> For the complete documentation index, see [llms.txt](https://docs.pharos.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.pharos.ai/platform-overview/pharos-platform-architecture/ai-analysis/issue-analysis.md).

# Issue Analysis

How AI-powered Issue analysis turns error investigation from hours into minutes with structured, context-aware insights.

When errors occur in Salesforce, the path to resolution traditionally involves building reports, filtering list views, manually correlating logs, and piecing together context across multiple interfaces. Pharos AI analysis eliminates this friction by providing intelligent, structured insights that accelerate your journey from problem discovery to resolution.

{% hint style="success" %}
**Quick Path to Resolution**

AI analysis provides structured insights in seconds, helping you understand Issues, identify root causes, and determine next steps—without manual investigation through dashboards and reports.
{% endhint %}

***

## The Traditional Troubleshooting Challenge

Salesforce provides powerful tools for viewing and analyzing data—dashboards, reports, list views, and custom analytics. However, these tools present information in a general-purpose format that demands significant manual effort to navigate effectively.

### The Manual Approach

Manual troubleshooting follows a familiar sequence of steps, each of which costs time and demands specialized knowledge.

{% hint style="warning" %}
**The Fundamental Problem**

You need to know what you're looking for before you can find it. This creates a catch-22: troubleshooting requires answers, but finding those answers requires already understanding the problem.
{% endhint %}

**When Issues arise, traditional troubleshooting means:**

| Activity                    | Time Investment | Challenge                                              |
| --------------------------- | --------------- | ------------------------------------------------------ |
| 📊 **Building Reports**     | 15-30 minutes   | Requires knowing which fields and filters to use       |
| 🔍 **Filtering List Views** | 10-20 minutes   | Limited operations, rigid interfaces                   |
| 🔗 **Correlating Changes**  | 30-60 minutes   | Manual cross-referencing between logs and audit trails |
| ⏱️ **Determining Severity** | 15-30 minutes   | Manually assessing impact and prioritization           |
| 🔄 **Context Switching**    | Ongoing         | Navigating between multiple tools and interfaces       |

**⏱️ Total Investigation Time: 1-3 hours per Issue**

While these methods work, they consume time and require intimate knowledge of your data model, Salesforce reporting tools, and your org's specific configuration.

***

## How Pharos AI Analysis Works

Pharos AI analysis operates on a carefully curated dataset that provides complete context for intelligent troubleshooting, eliminating the manual correlation work.

### Data Foundation

AI analysis draws on three categories of data that together provide complete context.

<table><thead><tr><th width="200">Data Category</th><th>What's Included</th></tr></thead><tbody><tr><td><strong>📊 Observability Data</strong></td><td>• Log records from runtime errors and instrumented code<br>• Issue records aggregating related logs<br>• Occurrence patterns, frequency, and timing data<br>• User impact and affected record tracking</td></tr><tr><td><strong>⚙️ Metadata Context</strong></td><td>• Salesforce configuration changes over time<br>• Flow versions, Apex code updates, validation rules<br>• Component relationships and dependencies<br>• Deployment history and timing</td></tr><tr><td><strong>📈 Historical Patterns</strong></td><td>• Previous occurrences of similar Issues<br>• Resolution history and effectiveness<br>• Related error patterns<br>• Seasonal or cyclical trends</td></tr></tbody></table>

This combination allows AI analysis to understand not just **what's breaking**, but **why it might be breaking**, **what changed to cause it**, and **how similar problems were resolved in the past**.

{% hint style="info" %}
**Data Grounding**

Pharos AI is grounded in your actual org data—it doesn't generate speculative answers. Every insight is based on logs, Issues, and metadata changes from your environment. This ensures recommendations are practical and relevant to your specific situation.
{% endhint %}

***

## The Pharos Trust Layer

Security and privacy are paramount. Pharos operates with strict data boundaries that ensure your sensitive information never leaves your control.

### Data Protection Model

Pharos draws a clear line between the metadata it sends to AI and the business data it keeps entirely within your org. What crosses that line is error messages and stack traces, configuration structure, statistical aggregations, and audit trail metadata. What never does is actual record data, transaction details, and sensitive or restricted fields.

