Forecast Call Center Metrics - Call Arrival Patterns and Handle Times

[dkstoneman]

I've been using SAS for a while but only running preestablished programs so I'm not sure what section I should be posting this request. I'm looking for a way to predict interval call arrival patterns, handle time, and staffing for multiple centers. Right now we do all these calculations in MS Excel files and the process has become very cumbersome. I currently have over 500K of records of interval call volume and handle time, stored in MS SQL server tables, over the last few years.

Looking at some of the current SAS programs being used:

• We use PROC TIMESERIES for loading information into tables
• We use PROC ARIMA for end-of-day information. (coding below)

proc arima data=&LOB.day ;
identify var = smoothnco (7) nlag=48 noprint ;
estimate p = (7)(14)(21)(28)(35)
method = ML noint converge = .0001 delta = .0001 maxiter = 5000 noprint ;
forecast id = date interval = day lead = 1 alpha = .05
out = &LOB.for1(keep=Date forecast rename=(forecast=forecast1)) noprint ;

Right now I'm exploring all options. Any recommendations are greatly appreciated.

[sbxkoenk]

For modelling (and forecasting) COUNT per time interval :

You can try PROC CNTSELECT (or PROC COUNTREG).

Another possible offering from SAS Econometrics is the Poisson Hidden Markov Model
(https://go.documentation.sas.com/doc/en/pgmsascdc/v_038/casecon/casecon_hmm_details17.htm).

Blog article: Poisson HMM: The model of count time series

​​​​​​​SAS' Ji Shen introduces you to an effective solution for modeling and forecasting count time series, Poisson HMM. It is a useful tool for modeling discrete time series and dealing with overdispersed and serially correlated data.

https://blogs.sas.com/content/subconsciousmusings/2023/05/11/poisson-hmm-the-model-of-count-time-ser...

Koen