Skip to main content

Converting Conversations to Cache

Learn how to preserve excellent chatbot responses by converting them into cached entries for faster, more consistent future responses.

Why Convert Conversations to Cache?​

Benefits of Caching Good Responses​

Speed: Cache responses in 50-500ms vs. 2-8 seconds for RAG search

Consistency: Same excellent answer every time

Cost Savings: Reduce AI API calls by 60-80%

Quality Control: Curate and verify best responses

User Experience: Instant answers to common questions

When to Convert a Conversation​

Convert conversations that are:

✅ High Quality

  • Accurate, verified information
  • Well-structured and clear
  • Comprehensive but concise
  • Received positive feedback

✅ Commonly Asked

  • Question asked 3+ times
  • Topic is frequently searched
  • Seasonal or recurring need

✅ Time-Sensitive

  • Response time > 5 seconds
  • Complex query that's slow to process
  • Users benefit from instant answer

✅ Consistent

  • Answer doesn't change often
  • Information is stable
  • Not dependent on current date/time

When NOT to Convert​

Avoid caching conversations that are:

❌ Inaccurate or Outdated

  • Wrong information
  • Superseded by recent changes
  • Needs verification

❌ Too Specific

  • One-time question
  • Unique user situation
  • Rarely asked topic

❌ Context-Dependent

  • Requires conversation history
  • Depends on "current date" or real-time info
  • Needs follow-up clarification

❌ Low Quality

  • Vague or incomplete answer
  • Poor formatting
  • Received negative feedback

How to Convert a Conversation to Cache​

Step-by-Step: Basic Conversion​

Steps:

  1. Find a High-Quality Conversation

    • Filter by Positive Feedback to find good candidates
    • Or browse recent conversations for excellent responses

  2. Click "View" to Review

    • Open the conversation detail modal
    • Read the full question and response
    • Verify accuracy and quality
    • Check sources used

  3. Click "Convert to Cache" Button

    • Look for button at the bottom of the modal
    • Label may be "Convert to Cache Entry" or "Cache This Response"

  4. Conversion Modal Opens

    • Pre-filled with conversation data:
      • Question: User's original question
      • Response: Chatbot's answer
      • Sources: Referenced documents
    • Other fields are empty (you'll fill them)

  5. Review and Edit the Question

    • The pre-filled question may need cleaning up
    • Remove typos or awkward phrasing
    • Rephrase into natural, clear language
    • Example:
      • Original: "what do i need before 3000"
      • Cleaned: "What are the prerequisites for HFIM 3000?"

  6. Review and Edit the Response

    • Verify accuracy
    • Fix any formatting issues
    • Add or remove details as needed
    • Ensure professional tone

  7. Add Question Variations

    • Click "Add Variation" or enter in text field (one per line)
    • Add 3-5 variations of how users might ask this question
    • Example variations: 
      What courses do I need before HFIM 3000?
      HFIM 3000 prerequisites
      What do I need to take before 3000?
      Requirements for HFIM 3000

  8. Set Confidence Score

    • Choose based on accuracy:
      • 0.95-1.0: Verified, official information
      • 0.85-0.94: Accurate, stable information
      • 0.70-0.84: Helpful but may vary slightly
    • Start with 0.85 if unsure

  9. Set TTL (Time-To-Live)

    • How long before this entry needs review:
      • 7 days: Frequently changing info (deadlines, current semester)
      • 30 days: Occasionally updated info (schedules, contacts)
      • 90 days: Stable, permanent info (requirements, history)
    • Start with 30 days if unsure

  10. Set Status to "Active"

    • Ensure the entry will be used immediately

  11. Add Admin Notes (Optional but Recommended)

    • Document the source of this entry
    • Example: "Converted from conversation #142 on 1/12/26. User gave positive feedback. Source: HFIM Handbook 2026, page 12."

