisfaction
• ☐ Sub-15 minute escalation response
• ☐ Proactive improvement culture
• ☐ Knowledge sharing system

Next Steps #

• Implement these best practices gradually
• Track your improvements
• Share what works with your team
• Join the Hay community to learn from others

Remember: Great customer support is a journey, not a destination. Keep learning, keep improving, and keep delighting your customers!