Revenue Prediction Setup

This allows you to fire your own logic e.g. pixel events based on high accurate revenue predictions and top n session percentages.

Requirements

1. one week of data
2. enabled server-side (please request)

Setup

To enable the predictions, set assertive.predict.enabled option to true.

window.assertive.predict = {
    enabled: true,
};

To access the revenue prediction, attach to the assertive_predictedRevenue event and check the events data object for impression, pageView and session revenue predictions. Note: predictions are in CPM

addEventListener(
    "assertive_predictedRevenue",
    function (e) {
        console.log("assertive_predictedRevenue", e.data);
    },
    false
);

The e.data object returned on every tracked revenue event:

{
    predictedRevenueCPM: {
        impression: 0.29518782454658005,
        pageView: 0.75443967454658,
        session: 3.82856877149944,
    },
    topNPercent: {
        session: 77,
    }
}

Prediction Performance Evaluation

Prediction performance should be judged according to the R-Squared (R2) metric. Other metrics of error (KMAE, KRMSE) are only useful to compare results across the same entity.

• R2 is a value between -∞ and 1.
• If R2 is smaller than 0, results are worse than using the average as the prediction.
• If R2 is bigger than 0, results are better than using the average as the prediction.
• If R2 is around 0.25, results are weak.
• If R2 is around 0.5, results are moderate.
• If R2 is around 0.75, results are substantial.

This is not a rule of thumb. You might get useful results even with "weak" R2s.