  12. Click "Save" or "Create Cache Entry"

    • Modal closes
    • Success message appears
    • New cache entry is created

  13. Verify the Cache Entry

    • Go to Cache Management
    • Find your new entry (sort by "Last Updated" or search)
    • Confirm it was saved correctly

Advanced: Improving During Conversion​

Enhancing the Question​

Original user questions are often messy:

  • Typos, misspellings
  • Informal language
  • Unclear phrasing
  • Too vague or too specific

Your job: Clean it up while preserving intent

Examples:

Original QuestionImproved Version
"what do i need before 3000""What are the prerequisites for HFIM 3000?"
"tell me bout internships""What internship opportunities are available for HFIM students?"
"whos the prof for 3000""Who teaches HFIM 3000?"
"how hard is the major""Is the HFIM major difficult?"

Enhancing the Response​

The chatbot's response may need refinement:

  • Remove redundant information
  • Fix formatting issues
  • Add clarity where needed
  • Improve structure

Before (verbose, redundant):

HFIM 3000 has prerequisites that you need to complete before you can take it.
The prerequisites for HFIM 3000 are HFIM 2100 and HFIM 2200. You need to
complete both of these courses. HFIM 2100 is Introduction to Hospitality
Management and HFIM 2200 is Food Service Operations. Both courses must be
completed with a grade of C or better before enrolling in HFIM 3000.

After (concise, structured):

To enroll in HFIM 3000, you must complete these prerequisites with a grade of
C or better:

• HFIM 2100 - Introduction to Hospitality Management
• HFIM 2200 - Food Service Operations

Additionally, you need junior standing (60+ credit hours).

Adding Strategic Variations​

Good variations match how users actually ask:

  1. Different Wording

    • "What are the prerequisites for HFIM 3000?"
    • "What courses do I need before HFIM 3000?"
    • "Requirements for HFIM 3000"

  2. Informal Phrasing

    • "what do i need before 3000"
    • "what comes before HFIM 3000"
    • "do i need anything before taking 3000"

  3. Abbreviations

    • "HFIM 3000 prereqs"
    • "3000 prerequisites"
    • "prereqs for hospitality 3000"

  4. Question Formats

    • "What are the prerequisites for HFIM 3000?" (question)
    • "HFIM 3000 prerequisites" (statement)
    • "Tell me about HFIM 3000 requirements" (command)

Goal: 5-15 variations covering common phrasings

Common Conversion Scenarios​

Scenario 1: Converting a Positive Feedback Conversation​

Situation: User asked "What is the HFIM program?" and gave positive feedback

Steps:

  1. Filter by Positive Feedback
  2. Find the conversation
  3. Click View → Convert to Cache
  4. Clean up the question (capitalize, fix punctuation)
  5. Review response quality
  6. Add 7 variations: 
    What is HFIM?
    Tell me about the HFIM program
    What does HFIM stand for?
    What is Hospitality and Food Industry Management?
    Explain the HFIM program
    HFIM program overview
    What is the Hospitality program at UGA?
  7. Set confidence: 0.90 (official program description)
  8. Set TTL: 90 days (program description doesn't change often)
  9. Add admin note: "Converted from conversation #142 with positive feedback"
  10. Save

Scenario 2: Converting a Frequently Asked Question​

Situation: 5 users asked variations of "Who teaches HFIM 3000?" in one week

Steps:

  1. Review all 5 conversations
  2. Pick the best response (most accurate and well-formatted)
  3. Click Convert to Cache
  4. Use the clearest question: "Who teaches HFIM 3000?"
  5. Ensure response includes current instructor name
  6. Add variations from all 5 conversations: 
    Who teaches HFIM 3000?
    Who is the professor for HFIM 3000?
    Who's the instructor for 3000?
    HFIM 3000 professor
    Who will be teaching HFIM 3000 this semester?
  7. Set confidence: 0.85 (accurate but may change each semester)
  8. Set TTL: 30 days (instructor may change)
  9. Add admin note: "Created from 5 similar conversations (IDs #142, #145, #148, #151, #153) in week of 1/12/26"
  10. Save

Scenario 3: Converting a Slow Response​

Situation: Question about "HFIM career paths" consistently takes 8-12 seconds to answer

Steps:

