Skip to main content

What is Predictive Routing?

Take a deep dive into moja's predictive routing feature

Written by Moja Bot

What is Predictive Routing?
​
Predictive Routing is an intelligent call distribution system that dynamically evaluates target performance based on Earnings Per Call (EPC) - a composite metric that considers both payout AND conversion quality (duration/connection rates). Unlike simple routing that just looks at static payouts, Predictive Routing continuously learns from actual call performance to route calls to the most profitable targets.
​Critical Concept: Predictive Routing operates BEFORE priority and weight calculations in the routing hierarchy. This means it acts as a pre-filter, potentially removing underperforming targets from consideration before traditional routing rules even apply.
​
​How Predictive Routing Works - The Technical Flow
​
​1. Performance Evaluation Window

  • Evaluation Window: The system evaluates performance over a configurable time period (e.g., 1 hour, 24 hours, 7 days).

  • Minimum Call Threshold: Targets must receive a minimum number of calls (e.g., 5 calls) within the window before being evaluated

  • New Target Grace Period: Targets without sufficient call history remain active until they meet the minimum threshold

2. EPC Calculation
For each target in the routing plan, the system calculates:

EPC = Average Payout × Conversion Rate

Where:

  • Average Payout = Actual money received per connected call

  • Conversion Rate = Percentage of calls that meet quality criteria (duration, connection status)

3. Performance Comparison
The system identifies the best-performing target(s) based on highest EPC, then evaluates all other targets:

  • If a target's EPC is X% below the best performer (e.g., 10% underperforming), it gets temporarily disabled

  • Disabled targets are re-evaluated continuously and can be re-enabled if performance improves


​
​Example Routing Plan Structure:

Routing Plan: "National Sales Campaign" ├── Target 1: Static Buyer ($45 payout) ├── Target 2: RTB Individual Buyer (variable payout) ├── Target 3: RTB Group (5 buyers, competing bids) ├── Target 4: Route to IVR (overflow handling) └── Predictive Routing: ENABLED     ├── Evaluation Window: 1 hour     ├── Minimum Calls: 5     └── Underperforming %: 10%

How It Processes the Plan Above:
​
​Hour 1 (Startup Phase):

  • All targets active (none have 5+ calls yet)

  • Calls distributed based on normal priority/weight rules

  • System collecting performance data

Hour 2 (Evaluation Begins):

  • Target 1 (Static): 10 calls, $45 payout, 80% conversion = $36 EPC

  • Target 2 (RTB Individual): 8 calls, $50 avg payout, 60% conversion = $30 EPC

  • Target 3 (RTB Group): 12 calls, $55 avg payout, 70% conversion = $38.50 EPC ✓ BEST

  • Target 4 (IVR): 3 calls (below minimum, not evaluated)

System Action:

  • Target 3 is best performer at $38.50 EPC

  • Target 2 is 22% below best (disabled - exceeds 10% threshold)

  • Target 1 is 6.5% below best (remains active)

  • Target 4 continues receiving calls until minimum met


​Predictive Routing UI Configuration Guide
​Field-by-Field Breakdown:
​1. Name (Required)

  • Purpose: Labels this specific Predictive Routing configuration within the routing plan

    • Best Practice Examples:"Aggressive EPC Optimization"

    • "Conservative Daily Evaluation"

    • "Testing Configuration"

  • Important: This names the PR configuration for this routing plan only (not a reusable template)

2. Minimum Call Count (Required)

  • Purpose: Number of calls a target must receive before performance evaluation begins

  • Valid Range: Typically 5-50 calls

    • Examples by Use Case:5 calls: Aggressive/rapid optimization, high call volume

    • 10 calls: Balanced approach, moderate volume

    • 20 calls: Conservative, ensures statistical significance

  • Key Behavior: Targets with fewer than this threshold remain active (won't be disabled for underperformance)

3. Desired Hours (Required)

  • Purpose: The time window for performance evaluation

  • Format: Numeric value in hours

    • Common Settings:1: Ultra-responsive, adapts to hourly patterns

    • 6: Quarter-day evaluation, good for shift-based routing

    • 24: Daily performance average, smooths out variations

    • 168: Weekly evaluation for stable, long-term optimization

4. Underperforming Percentage (Required)

  • Purpose: Threshold for disabling underperforming targets

  • Format: Percentage value (without % sign)

    • How It Works:System identifies best performing target (highest EPC)

    • Any target performing X% below the best gets disabled

    • Example: If best target has $40 EPC and threshold is 10, any target below $36 EPC is disabled

    • Recommended Settings:5-10: Aggressive optimization, only keep top performers

    • 15-20: Balanced approach, allows moderate variance


​
​Key Advantages:

  1. Self-Optimizing: No manual intervention needed as buyer performance changes

  2. Protects Revenue: Automatically disables poor performers before they waste calls

  3. Fair Testing: Every target gets a chance to prove performance

  4. Flexible Time Windows: Adapt to hourly, daily, or weekly performance patterns

  5. Comprehensive Metrics: Considers both payout AND quality, not just one or the other

Did this answer your question?