**Why this is safe:** the information sent describes system behavior and configuration, not your business data. It's analogous to sharing error logs with a support team — helpful for troubleshooting without revealing customer or transaction details. Your competitive data, customer information, and business intelligence remain entirely within your Salesforce org; the AI sees patterns and configurations, not your proprietary information.

Two design choices reinforce that boundary. Indexed data is stored in **vectorized form** — mathematical representations that are extremely difficult to reverse engineer — which puts a second layer beyond access controls in the way of unauthorized extraction. And AI analysis operates **inside your existing Salesforce security model** rather than adding a parallel one, so sharing rules, profiles, and permission sets determine who sees what, with nothing separate to administer.

{% hint style="info" %}
For the itemized boundary — exactly what is and isn't sent, per integration — see [Security & Data Access](https://docs.pharos.ai/configuration-guide/security.md#ai-data-handling).
{% endhint %}

***

## AI Analysis for Individual Issues

When examining a specific Issue, Pharos provides structured AI analysis designed to jumpstart your troubleshooting process. Every analysis follows a consistent format, making it easy to quickly extract the information you need.

{% hint style="info" %}
**Consistent Structure**

Every analysis includes the same components in the same order, allowing you to develop muscle memory for consuming insights. This consistency is especially valuable when triaging multiple Issues quickly.
{% endhint %}

### Analysis Components

Every analysis contains the same five components, always in the same order.

<table><thead><tr><th width="220">Component</th><th>Purpose &#x26; Value</th></tr></thead><tbody><tr><td>🎯 <strong>Severity Assessment</strong></td><td><strong>What it is:</strong> AI-driven criticality evaluation (Critical, High, Medium, Low)<br><strong>Based on:</strong> Occurrence patterns, user impact, affected data volume, business process criticality<br><strong>Why it matters:</strong> Helps prioritize which problems to tackle first in triaging workflows</td></tr><tr><td>💡 <strong>Problem Explanation</strong></td><td><strong>What it is:</strong> Clear description in accessible language<br><strong>Interprets:</strong> Stack traces, error codes, Salesforce-specific failures<br><strong>Why it matters:</strong> Makes technical errors understandable without requiring deep Salesforce expertise</td></tr><tr><td>🔍 <strong>Root Cause Indicators</strong></td><td><strong>What it is:</strong> Potential causes identified through correlation<br><strong>Analyzes:</strong> Recent metadata changes, related errors, historical patterns<br><strong>Why it matters:</strong> Points you toward likely causes instead of requiring hypothesis generation from scratch</td></tr><tr><td>⚠️ <strong>Impact Analysis</strong></td><td><strong>What it is:</strong> Consequences of not addressing the Issue<br><strong>Considers:</strong> Business process implications, user experience degradation, cascading effects<br><strong>Why it matters:</strong> Helps communicate urgency to stakeholders and justify prioritization decisions</td></tr><tr><td>✅ <strong>Recommended Next Steps</strong></td><td><strong>What it is:</strong> Practical, actionable guidance<br><strong>Tailored to:</strong> Your org's configuration, the specific error type, and available context<br><strong>Why it matters:</strong> Provides immediate direction without requiring extensive investigation</td></tr></tbody></table>

### Example Analysis Output

The tabs below show how these components look for a real Case Assignment failure.

{% tabs %}
{% tab title="🎯 Severity Assessment" %}

```
SEVERITY: HIGH

This issue warrants immediate attention due to:
• 156 occurrences affecting 47 users over 2 days
• Impacts critical "Case Assignment" business process
• Failure rate: 89% (blocking most case routing)
• User complaints detected via support channel

Recommended SLA: Resolve within 4 hours
```

{% endtab %}

{% tab title="💡 Problem Explanation" %}

```
PROBLEM:

The "Case Assignment Flow" is failing when attempting to assign 
cases to the Support queue. The error "NullPointerException" 
occurs because the Flow expects the Contact.Department field 
to have a value, but approximately 60% of incoming cases have 
Contacts with null Department fields.

In plain language: The automation that routes support cases 
is breaking when it tries to use department information that 
doesn't exist on the contact record.
```

