Improvements to Insights Algorithms

Mrunal Deshpande Updated by Mrunal Deshpande

Impact score improvements are made to Insights algorithms.

The prior algorithm highlighted potential top contributors only based on the absolute difference. In the case of ratio or average, absolute change can be on the higher side, but the impact of that on the overall change between the two time periods or cohorts can be minimal due to low population.

Insights algorithm was updated to calculate the impact score in order to suppress the impact score for potential top contributors that represent a small amount or a large amount of the data and are therefore not interesting.

This modification includes an exponential decay factor below a certain low population threshold and above a certain high population threshold. The exponential decay has a similar effect to simply filtering out candidate top contributors outside of these limits but allows for smoother suppression around the limits.

For example, for the change in average interest rate between Q1 and Q2, the top contributor based on the absolute change might be "state = XYZ," but the number of loans associated with that state might be 0.01% of the overall data. Such low population contributors will be filtered out with the updated algorithm.

How Did We Do?

Live Insights