Workforce Engagement Management

I can submit a support case as well, but I wanted to get everyone else's feedback on the below. 

Organization is attempting to use the Automatic Best Method Forecast.  They have been live approximately ~2 weeks.  For Thursday, January 8th, the system is forecasting a very high, 10+ minute Average Handle Time (small adjustments up and down, but nearly flat lined).  I recognize that more data will obviously smooth this out and make it more accurate, but the variance between actual historical and ABM seems inexplicable, even with so little data. 

Details below:

• Organization is using 2 planning groups: English & Spanish.  English includes 3 Route Paths; Spanish includes 2. 
• There is nothing in the historical data to support a 2 minute difference between ABM & WHI for English at minimum.  I can almost accept the 30-40 second difference on the Spanish planning group, but the dramatic difference between English forecasting is not acceptable nor understandable. 
  • ABM English Planning Group AHT forecasting 10:20
  • ABM Spanish Planning Group AHT forecasting 9:06
  • WHI (6 weeks) w/ no modifications source data updates: English Planning Group AHT forecasting: 8:15
  • WHI (6 weeks) w/ no modification source data updates: Spanish Planning Group AHT forecasting: 9:40
  • WHI (6 weeks) w/ test data/holidays removed English Planning Group: AHT Forecasting: 8:14
  • WHI (6 weeks) w/ test data/holidays removed Spanish Planning Group: AHT Forecasting: 9:35
• Queue performance shows an Average Handle time of no more than 8 minutes for English no matter how you cut it/filter it; 9 minutes average for Spanish. 
  • In the English Planning Group: the only exception to that rule is a single queue + skill combo that accounts for a mere 5% of offered volume in the last 6 weeks with an AHT varying between 8-9 minutes vs. the other route paths' 6.25-7.5 minute AHT. 
• When I "view history" on the AHT metric within the forecast...
  • For the English planning group the historical weeks average under 9 minutes with a few outliers in the last week of November; none of them are consistently at or near the 10 minute mark.
  • For the the Spanish Planning Group: Majority of the historical data (contrary to the performance view) has several (majority) data points over the forecasted handle time when I view the history
• My initial thought was that this could be affected by test data, but I looked back over a year and a half in Queue Performance and could not find even test data to support this.  
• My next thought was "maybe outbound is affecting the Queue Performance AHT" I filtered to Direction: Inbound and then, separately, to Initial Direction: Inbound, and neither yielded data that would support a 9+ minute handle time forecast. 
• Thanksgiving was one of the few Thursdays they were live for, but there was no data on that day, so at most it should have brought the AHT time DOWN.  Even still, when a weighted historical is pulled and test data+thanksgiving are left in, we get a more accurate AHT forecast than the ABM. 
• Organization has made a few manual modifications, but even with the disablement of these modifications, the AHT time goes higher. 

Obviously more data will help smooth this out, but it has prompted the question that I'd like some re-clarification on:

How is average handle time be calculated in the forecast vs. in the Queue Performance? I assume they are slightly different, but I can't imagine they are 2+ minutes different. 

• Is is still talk+hold+after call work time like the Queue Performance?
• What is the denominator? 
• Is there an adjustment for "higher volume=higher handle times"? or something similar that I am not considering?
• Would it make a dramatic difference to separate each route path into its own planning group vs. grouping them together? Does it not weight the number proportionally based on volume within the planning group?  

#Forecasting

------------------------------
Shelby Cronk
------------------------------

Average handle time for WFM uses the same as other views: tHandle.sum / Completed (using Analytics API terms).
Comparing historical AHT versus a forecast/predicted AHT is a bit asymmetric. They are very different.

ABM is not really a 'calculation' as it employs many techniques. Comparing an ABM forecast to WHI forecast is also asymmetric. WHI is a very simple weighted-weighted (source day weight; completed count) highly influenced by the source days chosen, the relative weights given, and any modifications applied). ABM uses all available data and looks at seasonality, trends, level shifts, outliers, etc. They are very different.

I would not recommend blowing out route paths into separate planning groups as you will lose economies of scale, hit the 'small number problem' inherent forecasting/predictions, and end up with significantly higher staffing requirements.

ABM gets better with more data. Two weeks of data is a paltry amount of data. I'd recommend using WHI until you have 6+ weeks of data for ABM. Ideally for annual seasonality you have 2+ years, but we still look at daily, weekly, monthly 'patterns' with less data. Two weeks isn't really enough to make meaningful daily pattern predictions - not enough data and too stochastic.