{% endtab %}

{% tab title="🔍 Root Cause" %}

```
ROOT CAUSE INDICATORS:

✅ Primary Suspect:
   Flow "Case_Assignment_Flow_v3" modified on January 10 at 2:10 PM
   - Added new decision element checking Contact.Department
   - No null handling added for Department field
   - First failure occurred 5 minutes after deployment

⚠️ Contributing Factor:
   Data Quality: 60% of Contact records missing Department value
   This pre-existing condition became problematic after Flow change

📊 Pattern Analysis:
   Similar Issue resolved 3 months ago by adding null checks in 
   related Flow (Issue #1247)
```

{% endtab %}

{% tab title="⚠️ Impact Analysis" %}

```
IMPACT IF NOT RESOLVED:

Business Impact:
• Support cases not being routed to appropriate teams
• Manual assignment required (5 min per case × ~80 cases/day = 6.5 hours)
• SLA violations likely as cases sit unassigned

User Impact:
• Support agents not receiving cases automatically
• Customers experiencing delayed responses
• Increased escalations to management

Technical Impact:
• Issue not self-resolving (requires code/configuration fix)
• May worsen if case volume increases
• Could cascade to other Flows using same pattern
```

{% endtab %}

{% tab title="✅ Next Steps" %}

```
RECOMMENDED NEXT STEPS:

Immediate Fix (15 minutes):
1. Edit Case_Assignment_Flow_v3
2. Add null check before Department evaluation:
   - If Contact.Department IS NULL, route to "General Support" queue
3. Save and activate Flow
4. Test with sample cases

Medium-Term (Next Sprint):
5. Data Quality: Review why Department field frequently null
6. Consider making Department required or adding validation
7. Audit other Flows using Contact.Department for similar issues

Verification:
8. Monitor Issue for 24 hours to confirm resolution
9. Mark as "Completed" if no new occurrences
```

{% endtab %}
{% endtabs %}

***

## Leveraging Analysis Output

The AI analysis output is designed for practical use across your entire workflow, not just for viewing in Pharos.

### Integration with Your Team's Workflow

The structured output supports several ways of working, from sharing to guided investigation.

{% tabs %}
{% tab title="📋 Copy and Share" %}
**Structured Format = Easy Communication**

The consistent structure makes it easy to copy analysis results into:

* Jira or Azure DevOps tickets
* Incident documentation
* Team Slack channels
* Sprint planning notes
* Executive status reports

**Why this matters:**\
Context isn't lost when handing off Issues to developers or other team members. Everyone receives the same comprehensive understanding.
{% endtab %}

{% tab title="🎯 Investigation Starting Point" %}
**Guided Troubleshooting**

Use the analysis as a starting point for deeper investigation:

* Root cause indicators tell you where to look
* Recommended next steps provide a roadmap
* You maintain control over the resolution approach

**Why this matters:**\
You spend time validating a hypothesis instead of generating one from scratch, dramatically accelerating resolution.
{% endtab %}

{% tab title="📚 Knowledge Building" %}
**Institutional Learning**

Over time, analysis patterns help your team:

* Build knowledge about common failure modes
* Recognize recurring patterns quickly
* Develop best practices for fixes
* Train new team members effectively

**Why this matters:**\
Your team becomes more efficient with each Issue resolved, compounding the benefits of AI analysis over time.
{% endtab %}
{% endtabs %}

***

## Automating AI Analysis

While on-demand AI analysis is valuable, the real power emerges when you automate analysis generation and delivery, ensuring critical context reaches the right people at the right time without manual intervention.

{% hint style="success" %}
**Proactive Intelligence**

Automation transforms AI analysis from a reactive tool (you ask for it when you need it) into a proactive system (it arrives automatically when conditions warrant attention).
{% endhint %}

### Delivering Analysis to Collaboration Channels

Pharos automatically delivers AI analysis results as comments or messages in your team collaboration tools.