  1. Find a conversation with good response (positive or no feedback, accurate answer)
  2. Click Convert to Cache
  3. Use clear question: "What career paths are available with an HFIM degree?"
  4. Review response - ensure it's comprehensive
  5. Add variations: 
    HFIM career options
    What jobs can I get with an HFIM degree?
    Career paths for HFIM graduates
    What can I do with a hospitality degree?
    HFIM job opportunities
  6. Set confidence: 0.80 (general career info, may vary by year)
  7. Set TTL: 30 days (career landscape changes)
  8. Add admin note: "Created to reduce response time from 10s to < 1s for common question"
  9. Save
  10. Verify improvement: After a week, check if this question now returns in < 1 second

After Conversion: What to Do Next​

Immediate Actions​

1. Test the Cache Entry

  • Go to chatbot interface (user-facing site)
  • Ask the exact cached question
  • Verify response is instant (< 1 second)
  • Verify response is correct

2. Generate Variations

  • Go to Cache Management
  • Find your new entry
  • Click Edit → Generate Variations (AI)
  • Review and refine generated variations
  • Save

3. Monitor Performance

  • Check Times Served after 1 week
  • If 0, review variations and question phrasing
  • If 10+, success!

Weekly Follow-Up​

After 7 days:

  1. Go to Cache Management
  2. Find your cache entry
  3. Check Times Served metric
  4. Check user feedback (if tracked separately)

If Times Served = 0:

  • Variations don't match user language
  • Question is too specific (not actually common)
  • Another cache entry is matching instead

Action: Add better variations or deactivate

If Times Served > 10:

  • Success! Entry is being used
  • Monitor for negative feedback
  • Update TTL if needed

Best Practices​

1. Convert Strategically, Not Excessively​

Quality over Quantity:

  • Better to have 20 excellent cache entries than 100 mediocre ones
  • Focus on high-traffic questions first
  • Aim for 60-80% cache hit rate, not 100%

Priority Order:

  1. Frequently asked questions (5+ times)
  2. Slow responses that can be cached (> 5 seconds)
  3. Excellent responses with positive feedback
  4. Seasonal questions (application deadlines, start dates)

2. Clean Up User Language​

Users type informally:

  • Fix typos and misspellings
  • Capitalize properly
  • Add punctuation
  • Make questions clear and professional

But keep variations informal:

  • Cache question: "What are the prerequisites for HFIM 3000?"
  • Variation: "what do i need before 3000" (matches user typing)

3. Verify Accuracy Before Caching​

Never cache unverified information:

  • Cross-check with official sources
  • Verify dates and requirements
  • Update outdated information
  • Ensure response answers the question

If uncertain:

  • Set lower confidence (0.70-0.80)
  • Set shorter TTL (7-30 days)
  • Add admin note: "Needs verification - review by [date]"

4. Set Appropriate TTL​

Match TTL to information stability:

Information TypeExampleTTL
Frequently changingApplication deadlines, current semester dates7 days
Occasionally updatedCourse schedules, faculty office hours30 days
StableProgram requirements (rarely change)90 days
PermanentProgram history, degree types, mission statement90 days

Troubleshooting​

Problem: "Convert to Cache" Button Missing​

Possible Causes:

  • Feature not enabled in your admin panel
  • Conversation already cached
  • Insufficient permissions

Solutions:

  1. Check if you're in the conversation detail modal (not just table row)
  2. Contact support to enable feature
  3. Manually create cache entry (go to Cache Management → Create New)

Problem: Converted Entry Not Matching Users​

Cause: Variations don't match how users actually ask

Solution:

  1. Review Conversations for similar questions
  2. Note exact user phrasing
  3. Edit cache entry → Add those variations
  4. Generate AI variations for more options
  5. Wait 1 week and check Times Served

Problem: Response Too Long After Conversion​

Cause: Chatbot generated verbose answer

Solution:

  1. Edit the cache entry
  2. Trim response to 100-300 words
  3. Use bullet points for structure
  4. Remove redundant information
  5. Save changes

Next Steps​

Now that you can convert conversations to cache:

  1. Analyze interaction patterns - Identify more conversion opportunities
  2. Troubleshoot common issues - Solve problems efficiently
  3. Review cache best practices - Optimize your cache strategy

Remember: Converting one excellent conversation can help hundreds of future users. Spend time finding and caching your best responses!

Last updated on