How to Use Predictive Analytics

Use AI-powered predictions to forecast conversation volume, optimize staffing levels, and prepare for busy periods before they happen.

Prerequisites

• ChatFlow Account (Enterprise plan)
• Historical data - At least 2 weeks of conversations
• Member role - At least Viewer access

Required Plan: Enterprise

What Predictive Analytics Offers

• Volume Forecasting - Predict future conversation counts
• Staffing Recommendations - Know how many agents you need
• Peak Hour Identification - Find your busiest times
• Performance Benchmarks - Compare against industry standards
• Accuracy Tracking - See how accurate predictions are

Accessing Predictive Analytics

1. Click Predictive Analytics in the sidebar
2. View the analytics dashboard

Understanding the Dashboard

Volume Predictions Chart

Shows predicted conversation volume:

• Blue line - Predicted volume
• Shaded area - Confidence interval
• Time axis - Hours or days ahead

Peak Hour Heatmap

Visual grid showing:

• Rows - Days of the week
• Columns - Hours of the day
• Color intensity - Volume level (darker = busier)

Staffing Recommendations

For each time period:

• Current staff count
• Recommended staff count
• Expected conversation volume
• AI deflection rate (% handled by bot)
• Escalation rate (% needing humans)

Using Volume Predictions

Reading the Chart

1. X-axis - Time (hours ahead or specific dates)
2. Y-axis - Predicted number of conversations
3. Line - Most likely volume
4. Shaded range - Possible variation

Time Ranges

Select prediction timeframe:

• Next 24 hours - Hour-by-hour forecast
• Next 7 days - Daily predictions
• Next 30 days - Weekly patterns

Acting on Predictions

Prediction Shows	Action to Take
Volume spike	Schedule more agents
Low period	Reduce staffing or assign other tasks
Sustained high	Consider hiring
Weekend pattern	Plan coverage

Using Staffing Recommendations

Understanding the Calculation

The system considers:

1. Predicted volume - Total conversations expected
2. AI deflection - % the bot handles alone (typically 85%)
3. Escalations - % needing human agents
4. Chats per agent - Capacity (typically 3 concurrent)

Example:

• 50 predicted conversations
• 85% handled by AI = 42 AI-handled
• 15% escalate = 8 need humans
• 8 ÷ 3 chats per agent = 3 agents needed

Staffing Table

Column	Description
Time Slot	Time period
Expected Volume	Total predicted chats
AI Handled	Bot resolves
Escalations	Human needed
Recommended Staff	Agents to schedule
Status	Understaffed/OK/Overstaffed

Taking Action

1. Review recommendations for upcoming shifts
2. Compare with your current schedule
3. Adjust staffing to match predictions
4. Monitor queue times during the shift

Setting Business Hours

Configure Operating Hours

1. Click Settings (gear icon) on the dashboard
2. Set your business hours for each day
3. Save changes

Why Business Hours Matter

• Staffing recommendations only for operating hours
• After-hours predictions show expected missed chats
• Helps plan coverage for extended hours

Performance Benchmarks

Available Benchmarks

Metric	Excellent	Good	Needs Work
First Response Time	< 30s	< 2 min	> 2 min
Avg Response Time	< 1 min	< 3 min	> 3 min
Resolution Rate	> 80%	> 70%	< 70%

Your Performance

The dashboard shows:

• Your current metrics
• Benchmark comparison
• Performance badge (Excellent/Good/Needs Work)

Accuracy Metrics

How Accuracy Is Measured

After predictions, we compare:

• Predicted volume vs Actual volume
• Displayed as accuracy percentage
• Updated daily

Improving Accuracy

Predictions improve with:

• More historical data
• Consistent patterns
• Fewer anomalies

Best Practices

Regular Review

• Check predictions daily
• Review weekly patterns
• Plan ahead for known events

Staffing Optimization

• Trust the AI deflection rate
• Staff for escalations, not total volume
• Build in buffer for peak times

Acting on Insights

• Adjust schedules proactively
• Train AI to handle more queries
• Identify why escalations happen

Feedback Loop

1. Review prediction accuracy
2. Note any discrepancies
3. Consider external factors (campaigns, events)
4. Adjust expectations accordingly

Troubleshooting

Predictions Not Available

• Need at least 2 weeks of conversation history
• Enterprise plan required
• Check organization has conversations

Inaccurate Predictions

• External events affect volume
• Marketing campaigns cause spikes
• System learns and improves over time

Staffing Seems Too Low

• AI handles most conversations (85%)
• Predictions are for human-needed chats
• Add buffer if uncertain

Data Not Updating

• Refresh the page
• Data updates every few hours
• Check your date filter

Next Steps

• Use Reports - Analyze historical data
• Manage Agent Settings - Configure capacity
• Add Team Members - Expand your team