<table><thead><tr><th width="150">Platform</th><th>Delivery Method</th><th>Benefits</th></tr></thead><tbody><tr><td>💬 <strong>Slack</strong></td><td>Threaded messages to designated channels</td><td>Teams see severity, root causes, and recommendations without leaving their communication flow</td></tr><tr><td>📧 <strong>Microsoft Teams</strong></td><td>Adaptive cards or message posts</td><td>Formatted analysis in relevant channels keeps ops and dev teams informed as Issues emerge</td></tr><tr><td>🎫 <strong>Jira</strong></td><td>Automatic ticket comments</td><td>Enriches tickets with contextual intelligence; updates create a complete AI-assisted timeline</td></tr><tr><td>📧 <strong>Email</strong></td><td>Formatted email notifications</td><td>Delivers structured analysis to stakeholders who prefer email over chat platforms</td></tr></tbody></table>

{% hint style="success" %}
**Contextual Delivery**

Analysis is delivered where your team already works, eliminating context switching and ensuring insights are immediately actionable. No need to log into Pharos to understand what's happening—the intelligence comes to you.
{% endhint %}

→ [**Learn more about Notification Channels**](https://docs.pharos.ai/platform-overview/pharos-platform-architecture/alerting/notification-channels.md)

### Rule-Based Analysis Automation

Configure notification and alerting rules to trigger automatic analysis based on specific conditions, ensuring you receive intelligence proactively.

{% tabs %}
{% tab title="📊 Threshold-Based Triggers" %}
**Generate analysis when Issues cross thresholds:**

* Log count exceeds a certain number (e.g., more than 100 occurrences)
* User impact reaches concerning levels (e.g., >20 affected users)
* Frequency accelerates beyond normal patterns (e.g., 3x increase day-over-day)
* Issues remain unresolved for extended periods (e.g., >7 days in "New" status)

**Example Rule:**

```
WHEN Issue.Log_Count__c > 100
AND Issue.Status__c = 'New'
THEN Generate AI Analysis + Notify #engineering-alerts via Slack
```

**Why this works:**\
High-occurrence Issues get immediate attention with AI-generated context, ensuring no critical problem goes unnoticed.
{% endtab %}

{% tab title="🔄 Status Change Triggers" %}
**Automatically analyze when status changes:**

* New Issues entering the system (catch problems immediately)
* Issues regressing from "Completed" to "New" (flag regressions)
* Issues escalating in severity (alert on worsening conditions)
* Issues remaining "Actioned" too long (identify stalled work)

**Example Rule:**

```
WHEN Issue.Regression__c = true
THEN Generate AI Analysis + Create Jira ticket + Notify assigned developer
```

**Why this works:**\
Regressions are automatically investigated and escalated, preventing previously-fixed problems from slipping through the cracks.
{% endtab %}

{% tab title="⚡ Pattern Detection Triggers" %}
**Generate analysis when Pharos detects anomalies:**

* Sudden spikes in error rates (e.g., 5x increase in 1 hour)
* New error types appearing (previously unseen error signatures)
* Issues spreading across multiple functional areas
* Anomalies in occurrence timing (errors outside normal hours)

**Example Rule:**

```
WHEN Issue.First_Occurred_On__c > LAST_7_DAYS
AND Issue.Average_Daily_Volume__c > 10
THEN Generate AI Analysis + Notify #ops-oncall via PagerDuty
```

**Why this works:**\
Emerging problems are surfaced proactively with AI insights before they become critical incidents.
{% endtab %}
{% endtabs %}

{% hint style="info" %}
**Proactive Surfacing**

These rules ensure AI analysis happens automatically, surfacing insights before someone has to manually request them. This transforms your team's posture from reactive (responding to problems) to proactive (intercepting problems early).
{% endhint %}

→ [**Learn more about Issue-Level Alerting**](https://docs.pharos.ai/platform-overview/pharos-platform-architecture/alerting/issue-level-alerts.md)

### Flow-Based Analysis Invocation

For sophisticated automation scenarios, Pharos exposes analysis capabilities through invocable actions callable from Salesforce Flows.

{% hint style="success" %}
**Advanced Automation**

Flows enable complex, customized triaging workflows tailored to your organization's specific needs—far beyond what simple rules can accomplish.
{% endhint %}

#### Invocable Actions

Flows can invoke AI analysis on specific Issues, passing parameters to control analysis scope and delivery:

**Capabilities:**

* 🔧 Build custom triaging processes that evaluate Issues systematically
* 🔗 Chain analysis with other automated actions (status updates, field updates, notifications)
* 🎯 Create conditional logic determining when analysis is valuable
* 🚀 Integrate analysis into broader business process automation
* 📊 Batch analyze multiple Issues and aggregate results

