AI Co-Pilot

The AI Co-Pilot is your conversational assistant for process exploration. Ask questions in natural language and get instant answers with automatic filters, visualizations, and insights. No SQL, no complex queries—just natural conversation.

What is the AI Co-Pilot?

The AI Co-Pilot is an intelligent assistant that understands your process data and answers questions through natural conversation. It's like having an expert process analyst available 24/7.

What You Can Do

Ask Questions

• "Show me loans that took longer than 14 days"
• "What's the average time from application to approval?"
• "Compare high-value vs standard loan processing"

Get Automatic Analysis

• AI generates filters to match your question
• Process map updates in real-time
• Charts and visualizations are created automatically
• Statistics are calculated and explained

Explore Iteratively

• Follow-up questions build on previous context
• Drill down into interesting patterns
• Compare different scenarios
• Test hypotheses conversationally

How It Works

Behind the scenes, the AI Co-Pilot uses sophisticated tools to answer your questions:

The AI's Toolkit

1. Filter Generator Creates process filters based on your question:

• Duration-based: "longer than 2 weeks"
• Attribute-based: "high-value loans"
• Event-based: "cases with manual review"
• Path-based: "cases that went through rejection"

2. Chart Generator Creates visualizations:

• Bar charts for distributions
• Line charts for trends over time
• Histograms for duration analysis
• Comparison charts for cohorts

3. Process Summary Retrieves key statistics:

• Case counts and percentages
• Average durations
• Variant information
• Performance metrics

4. Attribute Influence Analyzes which factors drive decisions:

• What attributes affect routing?
• Which values correlate with outcomes?
• How do different groups compare?

The Conversation Flow

1. You ask a question in natural language
2. AI analyzes what you're asking for
3. AI selects tools to answer (filter, chart, summary)
4. AI executes tools in the right order
5. Response streams in real-time
6. Process map updates automatically (if filter applied)
7. You follow up with more questions

Example Conversations

Let's walk through real conversations with the AI Co-Pilot:

Example 1: Finding Slow Cases

You: Show me loans that took longer than 14 days

AI: I've applied a filter to show loans with a total duration over 14 days. Found 45 loans (18% of all cases).

[Process map updates to show only these 45 cases]

Here's the breakdown:

• Average duration: 18.3 days
• Median duration: 16.2 days
• Slowest case: 28.7 days

You: What's the main bottleneck for these slow cases?

AI: For the 45 long-duration loans, the biggest bottleneck is the Manager Approval step:

[Shows chart comparing step durations]

• Risk Assessment → Manager Approval: 6.8 days average
• Other steps average: 1.2 days

This single transition accounts for 37% of the total process time for these cases.

You: Why does manager approval take so long?

AI: Based on the data, I see these patterns for slow manager approvals:

1. Workload concentration: 70% occur during month-end periods
2. Loan complexity: Average of $127K (vs $52K overall)
3. Multiple reviewers: Cases touch 2.3 approvers on average

[Shows chart of approval times by month day]

Recommendations:

• Implement approval thresholds (auto-approve under certain criteria)
• Add capacity during month-end periods
• Create fast-track for straightforward high-value cases

Example 2: Comparing Cohorts

You: Compare high-value vs standard loans

AI: I'll compare loans over $75K (high-value) vs loans under $75K (standard).

[Applies two filters, shows both on process map in different colors]

Key Differences:

Volume:

• High-value: 62 loans (25%)
• Standard: 188 loans (75%)

Duration:

• High-value: 7.2 days average
• Standard: 3.8 days average
• Difference: High-value take 89% longer

Path Differences:

• High-value: 100% require Manager Approval
• Standard: 45% auto-approved, 55% Manager Approval

[Shows comparison chart]

You: What causes the extra time for high-value loans?

AI: Breaking down the duration difference:

[Shows step-by-step comparison]

Additional time in high-value loans:

• Manager Approval: +2.1 days (requires additional sign-offs)
• Risk Assessment: +1.4 days (more thorough evaluation)
• Document Verification: +0.8 days (additional documents required)

Total extra time: 4.3 days (accounts for 89% of the difference)

Example 3: Investigating Rejections

You: Show me all rejected applications

AI: Applied filter for applications ending in Rejection. Found 28 cases (11%).

[Process map updates to show rejection path]

Rejection Statistics:

• Average time to rejection: 3.2 days
• Most common rejection point: After Risk Assessment (82%)
• Early rejections (after Credit Check): 18%

You: Why were they rejected?

AI: Analyzing rejection attributes:

Top Rejection Reasons:

1. Credit Score (43% of rejections)

   • All had scores below 650
   • Average: 612

2. Debt-to-Income Ratio (32%)

   • All exceeded 45% DTI
   • Average: 52%

3. Insufficient Income (25%)

   • Income below $40K threshold
   • Average: $32K

[Shows distribution chart]

Asking Effective Questions

Get better answers by asking clear, specific questions:

Good Question Patterns

Duration Questions

• ✅ "Show me cases that took longer than 2 weeks"
• ✅ "What's the average approval time?"
• ✅ "How long does Risk Assessment typically take?"

Comparison Questions

• ✅ "Compare approved vs rejected applications"
• ✅ "How do high-value loans differ from standard loans?"
• ✅ "Compare this month vs last month"

Investigation Questions

• ✅ "What causes delays in manager approval?"
• ✅ "Why do some cases skip credit checks?"
• ✅ "Which attributes predict approval?"

