Skip to main content

Best Practices

Comprehensive guidelines for effective prompt template management.

Testing Best Practices​

Pre-Deployment Testing​

✅ Always Test:

  1. Happy Path: Typical user inputs
  2. Edge Cases: Empty, null, very long
  3. Special Characters: <>, &, ", '
  4. Unicode: Non-English characters
  5. Consistency: Multiple runs with same input
  6. Performance: Response time and token count

Testing Checklist​

Before activating:

  • ✅ Tested with realistic production data
  • ✅ Tested with edge cases (empty, null, long)
  • ✅ All variables substitute correctly
  • ✅ Token count is reasonable (<800)
  • ✅ Tested with different AI models (if applicable)
  • ✅ Output format matches expectations
  • ✅ Required fields validated
  • ✅ Optional fields handle null gracefully
  • ✅ Special characters handled correctly
  • ✅ Tested multiple times for consistency
  • ✅ Performance is acceptable (<3s)
  • ✅ Cost per execution within budget

Token Optimization Best Practices​

Keep Templates Concise​

❌ Verbose:

You are a helpful customer service agent.
Please help the customer with their question.
Be helpful and professional in your response.
Please make sure to address all of their concerns.

Tokens: ~30

✅ Concise:

Help customer professionally.
Address all concerns.

Tokens: ~7

Savings: 77%

Token Targets​

Template TypeTargetMax
Simple<300500
Standard<600900
Complex<9001500

See: Token Optimization for strategies

Monitoring Best Practices​

Regular Review Schedule​

Daily:

  • ✅ Check for execution errors
  • ✅ Monitor unusual token spikes
  • ✅ Verify template is working

Weekly:

  • ✅ Review token usage trends
  • ✅ Compare performance across versions
  • ✅ Identify optimization opportunities
  • ✅ Check response quality

Monthly:

  • ✅ Analyze cost vs. performance
  • ✅ Update templates based on learnings
  • ✅ Archive unused templates
  • ✅ Plan optimization initiatives

See: Execution History for monitoring guide

Version Management Best Practices​

Version Comparison​

Before switching versions:

┌─────────────────────────────────────────────────────────────┐
│  Compare Versions: v2 vs v3                                 │
├─────────────────────────────────────────────────────────────┤
│                                                              │
│  ┌─ v2 (Previous) ──────────────┬─ v3 (Current) ─────────┐ │
│  │                               │                         │ │
│  │ Avg Tokens: 800               │ Avg Tokens: 600        │ │
│  │ Avg Time: 2.1s                │ Avg Time: 1.8s         │ │
│  │ Executions: 1,247             │ Executions: 342        │ │
│  │                               │                         │ │
│  │ Template Length: 450 chars    │ Length: 320 chars      │ │
│  │ Variables: 5                  │ Variables: 4           │ │
│  │                               │                         │ │
│  └───────────────────────────────┴─────────────────────────┘ │
│                                                              │
│  📊 v3 is 25% faster and 25% cheaper                        │
└─────────────────────────────────────────────────────────────┘

Evaluate:

  1. Token count difference
  2. Response time change
  3. Output quality
  4. Cost impact
  5. User feedback

Prompt Engineering Best Practices​

Workflow​

Prompt Structure​

Effective Template Structure:

1. Role/Context (1 line)
2. Task (1-2 lines)
3. Constraints (optional, 1-2 lines)
4. Output Format (optional, 1 line)

Example:

You are a customer service expert.
Help {!Customer_Name} with {!Issue_Type}.
Be professional and concise.
Provide actionable steps.

Security Best Practices​

✅ Do's​

  • ✅ Validate all input data
  • ✅ Sanitize special characters
  • ✅ Use type-safe variables
  • ✅ Test with malicious inputs
  • ✅ Limit template permissions

❌ Don'ts​

  • ❌ Trust user input blindly
  • ❌ Allow template injection
  • ❌ Expose sensitive data in prompts
  • ❌ Store credentials in templates
  • ❌ Allow unrestricted variable access

Performance Best Practices​

Response Time​

Targets:

  • Simple templates: less than 1s
  • Standard templates: less than 2s
  • Complex templates: less than 3s

If Slow:

  1. Reduce template length
  2. Use smaller AI model
  3. Remove unnecessary context
  4. Optimize variable lookups

Token Efficiency​

Track:

  • Average tokens per execution
  • Token cost trend over time
  • High-token outliers

Optimize:

  • Remove redundant text
  • Use concise language
  • Limit examples to 2-3

Documentation Best Practices​

Template Description​

✅ Good Description:

Helps customers with billing inquiries.
Uses customer name and account details.
Returns step-by-step resolution.

❌ Bad Description:

Customer template

Variable Documentation​

For Each Variable:

  • Purpose: What is it for?
  • Format: Expected format
  • Examples: Sample values
  • Default: Default value if any

Deployment Best Practices​

Pre-Deployment Checklist​

  • All tests pass
  • Token count optimized
  • Documentation complete
  • Peer review done
  • Security validated
  • Performance acceptable

Rollout Strategy​

  1. Test in Sandbox first
  2. Pilot with Small Group (10% of users)
  3. Monitor Closely (daily checks)
  4. Gradual Rollout (25% → 50% → 100%)
  5. Keep Previous Version ready for rollback

Rollback Plan​

If issues arise:

  1. Identify the problem quickly
  2. Revert to previous version
  3. Fix issue in new version
  4. Re-test thoroughly
  5. Re-deploy when stable

Common Pitfalls to Avoid​

❌ Don't​

  1. Skip Edge Case Testing

    • Always test empty, null, long values

  2. Ignore Token Costs

    • Monitor and optimize regularly

  3. Deploy Without Testing

    • Test in sandbox first

  4. Make All Fields Required

    • Only require truly necessary fields

  5. Use Vague Variable Names

    • Use clear, descriptive names

  6. Forget to Monitor

    • Check execution history regularly

  7. Over-Engineer Templates

    • Keep prompts simple and focused

  8. Neglect Version Comparison

    • Always compare before switching

Success Metrics​

Track These KPIs​

MetricTargetAction If Off
Token Count<800Optimize template
Response Time<2sSimplify or use smaller model
Error Rate<1%Fix validation issues
Cost/1K CallsVariesOptimize tokens
User Satisfaction>90%Improve output quality

Following these best practices ensures reliable, cost-effective templates that deliver high-quality results.