#### Complex Triaging Scenarios

{% tabs %}
{% tab title="🌅 Automated Daily Triage" %}
**Scheduled Flow - Morning Review**

A Flow runs each morning at 8 AM, querying for new or updated high-severity Issues. For each Issue meeting criteria, it invokes AI analysis and routes accordingly:

**Logic:**

```
1. Query: Issues WHERE (Status = 'New' OR LastModified = LAST_24_HOURS)
          AND Priority IN ('Highest', 'High')
2. For Each Issue:
   a. Invoke AI Analysis
   b. If AI-assessed severity is critical:
      - Create Jira ticket with analysis
      - Notify on-call team via Slack
      - Set Issue status to 'Actioned'
   c. If AI Severity = 'High':
      - Add to triage queue
      - Post analysis to #dev-triage channel
   d. If AI Severity = 'Medium':
      - Tag for next sprint review
      - Update Issue description with analysis
```

**Value:**\
Team starts each day with critical Issues already analyzed, prioritized, and routed—no manual triage meeting required.
{% endtab %}

{% tab title="🚨 Smart Escalation" %}
**Record-Triggered Flow - Spike Detection**

Monitors Issue updates. When an Issue's occurrence count doubles within 24 hours:

**Logic:**

```
1. Trigger: WHEN Issue.Log_Count__c CHANGES
2. Calculate: Previous_24h_Count vs Current_24h_Count
3. IF Current_24h_Count > (Previous_24h_Count * 2):
   a. Invoke AI Analysis with "change correlation" parameter
   b. Check if metadata modifications correlate with spike
   c. Post urgent analysis to #incident-response channel
   d. IF Root_Cause includes "critical component":
      - Page senior engineering via PagerDuty
      - Create Jira ticket tagged "P0"
   e. ELSE:
      - Notify team lead via Slack
      - Create standard Jira ticket
```

**Value:**\
Escalating Issues are automatically investigated and routed based on severity, preventing small problems from becoming incidents.
{% endtab %}

{% tab title="🚀 Post-Deployment Verification" %}
**Scheduled Flow - Release Validation**

After deployments, checks for Issues created in the past hour and correlates with deployment:

**Logic:**

```
1. Trigger: Scheduled 1 hour after deployment window
2. Query: Issues WHERE First_Occurred_On__c = LAST_1_HOUR
3. Generate deployment baseline report (errors before deployment)
4. For Each New Issue:
   a. Invoke AI Analysis with "deployment correlation" parameter
   b. Compare against pre-deployment baseline
   c. IF likely caused by deployment:
      - Alert deployment team with analysis
      - Include rollback recommendations from AI
      - Create Jira ticket linked to deployment
5. IF 3+ Issues correlated with deployment:
   - Escalate to engineering leadership
   - Recommend considering rollback
```

**Value:**\
Deployment impact is assessed automatically, enabling rapid rollback decisions based on AI correlation analysis.
{% endtab %}

{% tab title="👥 Custom Assignment" %}
**Record-Triggered Flow - Intelligent Routing**

When Issues are marked for action, invokes analysis and uses insights for smart routing:

**Logic:**

```
1. Trigger: WHEN Issue.Status__c CHANGES TO 'Actioned'
2. Invoke AI Analysis
3. Parse AI Output:
   a. Extract: Affected_Component (e.g., "AccountTrigger", "Case Assignment Flow")
   b. Extract: Recommended_Team (e.g., "Platform", "Integration", "Admin")
   c. Extract: Severity
4. Route Issue:
   a. IF Component = Apex → Assign to Platform Team
   b. IF Component = Flow → Assign to Admin Team
   c. IF Component = Integration → Assign to Integration Team
5. Set Priority based on AI Severity:
   a. Critical → Priority 1
   b. High → Priority 2
   c. Medium → Priority 3
6. Populate Jira ticket template with AI analysis insights
7. Notify assignee via Slack with context already in place
```

**Value:**\
Issues are automatically routed to the right team with full context, eliminating manual triage and assignment overhead.
{% endtab %}
{% endtabs %}

{% hint style="warning" %}
**Governor Limits Consideration**

When designing complex automation, be mindful of Salesforce governor limits. AI analysis invocations count toward API limits, so design Flows to target specific Issue subsets (using filters and criteria) rather than broad, high-volume queries.

