The new version of CPI detects anomalies in the metrics and acts on them by lowering the importance of these metrics in calculations. As a result, we have an algorithm that prevents one metric going wild and taking the whole CPI with it. Also, articles that don't fit in the general pattern have a better chance of scoring higher values than before. 

Understanding the new Exposure CPI score

Let's see how the new calculations affects the Exposure CPI score using the example of an article with a high number of social actions, but lower than expected visits on the site.

The second version of the algorithm would allow the high number of social actions to push Exposure CPI higher than deserved.

But, with its third, perfected version, the algorithm will know that behavior on social networks is not in-tune with the behavior of readers on the site. It is smart enough to adjust the importance of the social actions for that article when calculating Exposure CPI.

Understanding the new Engagement CPI score

As for the Engagement CPI score, the algorithm is smart enough to know which articles are different from the others published on the same website, and what behavior could be considered as expected.

For example, in case of the longer articles, the algorithm knows that it cannot expect the same values of Read Depth as it can with shorter articles. This allows a fairer treatment of longer pieces when it comes to engagement.

The new version of the CPI values harmony among metrics - even when all numbers paint the same picture about reader behavior.

This is accomplished by taking into account the weight of metrics, which are upgraded and calculated on-the-fly by using non-linear functions. Functions establish what "normal" is and what should be considered as an "anomaly" during each calculation.