Relayby Switch Labs
Back to User Guide

AI Feature Best Practices

Get the most value from AI-powered features.

Philosophy

AI should assist, not replace:

  • You make final decisions
  • AI provides suggestions
  • Human judgment is key
  • AI saves time, you add value

AI Summaries

When They Help Most

  • Long conversation threads
  • Complex multi-issue tickets
  • Returning to old tickets
  • Escalations with history

Using Summaries Effectively

  1. Skim the summary first

    • Get overall context
    • Identify key issues
    • Note customer sentiment

  2. Dive deeper when needed

    • Summary may miss nuance
    • Read full thread for complex issues
    • Don't rely solely on summary

  3. Check for accuracy

    • Summaries are usually right
    • But verify critical details
    • Especially dates, numbers, names

Pro Tip

Use summaries for context when:

  • Ticket was handled by colleague
  • Long time since last reply
  • Multiple issues in one thread

Reply Suggestions

Best Use Cases

  • Common questions
  • Standard procedures
  • Starting point for responses
  • Learning new topics

How to Use Well

  1. Review before using

    • Never send without reading
    • Check it answers the question
    • Verify accuracy

  2. Personalize

    • Add customer's specific details
    • Adjust tone for situation
    • Add context only you know

  3. Edit freely

    • Suggestions are starting points
    • Combine with your knowledge
    • Make it sound like you

When to Skip AI Suggestions

  • Emotional situations (frustrated customer)
  • Sensitive topics
  • Very unique situations
  • When you know the answer already

Improving Suggestions

AI learns from your choices:

  • Use good suggestions
  • Modify okay suggestions
  • Skip poor suggestions
  • Quality improves over time

Sentiment Analysis

Understanding Sentiment

AI detects customer emotions:

  • Positive: Happy, satisfied
  • Neutral: Informational, matter-of-fact
  • Negative: Frustrated, upset, angry

Using Sentiment

  1. Prioritize upset customers

    • Negative sentiment → extra care
    • May need faster response
    • Softer, more empathetic tone

  2. Spot escalation risks

    • Sentiment declining → attention needed
    • Multiple negative tickets → VIP treatment
    • Save relationships proactively

  3. Recognize opportunities

    • Positive customers → ask for review
    • Satisfied resolution → upsell possibility

Don't Over-Rely

  • Sentiment isn't always accurate
  • Sarcasm is hard to detect
  • Context matters
  • Trust your judgment too

Priority Predictions

How It Works

AI suggests priority based on:

  • Language used
  • Issue type
  • Customer history
  • Urgency indicators

Using Priority Predictions

  1. As a starting point

    • Review AI suggestion
    • Adjust based on judgment
    • Your decision is final

  2. Spot urgent issues

    • AI catches "ASAP", "urgent", "down"
    • Helps surface critical tickets
    • Faster than manual scanning

  3. Override when needed

    • You know the customer better
    • You understand business context
    • AI can't see everything

Tag Suggestions

Using Suggested Tags

  1. Review suggestions

    • Usually relevant
    • May miss some
    • May suggest extras

  2. Accept accurate tags

    • One click to add
    • Saves typing
    • Consistent tagging

  3. Add missing tags

    • AI doesn't know everything
    • Add specific tags
    • Improve categorization

Improving Tag Suggestions

  • Use tags consistently
  • Create clear tag definitions
  • Remove unused tags
  • AI learns from patterns

Article Recommendations

When Articles Help

  • Customer asking common question
  • Issue documented in KB
  • Self-service could resolve

Using Recommendations

  1. Check relevance

    • Is article actually helpful?
    • Does it answer their question?
    • Is it current?

  2. Share appropriately

    • Include article link
    • Summarize key points
    • Add personal context

  3. Don't just send links

    • Bad: "See this article"
    • Good: "Here's how to reset your password, with detailed steps: [link]"

Improving Recommendations

  • Write clear article titles
  • Use relevant keywords
  • Categorize properly
  • Update outdated content

Similar Tickets

Use Cases

  • Finding duplicates
  • Learning from past solutions
  • Checking for patterns

Using Similar Tickets

  1. Check for duplicates

    • Same customer, same issue?
    • Merge if appropriate
    • Avoid duplicate effort

  2. Learn from history

    • How was similar issue solved?
    • What worked well?
    • What to avoid?

  3. Spot patterns

    • Multiple similar tickets = bigger issue
    • May indicate bug or documentation gap
    • Report patterns to team

General AI Tips

Trust But Verify

  • AI is helpful, not infallible
  • Verify important details
  • Use your expertise
  • You're responsible for responses

Provide Feedback

When AI is wrong:

  • Skip the suggestion
  • This trains the system
  • Accuracy improves over time

Know the Limits

AI struggles with:

  • Company-specific knowledge
  • Very recent events
  • Sarcasm and nuance
  • Unique situations

Efficiency, Not Replacement

Use AI to:

  • Get started faster
  • Handle routine tasks
  • Catch things you'd miss
  • Focus on what matters

Don't use AI to:

  • Avoid thinking
  • Replace empathy
  • Skip verification
  • Handle everything

Measuring AI Value

Track Time Savings

Notice how AI helps:

  • Faster ticket handling
  • Quicker context gathering
  • More consistent responses

Quality Metrics

Monitor:

  • CSAT with AI-assisted responses
  • First contact resolution
  • Response accuracy

Continuous Improvement

  • Review what works
  • Adjust what doesn't
  • Share tips with team
  • Suggest improvements

← Inbox Zero | Team Collaboration →