**Recommendation:** For high-volume scenarios, use scheduled Flows with batching instead of record-triggered Flows.
{% endhint %}

→ [**Learn more about Rule-Based Configuration**](https://docs.pharos.ai/platform-overview/pharos-platform-architecture/rule-based-configuration.md)

### Building a Comprehensive Automation Strategy

The most effective implementations combine multiple automation approaches in a layered strategy, ensuring AI intelligence flows through your entire incident management lifecycle.

<table><thead><tr><th width="200">Layer</th><th>Purpose</th><th>Example Implementation</th></tr></thead><tbody><tr><td>🎯 <strong>Baseline Rules</strong></td><td>Immediate notification for obvious problems</td><td>Alert on high occurrence counts (>100), critical errors, production failures</td></tr><tr><td>🔀 <strong>Flow-Based Logic</strong></td><td>Nuanced scenarios requiring conditional evaluation</td><td>Post-deployment validation, smart escalation, custom routing</td></tr><tr><td>📅 <strong>Scheduled Processes</strong></td><td>Regular review of Issue backlogs</td><td>Daily triage at 8 AM, weekly health checks, monthly trend analysis</td></tr><tr><td>🔔 <strong>Integration Delivery</strong></td><td>Insights reach the right channels automatically</td><td>Slack notifications, Jira enrichment, email digests, PagerDuty alerts</td></tr></tbody></table>

**How layers work together:**

1. **Baseline rules** catch obvious high-priority Issues and trigger immediate AI analysis
2. **Flow-based logic** evaluates nuanced conditions and makes intelligent routing decisions
3. **Scheduled processes** ensure nothing falls through the cracks with regular review
4. **Integration delivery** ensures insights reach the right people in their preferred tools

{% hint style="success" %}
**Compounding Benefits**

This layered approach means your team gets AI insights when they're most valuable—as Issues emerge, as patterns evolve, and as triaging decisions need to be made. Over time, this transforms your operational posture from reactive firefighting to proactive system stewardship.
{% endhint %}

### Monitoring Automation Effectiveness

Track your automation's impact to refine processes over time and demonstrate value to stakeholders.

**Key Metrics to Monitor:**

| Metric                               | What It Measures                              | Target Direction         |
| ------------------------------------ | --------------------------------------------- | ------------------------ |
| ⏱️ **Mean Time to First Response**   | How quickly team engages after Issue creation | ⬇️ Decrease              |
| 📊 **AI Analysis Usage Rate**        | % of Issues receiving AI analysis             | ⬆️ Increase              |
| 🎯 **Auto-Routing Accuracy**         | % of Issues routed to correct team            | ⬆️ Increase              |
| ✅ **Resolution Time (AI vs Manual)** | Time to resolve Issues with vs without AI     | ⬇️ AI should be faster   |
| 🔔 **Alert Action Rate**             | % of AI-triggered alerts acted upon           | ⬆️ High = good targeting |

{% hint style="info" %}
**Continuous Improvement**

Use this feedback to refine your automation rules and Flows, creating increasingly effective triaging processes. Low action rates might indicate too many low-value alerts; high resolution times might suggest AI recommendations need refinement or that Issues need better initial data.
{% endhint %}

***

## Org Genie: Conversational Analysis

Beyond structured Issue analysis, Pharos includes **Org Genie**—a conversational AI interface for asking freeform questions about your org's health and behavior.

### Two Contexts for Different Needs

Org Genie works in two contexts, each suited to a different kind of question.

{% tabs %}
{% tab title="🌐 Org-Level Analysis" %}
**Ask questions across your entire environment:**

* Generate visualizations showing error trends over time
* Identify outliers and anomalies in Issue patterns
* Compare time periods (this week vs last week)
* Explore relationships between different Issue types
* Prioritize which Issues deserve immediate attention

**Perfect for:**\
Health checks, weekly reviews, sprint planning, understanding overall patterns, and executive reporting.

