Skip to main content
Suggestions are actionable recommendations for improving your agent’s performance, generated by analyzing patterns in your data.

How Suggestions Work

Duckie analyzes:
  • Common questions that lack good answers
  • Frequently escalated scenarios
  • Response patterns and quality
  • Knowledge gaps and usage
Then generates specific, actionable recommendations.

Types of Suggestions

Knowledge Article Suggestions

New articles to create based on common questions:
Suggested: Create article about “two-factor authentication setup” Reason: 45 conversations this week asked about 2FA with no matching knowledge. Impact: Could resolve ~40 conversations per week automatically.
Action: Click to create a pre-filled article.

Runbook/Workflow Suggestions

Process improvements based on execution patterns:
Suggested: Add step to verify email before password reset Reason: 30% of password reset escalations were due to unverified identity. Impact: Could reduce escalations by ~25%.

Guideline Suggestions

Communication improvements based on response analysis:
Suggested: Add guideline to always confirm the specific issue before troubleshooting Reason: Many conversations required clarification after initial response.

Guardrail Suggestions

Safety improvements:
Suggested: Add escalation rule for pricing negotiations Reason: 15 conversations about custom pricing were handled inconsistently.

Viewing Suggestions

1

Navigate to Suggestions

Go to Analyze → Suggestions.
2

Browse Suggestions

View pending suggestions sorted by impact.
3

Review Details

Click any suggestion for full context.

Suggestion Details

Each suggestion includes:
SectionContent
RecommendationWhat to do
ReasonWhy it’s suggested
EvidenceData supporting the suggestion
ImpactExpected improvement
Related conversationsExamples that drove this suggestion

Acting on Suggestions

Accept

Implement the suggestion:
1

Click Accept

Click the Accept button on the suggestion.
2

Implement

Follow the guided flow to implement:
  • Knowledge: Opens editor with pre-filled content
  • Runbook: Opens editor with suggested changes
  • Guideline: Opens settings with recommendation
3

Complete

Suggestion is marked as implemented.

Dismiss

Mark as not relevant:
  1. Click Dismiss
  2. Select a reason (optional)
  3. Suggestion won’t reappear

Save for Later

Keep for future consideration:
  1. Click Save for Later
  2. Suggestion moves to saved queue
  3. Review when ready

Suggestion Priority

Suggestions are ranked by:
FactorDescription
ImpactHow much improvement expected
ConfidenceHow certain the analysis is
FrequencyHow often the issue occurs
EffortHow much work to implement
High-impact, high-confidence, low-effort suggestions appear first.

Suggestion Sources

Where suggestions come from:

Knowledge Gaps

Questions the agent couldn’t answer well:
  • Reviewed in the Gaps tab
  • Cross-referenced with conversation patterns
  • Suggested when gap is significant

Escalation Analysis

Patterns in escalated conversations:
  • Common topics in escalations
  • Missing knowledge that caused escalation
  • Guardrail trigger patterns

Response Quality

Analysis of agent responses:
  • Conversations that needed clarification
  • Low-rated responses
  • Repeated customer follow-ups

Performance Data

Metrics that indicate issues:
  • Categories with low resolution
  • Increasing escalation trends
  • Response time outliers

Suggestion History

View past suggestions:
  • Implemented: Suggestions you accepted
  • Dismissed: Suggestions you skipped
  • Expired: Old suggestions no longer relevant

Next Steps