Skip to main content

What Is Predictive Routing?

Learn how Moja’s Predictive Routing feature uses recent performance to distribute calls.

E
Written by Eric Gardona

What Is Predictive Routing?

Predictive Routing is an intelligent call-distribution feature that uses recent Earnings Per Call, or EPC, performance to favor the strongest eligible route while reserving some traffic for other eligible routes.

This balance between exploitation and exploration allows the system to route most calls to the current best performer without permanently starving new or lower-volume routes of the traffic needed to prove themselves.

Important: Predictive Routing evaluates only routes that are already eligible for the current call. It does not override routing conditions, schedules, caps, or other eligibility rules.

How Predictive Routing Works

1. Build the eligible route set

The routing plan first determines which routes are eligible for the current call based on the plan’s normal routing rules.

Predictive Routing then evaluates that eligible route set. Ineligible routes are not considered.

2. Review recent performance

The system reviews route performance over the configured Desired Hours window.

For each eligible route, EPC is calculated as:

EPC = Total Buyer Payout ÷ Total Calls

EPC represents the average buyer revenue generated per call during the evaluation window.

3. Identify the best route

A route must reach the configured Minimum Call Count before it can qualify as the best route.

Once at least one eligible route reaches that minimum, the qualified eligible route with the highest EPC is categorized as Best.

All other eligible routes are treated as non-best eligible routes for traffic-allocation purposes. This includes:

  • qualified routes with lower EPC;

  • routes below the minimum call count;

  • new routes with no traffic history.

In this context, “underperforming” does not mean that a route is disabled or has failed a specific EPC threshold. It means only that the route is not currently the qualified best performer.

4. Allocate traffic

The Underperforming Percentage controls how much traffic is reserved for routes other than the current best route.

Best-route target percentage = 100% − Underperforming Percentage

For example, if Underperforming Percentage is set to 10:

  • approximately 90% of calls are directed to the current best route;

  • approximately 10% of calls are distributed across the remaining eligible routes.

This exploration traffic allows other routes to accumulate performance data and potentially become the new best route.

When selecting among non-best eligible routes, the system favors the route with the fewest recent calls. If multiple routes have the same call count, the route with the higher EPC is preferred.

Important: Underperforming Percentage is a traffic-allocation setting. It is not an EPC-difference threshold and does not disable routes whose EPC is a specified percentage below the best route.

Startup and Minimum Call Behavior

Predictive Routing requires a reliable performance baseline.

If no eligible route has reached the configured Minimum Call Count, Predictive Routing does not make a selection. Calls continue through the routing plan’s normal priority and weight behavior.

Once at least one eligible route reaches the minimum:

  • the qualified route with the highest EPC becomes the current best route;

  • the best route receives the targeted best-route share;

  • all other eligible routes share the exploration allocation.

Routes below the minimum cannot become the best route yet, but they can continue receiving exploration traffic so they can accumulate enough calls to qualify.

Example

Consider this routing plan:

National Sales Campaign

  • Target 1: Static buyer

  • Target 2: Individual RTB buyer

  • Target 3: RTB group

  • Target 4: IVR overflow

Predictive Routing configuration

  • Desired Hours: 1

  • Minimum Call Count: 5

  • Underperforming Percentage: 10

Performance during the evaluation window:

Route

Calls

Total buyer payout

EPC

Classification

Target 1

10

$360

$36.00

Non-best eligible route

Target 2

8

$240

$30.00

Non-best eligible route

Target 3

12

$462

$38.50

Best

Target 4

3

$0

$0.00

Non-best eligible route; below minimum

Target 3 is the qualified route with the highest EPC, so it becomes the current Best route.

With an Underperforming Percentage of 10:

  • the system targets approximately 90% of calls to Target 3;

  • the remaining approximately 10% is distributed across Targets 1, 2, and 4;

  • no target is disabled because its EPC is a certain percentage below Target 3;

  • Target 4 remains eligible for exploration traffic but cannot become the best route until it reaches five calls.

Actual traffic percentages may vary over small call volumes, but should trend toward the configured allocation over time.

As performance changes, another qualified route may become the new best route.

Predictive Routing Configuration

Name

Labels the Predictive Routing configuration.

The configuration can be selected for use by a routing plan.

Minimum Call Count

The minimum number of calls a route must receive within the evaluation window before it can qualify as the best route.

Key behavior

  • If no eligible route reaches the minimum, Predictive Routing falls back to normal routing behavior.

  • A route below the minimum cannot become the best route.

  • Once a qualified best route exists, below-minimum and new routes can receive exploration traffic.

Use a minimum large enough to avoid declaring a winner from an unreliable sample, while keeping it low enough for the system to begin optimizing within a reasonable period.

Desired Hours

The rolling lookback window used to calculate route performance and measure recent routing distribution.

For example:

  • 1 evaluates the most recent hour;

  • 12 evaluates the most recent 12 hours;

  • 24 evaluates the most recent 24 hours.

A shorter window reacts more quickly to performance changes but may be more volatile. A longer window provides a larger sample but responds more slowly.

Underperforming Percentage

The percentage of eligible call traffic reserved for routes other than the current best route.

Best-route target percentage = 100 − Underperforming Percentage

Underperforming Percentage

Best-route target

Other eligible routes

5

95%

5%

10

90%

10%

20

80%

20%

30

70%

30%

A lower value concentrates more traffic on the current best route. A higher value increases exploration and gives alternative routes more opportunities to establish or improve their performance.

The exploration percentage applies collectively to all non-best eligible routes; it is not assigned separately to each one.

How Predictive Routing Interacts With Normal Routing

Predictive Routing does not make an ineligible route eligible.

The system first applies the routing plan’s normal eligibility controls. Predictive Routing then chooses which eligible route should be prioritized for that call.

If Predictive Routing cannot make a selection — for example, because no eligible route has reached the minimum call count or performance data is unavailable — the routing plan continues using its normal priority and weight behavior.

Key Points

  • Predictive Routing ranks eligible routes using recent EPC.

  • EPC is calculated as total buyer payout divided by total calls.

  • A route must meet the minimum call count before it can become the best route.

  • Only one qualified eligible route is treated as the current best route.

  • All other eligible routes share the exploration allocation.

  • Underperforming Percentage is a traffic-allocation percentage, not an EPC variance threshold.

  • Predictive Routing does not disable routes merely because their EPC is below the best route.

  • If no route has enough data, the system falls back to normal routing.

  • Route classifications can change as new performance data enters the rolling evaluation window.

Did this answer your question?