**Example Questions:**

* "Show me errors over the last month broken down by category"
* "Which Issues from the last 7 days warrant immediate attention?"
* "Were there any anomalies this week compared to last week?"
  {% endtab %}

{% tab title="🔍 Issue-Level Investigation" %}
**Focus on specific problems with detailed questions:**

* Investigate a single Issue's occurrence history
* Correlate with related configuration changes
* Understand affected users and records
* Identify potential root causes specific to that Issue
* Determine if similar Issues exist elsewhere

**Perfect for:**\
Deep-dive troubleshooting, root cause analysis, understanding why an Issue started or stopped, and investigating regressions.

**Example Questions:**

* "This Issue stopped on December 17th. What changed that could have fixed it?"
* "Show me all metadata changes around when this Issue first appeared"
* "Are there related Issues that might have the same root cause?"
  {% endtab %}
  {% endtabs %}

{% hint style="success" %}
**Complementary Approaches**

Structured Issue analysis provides consistency and automation; Org Genie provides flexibility and exploratory capability. Use both together for comprehensive observability intelligence.

→ [**Learn more about Org Genie**](https://docs.pharos.ai/platform-overview/pharos-platform-architecture/ai-analysis/org-genie.md)
{% endhint %}

***

## Getting Started with AI Analysis

The AI capabilities in Pharos are designed to complement, not replace, your existing troubleshooting workflows. Start simple and gradually increase sophistication as your team becomes comfortable.

### Recommended Workflow

Adopt AI analysis progressively—start with simple on-demand use, then layer in automation as you build trust.

{% hint style="info" %}
**Progressive Adoption**

Don't try to implement everything at once. Start with basic on-demand analysis, validate its value, then layer in automation incrementally.
{% endhint %}

<table><thead><tr><th width="50">Step</th><th width="220">Action</th><th>When to Use</th></tr></thead><tbody><tr><td>1️⃣</td><td><strong>Issue Analysis</strong></td><td>For problems that have already been identified and grouped—get structured insights immediately</td></tr><tr><td>2️⃣</td><td><strong>Org Genie (Org-Level)</strong></td><td>When exploring your org's health, identifying trends, or prioritizing across multiple Issues</td></tr><tr><td>3️⃣</td><td><strong>Org Genie (Issue-Level)</strong></td><td>When you need to dig deeper into a specific Issue's root cause or correlation with changes</td></tr><tr><td>4️⃣</td><td><strong>Combine Both Approaches</strong></td><td>Use Org Genie to discover Issues, review structured analysis, then return to Org Genie for follow-up questions</td></tr><tr><td>5️⃣</td><td><strong>Automate Delivery</strong></td><td>Once you trust AI insights, configure rules to automatically deliver analysis to Slack, Jira, or email</td></tr></tbody></table>

### First Steps

The following four-week plan helps teams ramp up gradually.

**Week 1: Manual Exploration**

1. Open any high-priority Issue in Pharos
2. Navigate to the AI Analysis tab
3. Review the structured output (severity, root cause, recommendations)
4. Validate insights against your manual investigation
5. Note time savings compared to manual approach

**Week 2: Conversational Queries** 6. Access Org Genie from the main navigation 7. Ask a simple question: "Show me errors from yesterday" 8. Request a visualization: "Show me a chart of errors over the last month" 9. Try issue-level Org Genie for a deep-dive investigation 10. Experiment with follow-up questions to refine understanding

**Week 3: Basic Automation** 11. Configure a simple rule: Alert on high-occurrence Issues 12. Set up Slack notification delivery with AI analysis included 13. Monitor team's response to automated insights 14. Adjust thresholds based on feedback

**Week 4: Advanced Automation** 15. Create a Flow for post-deployment validation 16. Implement automated daily triage with routing logic 17. Track metrics (time to resolution, team engagement) 18. Refine automation based on effectiveness data

{% hint style="success" %}
**AI as a Tool, Not a Replacement**

AI analysis accelerates your troubleshooting and surfaces insights you might miss. Final decisions about prioritization and resolution remain with your team, informed by AI-generated intelligence but guided by your business knowledge and technical expertise.
{% endhint %}

***

## Best Practices for Effective AI Analysis

### For Consuming Analysis

When reading AI output, treat it as a well-informed starting point rather than a final answer.

**Validate Recommendations:**\
AI provides likely root causes, not guaranteed answers. Always validate suggestions with manual review before making changes, especially in production environments.

