> ## Documentation Index
> Fetch the complete documentation index at: https://help.craftcx.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Failure Reason Clusters

> Identify and fix systemic AI support issues with intelligent failure pattern detection

Failure Reason Clusters automatically group similar AI support failures to help you quickly identify and resolve systemic issues. Instead of analyzing failures one by one, see patterns across multiple conversations and fix root causes that impact many customers.

## Key Benefits

### **Systemic Issue Detection**

Spot recurring problems before they become major customer experience issues. When your AI struggles with similar topics repeatedly, clusters surface these patterns immediately so you can take action.

### **Prioritized Improvement Roadmap**

Focus your optimization efforts on failures that affect the most customers. Clusters are ranked by impact, showing you which issues to tackle first for maximum improvement.

### **Root Cause Analysis**

Drill into representative conversations within each cluster to understand exactly why failures are happening. This targeted analysis saves time and leads to more effective solutions.

### **Measurable Impact Tracking**

Monitor how fixing clustered issues improves your overall AXIS scores. See the direct correlation between your optimizations and customer experience improvements.

## How Failure Clustering Works

Our AI analyzes every low-scoring conversation to identify common failure patterns. When multiple conversations fail for similar reasons, they're automatically grouped into clusters that reveal systemic issues.

<Tabs>
  <Tab title="Pattern Recognition">
    The system identifies failures across all three AXIS dimensions:

    **Resolution Accuracy Issues:**

    * Outdated knowledge base articles
    * Missing information on specific topics
    * Incorrect routing decisions

    **Interaction Effort Problems:**

    * Repetitive clarification requests
    * Inefficient conversation flows
    * Complex multi-step processes

    **Handoff Smoothness Failures:**

    * Poor escalation triggers
    * Inadequate context transfer
    * Timing issues with human takeover
  </Tab>

  <Tab title="Cluster Formation">
    Similar failures are grouped by:

    * **Topic similarity** - Issues related to the same product area
    * **Failure type** - Common patterns in how conversations break down
    * **Impact level** - How severely the failure affects customer experience
    * **Frequency** - How often the same issue occurs

    Each cluster includes representative conversations that best illustrate the problem.
  </Tab>
</Tabs>

## Understanding Your Clusters

### Cluster Priority Levels

**High Impact Clusters**

* Affect many customers frequently
* Cause significant AXIS score reductions
* Often involve critical business processes

**Medium Impact Clusters**

* Moderate frequency or severity
* May be product-specific issues
* Good targets for iterative improvement

**Low Impact Clusters**

* Edge cases or rare scenarios
* Minor score impacts
* Consider for future optimization rounds

### Cluster Insights

Each cluster provides:

* **Failure summary** - Common thread across conversations
* **Affected conversation count** - Scale of the problem
* **Average impact** - How much scores are affected
* **Representative examples** - Best conversations to review for context

## Using Clusters to Improve

### Step 1: Review High-Impact Clusters First

Start with clusters affecting the most customers or causing the largest score drops. These offer the biggest improvement opportunities.

### Step 2: Analyze Representative Conversations

Examine 2-3 conversations from each priority cluster to understand the failure pattern. Look for:

* What information was missing or incorrect
* Where the conversation went off track
* How the failure could have been prevented

### Step 3: Implement Targeted Fixes

Common solutions include:

* **Knowledge base updates** - Fix outdated or missing content
* **Training improvements** - Address AI knowledge gaps
* **Process optimization** - Streamline complex workflows
* **Escalation refinement** - Improve handoff triggers and timing

### Step 4: Monitor Cluster Resolution

Track how your fixes affect cluster size and new conversation failures. Successful optimizations should reduce cluster growth or eliminate them entirely.

## Real-World Examples

<Accordion title="Outdated Documentation Cluster">
  **Scenario:** Multiple customers asking about a feature that changed significantly, but documentation wasn't updated.

  **Pattern:** AI consistently references old article, customers correct the information, leading to confusion and escalation.

  **Solution:** Update knowledge base article with current feature details and capabilities.

  **Result:** Cluster disappears, Resolution Accuracy scores improve for this topic area.
</Accordion>

<Accordion title="Complex Setup Process Cluster">
  **Scenario:** Customers struggling with multi-step account configuration, requiring excessive back-and-forth.

  **Pattern:** AI can answer individual setup questions but fails to guide customers through the complete process efficiently.

  **Solution:** Create step-by-step setup guide and train AI to reference it proactively.

  **Result:** Interaction Effort scores improve, fewer escalations needed.
</Accordion>

## Frequently Asked Questions

<Accordion title="How often are clusters updated?">
  Conversations will be assigned to clusters as they are analyzed. Clusters are recalculated regularly (at least once weekly) as new conversations are analyzed. You'll see new clusters emerge as patterns change.
</Accordion>

<Accordion title="Can I customize cluster detection?">
  The clustering algorithm uses standardized pattern recognition to ensure consistent, objective analysis. This provides reliable insights across different conversation types and time periods.
</Accordion>

<Accordion title="What if I don't see any clusters?">
  No clusters typically means your AI is performing consistently well without systemic issues, or that you have not yet analyzed enough conversations. Individual conversation failures may still occur, but they're isolated rather than part of larger patterns.
</Accordion>

<Accordion title="How do I know if my fixes worked?">
  Monitor cluster size and growth rate after implementing solutions. Successful fixes should reduce the number of new conversations joining a cluster, and may eliminate clusters entirely.
</Accordion>

## Next Steps

Ready to eliminate systemic AI support issues? Start by:

1. **Reviewing your highest-impact clusters** - Which patterns affect the most customers?
2. **Analyzing representative conversations** - What's the root cause of each cluster?
3. **Implementing targeted solutions** - Focus on fixes that address multiple conversations at once
4. **Tracking improvement progress** - Monitor how clusters shrink as you optimize

Failure Reason Clusters turn scattered conversation failures into actionable improvement opportunities, helping you systematically enhance your AI support quality.