lagged metrics

In this project we want to investigate main differences between the metrics on the call and lagged metrics

difference

The lagged metrics considers a time span of 25 days when the customer could call again and turn around. The lagged metric loses around 10% customers

time series of on-call and lagged metrics

We have different metrics to predict

metrics to predict, on call and lagged

We clean the features

normalized feature distribution

We check the feature independency

correlation between features

Effect of afiniti agent on feature distribution

overlay of feature distribution

Overlay distribution of features on saved customers

overlay of feature distribution

Overlay distribution of featured of saved/lost customers

Overlay distribution of features: joyplot

We build a predictive model for the on-call metric

confusion matrix on prediction: on-call

We use the prediction of the previous model to predict the new metric

confusion matrix on prediction: lagged

We knock-out features to calculate the relative importance of that feature

knock-out of features from trained model: on-call

knock-out of features from trained model: lagged

We finally check the relative difference between on-call and lagged

relative difference between feature importance in lagged metrics

We can study the relative feature importance

feature importance (week overlay)

and see how it evolves week by week

relative importance week by week

We study over week how the relative importance of that metric dropw over time

importance difference relative to lagged metrics

We calculate the lift depending on categorical metrics

lift = SR_on − SR_off

lift relative to categorical metrics

cohort reshuffling

We look month per month how the association of agents per area varies over time

time series volume changes in cohorts

Shuffling among cohorts

reshuffling of agents cohorts