Average Speed of Answer, Average Handling Time, and After Call Work for Field Support Center

Chris Rucker, Associate Data Scientist

14 Nov 2016

Observation

Three discrete quantitative variables were chosen for exploration from a data set of 42 observations from the Field Support Center's (FSC) CS Monthly Flash reports.

  • Average Handling Time (AHT) is the amount of time it takes an Agent to deal with all aspects of a call including talk time plus After Call Work increases.
  • After Call Work (ACW) is the period of time immediately after contact with the customer is completed and any supplementary work is undertaken by the Agent.
  • Average Speed of Answer (ASA) is the amount of time it takes to answer a typical call once it has been routed to the FSC.

In [67]:
import pandas as pd
import seaborn as sns

In [68]:
data = pd.read_csv('C:\Users\crucker\calls.csv')
data.head()


Out[68]:
date calls offered to queue sales retention AHT ACW ASA
0 1 548792 71442 61461 346 9 33
1 2 579488 67364 65161 349 9 24
2 3 607728 69317 71092 348 9 22
3 4 662836 77384 76004 355 9 36
4 5 610952 70745 66888 356 10 33

ASA ↑ AHT ↑

As the amount of time it takes to answer a typical call once it has been routed to the FSC increases, the amount of time it takes an Agent to deal with all aspects of a call including talk time plus ACW increases.

A bivariate distribution of ASA/AHT variables along with the univariate (or marginal) distribution of each on separate axes shows a Pearson correlation coefficient of 0.77 and a p-value of 1.8e-09. The p-value roughly indicates the probability of an uncorrelated system producing datasets that have a Pearson correlation at least as extreme as the one computed from these datasets. The p-values are not entirely reliable but are probably reasonable for datasets larger than 500 or so. Positive correlations imply that as AHT increases, so does ASA.

An altogether different approach is to fit a nonparametric regression using a lowess smoother. This approach has the fewest assumptions, although it is computationally intensive and so currently confidence intervals are not computed at all.


In [69]:
sns.lmplot(x='AHT', y='ASA', data=data, robust=True)


Out[69]:
<seaborn.axisgrid.FacetGrid at 0x13cfb3c8>

In [70]:
sns.lmplot(x='AHT', y='ASA', data=data, lowess=True)


Out[70]:
<seaborn.axisgrid.FacetGrid at 0x13ebd2b0>

In [25]:
sns.jointplot(x='AHT', y='ASA', data=data, kind="reg", robust=True);


ACW ↓ ASA ↑

As the period of time immediately after contact with the customer is completed and any supplementary work is undertaken by the Agent decreases, the amount of time it takes to answer a typical call once it has been routed to the FSC increases.

A bivariate distribution of ACW/ASA variables along with the univariate (or marginal) distribution of each on separate axes shows a Pearson correlation coefficient of -0.48 and a p-value of 0.0013. The p-value roughly indicates the probability of an uncorrelated system producing datasets that have a Pearson correlation at least as extreme as the one computed from these datasets. The p-values are not entirely reliable but are probably reasonable for datasets larger than 500 or so. Negative correlations imply that as ASA increases, ACW decreases.

An altogether different approach is to fit a nonparametric regression using a lowess smoother. This approach has the fewest assumptions, although it is computationally intensive and so currently confidence intervals are not computed at all.


In [71]:
sns.lmplot(x='ASA', y='ACW', data=data, robust=True)


Out[71]:
<seaborn.axisgrid.FacetGrid at 0x1370db38>

In [72]:
sns.lmplot(x='ASA', y='ACW', data=data, lowess=True)


Out[72]:
<seaborn.axisgrid.FacetGrid at 0x13dd08d0>

In [28]:
sns.jointplot(x='ASA', y='ACW', data=data, kind="reg", robust=True);


ACW ↓ AHT ↑

As the period of time immediately after contact with the customer is completed and any supplementary work is undertaken by the Agent decreases, the amount of time it takes an Agent to deal with all aspects of a call increases.

A bivariate distribution of AHT/ACW variables along with the univariate (or marginal) distribution of each on separate axes shows a Pearson correlation coefficient of -0.096 and a p-value of 0.55. The p-value roughly indicates the probability of an uncorrelated system producing datasets that have a Pearson correlation at least as extreme as the one computed from these datasets. The p-values are not entirely reliable but are probably reasonable for datasets larger than 500 or so. Negative correlations imply that as AHT increases, ACW decreases.

An altogether different approach is to fit a nonparametric regression using a lowess smoother. This approach has the fewest assumptions, although it is computationally intensive and so currently confidence intervals are not computed at all.


In [73]:
sns.lmplot(x='AHT', y='ACW', data=data, robust=True)


Out[73]:
<seaborn.axisgrid.FacetGrid at 0x14520320>

In [74]:
sns.lmplot(x='AHT', y='ACW', data=data, lowess=True)


Out[74]:
<seaborn.axisgrid.FacetGrid at 0x1c123d30>

In [31]:
sns.jointplot(x='AHT', y='ACW', data=data, kind="reg", robust=True);


Conclusion

As the amount of time it takes to answer a typical call once it has been routed to the FSC increases, the amount of time it takes an Agent to deal with all aspects of a call including talk time plus ACW increases.

As the period of time immediately after contact with the customer is completed and any supplementary work is undertaken by the Agent decreases, the amount of time it takes to answer a typical call once it has been routed to the FSC decreases.

As the period of time immediately after contact with the customer is completed and any supplementary work is undertaken by the Agent decreases, the amount of time it takes an Agent to deal with all aspects of a call increases.