**Use as a Starting Point:**\
Let AI analysis guide your investigation, not define it. Your domain knowledge combined with AI insights produces the best outcomes.

**Document Findings:**\
Copy AI analysis into Jira tickets, incident reports, and documentation to build institutional knowledge and justify decisions to stakeholders.

**Provide Feedback:**\
If AI analysis misses the mark, note what it got wrong. While you can't directly train the model, patterns in AI accuracy can inform configuration improvements.

### For Automation Design

When automating analysis, expand scope gradually and keep governor limits in mind.

**Start Conservative:**\
Begin with high thresholds and narrow criteria for automated analysis. Expand as you validate effectiveness.

**Target High-Value Scenarios:**\
Automate analysis for Issues that historically consume significant investigation time—high-occurrence errors, production failures, user-impacting problems.

**Respect Governor Limits:**\
Design Flows and rules with Salesforce limits in mind. Batch operations where possible; use selective criteria to avoid high-volume invocations.

**Monitor and Refine:**\
Track which automated analyses lead to action. Low action rates indicate over-automation or poor targeting—adjust accordingly.

### For Data Quality

AI analysis is only as good as the data behind it, so invest in log quality and change tracking.

**Use Rich Logging:**\
AI analysis quality depends on log quality. Use [Triton instrumentation](https://docs.pharos.ai/platform-overview/pharos-platform-architecture/logs/log-ingestion-methods/pharos-triton-instrumentation.md) to log meaningful context in custom code.

**Enable Change Tracking:**\
Configure [Change Tracking](https://docs.pharos.ai/configuration-guide/change-tracking.md) to enable metadata correlation—one of the most powerful AI analysis features.

**Maintain Clean Data:**\
Exclude test users, sandbox activity, and known Issues from triggering automated analysis to reduce noise.

***

## Limitations and Considerations

### AI is Probabilistic, Not Deterministic

AI analysis surfaces probabilities, not certainties, and should be validated before you act on it.

{% hint style="warning" %}
**Important Understanding**

Org Genie provides **likely** root causes and **suggested** next steps based on patterns. It's not guaranteed to be correct 100% of the time.
{% endhint %}

**What this means in practice:**

* Severity assessments are estimates based on data, not absolute truth
* Root cause indicators point to probable causes, not definitive ones
* Recommendations are starting points for investigation, not mandated actions

**Recommendation:**\
Always validate AI suggestions with manual review before taking action, especially for critical systems. Use AI to accelerate investigation, not replace human judgment.

### Data Quality Impacts Results

**If Your Logs Lack Context:**\
AI analysis is only as good as the data it has. If logs don't include meaningful error messages or contextual information (affected records, user context, business process), AI insights will be limited.

**Recommendation:**\
Use Triton custom instrumentation to log rich contextual data in business-critical processes. The investment in logging pays dividends in AI analysis quality.

### Metadata Changes May Not Capture Everything

Some changes fall outside what Pharos can track, leaving blind spots in correlation analysis.

**Blind Spots:**

* Changes made via API (not all tracked in Setup Audit Trail)
* External system changes (outside Salesforce)
* Data issues (bad data quality, not configuration problems)
* Environment-specific problems (network issues, infrastructure)

**Recommendation:**\
Org Genie is a powerful tool, but it's not omniscient. Combine AI insights with domain knowledge, system monitoring, and external context for complete understanding.

### Historical Data Requirements

Meaningful trend analysis depends on accumulated history, so results improve as data builds up over time.

**For Meaningful Analysis:**

* AI needs sufficient historical data to identify patterns and trends
* Recently-installed Pharos instances have limited historical context
* Change tracking must accumulate over time to correlate with Issues

**Recommendation:**\
AI analysis improves over time as more data accumulates. Early analyses may be less sophisticated; give the system time to build comprehensive history.

***

{% hint style="success" %}
**Ready to Experience AI Analysis?**

1. Navigate to any Issue in Pharos with >10 occurrences
2. Open the AI Analysis tab
3. Review the structured insights in under 30 seconds
4. Note how long the same investigation would have taken manually
5. Copy the analysis into a Slack message or Jira ticket
6. Observe your team's reaction to having this context immediately available

The best way to understand AI analysis value is to experience it firsthand with real Issues from your org.
{% endhint %}