Filtering Questions

• ✅ "Show me only expedited cases"
• ✅ "Filter to loans over $100K"
• ✅ "Show cases from Q1 2024"

Less Effective Questions

Too Vague

• ❌ "Tell me about my process"
  • ✅ Better: "What's the average duration from start to finish?"

Multiple Questions at Once

• ❌ "Show me slow cases and compare high-value vs standard and tell me the bottlenecks"
  • ✅ Better: Ask one question, then follow up

Asking for Data Not Available

• ❌ "What will happen next month?" (AI can't predict future)
  • ✅ Better: "What was the trend over the past 6 months?"

Too Complex

• ❌ "Show me cases where if the loan amount is over $50K and the risk level is medium unless it's from branch A then show high risk..."
  • ✅ Better: Break into multiple simpler questions

Understanding Responses

The AI provides different types of responses:

Filter Applications

When AI applies a filter:

I've filtered the process to show [description].
Found X cases (Y% of total).

[Statistics about filtered cases]

• Process map updates automatically
• Filter appears in active filters list
• You can remove it anytime

Chart Generation

When AI creates a visualization:

Here's a chart showing [what it displays]:

[Interactive chart appears]

Key observations:
- [Insight 1]
- [Insight 2]

• Charts are interactive (hover for details)
• Can be exported or saved
• Different chart types for different questions

Statistical Summary

When AI provides statistics:

Process Statistics:
- Total cases: X
- Average duration: Y days
- Median duration: Z days
- [Other relevant metrics]

• Based on current filtered view
• Updates as you apply more filters
• Can drill down for more detail

Analysis and Insights

When AI analyzes patterns:

Based on the data, I observe:

1. [Finding 1 with supporting data]
2. [Finding 2 with supporting data]

This suggests [interpretation].

• AI connects multiple data points
• Provides business context
• Often includes recommendations

Starter Questions

When AI suggests follow-ups:

You might want to explore:
- "Show me [related scenario]"
- "Compare [dimension A] vs [dimension B]"
- "What causes [observed pattern]?"

• Click to ask immediately
• Based on current context
• Help guide exploration

Advanced Features

Voice Input

Speak your questions instead of typing:

1. Click microphone icon
2. Speak your question
3. AI transcribes and processes
4. Same responses as typed questions

Good for:

• Complex questions (easier to speak)
• Rapid-fire exploration
• Hands-free analysis
• Presentations and demos

Conversation History

Access past conversations:

• Sidebar shows all conversations
• Click any to resume
• Each maintains its own context
• Filters and state preserved

Use cases:

• Return to previous analysis
• Share conversations with teammates
• Build on prior explorations
• Document investigation paths

Multi-Turn Context

AI remembers conversation history:

You: Show me high-value loans
AI: [Filters to high-value loans, shows 62 cases]

You: What's the rejection rate?
AI: Of the 62 high-value loans, 8 were rejected (13%)

You: Show me those rejected ones
AI: [Filters further to 8 rejected high-value loans]

The AI understands "those" refers to high-value rejected loans from context.

Tool Coordination

AI coordinates multiple tools:

You: Compare approval times for different risk levels

AI internal process:
1. Apply filter for each risk level
2. Wait for filters to complete
3. Generate comparison chart
4. Calculate statistics
5. Present unified response

You just see the final answer—AI handles the coordination.

Tips and Best Practices

Start with Exploration

Begin conversations with discovery:

• "What's the overall process duration?"
• "How many variants exist?"
• "What are the most common paths?"

Build on Previous Answers

Use follow-up questions:

• "Show me only the slow ones"
• "What causes that?"
• "Compare that to the fast ones"

Be Specific About Time Periods

Include time context:

• "Show me cases from last month"
• "Compare Q1 vs Q2"
• "What's the trend over the past 90 days?"

Use Comparisons

The AI excels at comparisons:

• "Compare X vs Y"
• "How do A and B differ?"
• "Show me the difference between..."

Ask "Why" Questions

Dig into root causes:

• "Why does X take so long?"
• "Why do some cases skip step Y?"
• "Why are rejection rates higher in Z?"

Iterate and Refine

Don't expect perfect answers immediately:

1. Ask initial question
2. Review AI response
3. Ask clarifying follow-up
4. Narrow down or expand scope
5. Reach insight

Troubleshooting

AI Doesn't Understand Question

Issue: AI says "I'm not sure what you're asking"

Solutions:

• Rephrase more specifically
• Break complex question into parts
• Use concrete examples
• Check for typos

Filter Doesn't Show Expected Cases

Issue: Applied filter but results seem wrong

Solutions:

• Check filter description AI provides
• Ask AI to explain what filter was applied
• Be more specific about criteria
• Try alternative phrasing

Response is Too Slow

Issue: AI takes a long time to respond

Solutions:

• Complex questions may take 10-20 seconds
• Multiple tool coordination adds time
• Large datasets increase processing
• Check internet connection

Can't Find Specific Data

Issue: AI says data isn't available

Solutions:

• Verify that attribute exists in your data
• Check if you're asking for computed values AI can't calculate
• Try asking for related information instead

Next Steps

The AI Co-Pilot is your gateway to deep process understanding:

Explore Your Process

• Start a conversation and ask about your process
• Use starter questions if you're not sure what to ask
• Follow interesting patterns wherever they lead

Combine with Other Views

• Process Map - See filters applied visually
• Process Variants - Ask about specific variants
• AI Insights - Start from generated insights

Learn More

• Filters & Exploration - Understand how filtering works
• Quick Start Guide - See co-pilot in action