The workforce management process begins with the forecast, right?  Well, not quite.  It really begins with the gathering of data that is used to create the forecast.  Since the foundation of any forecast is the set of historical patterns, volumes, and handle times that have been experienced, searching through this history for relevant data that is both appropriate for the time frame being forecast and validated for accuracy is critical.

Let’s say that we are doing a forecast for the upcoming 6-month period – perhaps to support a shift bid or hiring plan.  We need to decide what period we want to use for the basis of the plan.  We could use an average week or we could choose a peak or low week instead.  If we choose the average week, we know that the schedules we put in place will be just right some of the time, too many people during low times, and not enough during the peaks.  We can fix these mismatches with overtime, time off without pay, restricted time off during peaks, or some extra shrinkage such as training or team meetings in lows, etc.  If we choose the peak week, we will more often be overstaffed and not too many companies have the budget luxury for that.

Now that we have a plan to use the average week, where can we find the data to support the forecast?  For most centers, the ACD feed directly into the WFM software package is the primary source.  This can be a valid and accurate source or not depending on how well the ACD interface and data have been managed.  Here are some issues that need to be addressed:

• Has  there been any reconfiguration of call routing since the interface and data mapping from the ACD into the WFM tool were last set up?  If so, the data mapping needs to be adjusted to match the new routing or the data may be feeding into the wrong storage locations for forecasting.
• Has there been any reconfiguration of personnel assignments to the skills and call types since the ACD interface was set up?  Once again, the agent skills and routing instructions assumed in the WFM process must match the ACD.
• In a skill-based routing configuration, maintaining the match between skills in the ACD and the WFM system becomes even more complex and critical.  Knowing how a call should have been routed, handled, and recorded is the best way to ensure an accurate forecast.  For example, what if a caller chose “sales” in the IVR menu but the sales team was overwhelmed and a service representative took the call?  It is important to record the calls in terms of sales workload rather than service workload to ensure that there are enough sales people in the future.  Make sure the feed from the ACD brings the best data for forecasting into the WFM tool rather than reporting how a call was handled in a backup situation.
• Where a third-party router is used in a multi-site environment, the call may reach the ACD directed to a backup team because there were no resources for the primary caller choice at any site.  In this case, it is important that the data for forecasting know how the caller wanted to be routed, not how it entered the ACD and this data may need to come from the router.
• Are agents using the after-call work (ACW) or wrap-up buttons correctly?  It is a rare center that gets everyone focused on using this feature only for the time that is directly related to the call just disconnected.  If the agents stay in the ACW state longer than needed, or use it for time that is unrelated to calls just disconnected, or don’t use it when it is appropriate, time that should have been accounted for in shrinkage winds up in average handle time (AHT).  This can throw off the total handling time by a considerable measure.
• Look for any long-term changes in the AHT.  For example, a new announcement may have been added to ensure compliance with some regulation or verification of email address may have been added to the identification portion of the call.  Some new feature or automation may have reduced the time to handle some call types.  Remember that AHT is an equal partner in the determination of the total workload when it is multiplied times the contact volume, so do not ignore the influences that will affect only AHT.

In some forecasting tools, the system requires the forecaster to select one or more historical weeks as the basis of the new forecast.  The most important selection criterion in this case is the pattern of the call arrivals across the days and half-hours.  If the volume is high or low compared to current workload, the forecaster can adjust it up or down by a percentage or arbitrary amount.  Anomalies in the data can be smoothed out statistically as well, but any data which is really out of the normal ranges should be discarded.  Most systems allow non-contiguous weeks to be selected for the process.

With a solid historical base to begin the process, it is time to apply the “art” to the forecasting.  What is known about the upcoming period that would make it different in any way from the history?  Some of the following issues should be considered:

• Are there any holidays in the planning period?  The patterns tend to shift in the week before and the week after a holiday, not just on the day itself.
• Is there a definite pattern by week of month to consider?  Some centers are much busier on one week than on the others.  Billing dates can be a significant driver for some organizations and cycle planning around the billing dates can be useful.
• Are there any planned marketing campaigns?  Applying cycle planning to these events can be very effective in gauging their effects.  Building a relationship with the marketing department to get advance warning on these events is invaluable.
• What are the unique business drivers that affect your center?  Does weather cause changes in volume?  What about interest rates or changes in the financial markets?  Perhaps there is some momentary spike caused by a popular pay-per-view TV option activation.  Identify any outside influences that might affect the center and build a history of such patterns so that they can be plugged in as needed.
• Are any changes planned in the company’s offerings or service plans?  For example, a bank changed the way that it processed checks for rejection due to insufficient funds and the call volume soared.  Additions to the IVR or website give callers more self-service choices and can drop the volume to the center but may raise the AHT for the remaining calls.  Keeping feelers out within the organization to identify any customer-affecting changes can help to identify times to make forecasting adjustments.
• Consider AHT in any of the adjustments for call volume as some of them are likely to take longer or be shorter than normal calls.  Identify how turnover and new-hire classes may affect the AHT of the work.  While long-term changes in AHT need to be factored in across the board, short-term changes are also appropriate in reaction to any kind of influence that will only last a few hours or days.

Some forecasters consider shrinkage as part of the forecast rather than part of the scheduling process.  It does not really matter where it is done, but once again accuracy is critical.
The most common mistake is to use entitlements and goals for the assumptions rather than measuring the actual losses experienced.  For example, agents may be entitled to 5 paid sick days per year but are actually taking 7 days (2 unpaid) on average.  Assuming shrinkage of only 5 days will leave the center understaffed.  Some of the things to look for in shrinkage assumptions include the following:

• The pattern of shrinkage changes by type of loss.  For example, many centers experience higher losses on Mondays or Fridays, or may have no training on one day of the week or few team meetings on the weekends.  It is important to adjust the shrinkage assumption by day or week and time of day to achieve the most accurate results.
• What is the actual loss for absenteeism?  Tracking in the WFM exceptions and within the human resources/payroll systems can provide good data.
• What is the actual loss for tardies?  In some centers, no WFM exceptions are entered for these as there is no “forgiveness” in the adherence scheme for them.  Identify when tardies occur and it is likely that they cluster at the beginnings of shifts so account for this time in the most appropriate periods.
• What is the loss for training?  While the new-hire training is generally not considered in this shrinkage, ongoing training is.  It could be that your staff needs very little ongoing training or may utilize on-line eLearning options while another center may have classroom sessions lasting from one hour to several days throughout the year.  While a long-term average shrinkage assumption for hiring plans and shift bids is appropriate, as the time for the work draws closer, adjustments to the actual training expectation is needed.  Make sure to take into account the times when the training is postponed due to understaffing situations.
• How much time is planned for team meetings?  It is common for these to be planned and scheduled ahead of time, but some centers just grab opportunities as the workload permits. Where there is little or no shrinkage assumption for it, the meetings are less likely to happen.
• What time needs to be allotted for other types of work?  There may be research, call-backs, email, web chat and other types of work that the agents need to complete during the day.  Assuming this can just be slotted in between incoming calls risks that it won’t be done on busy days.  If there are deadlines for this work or specific times to be set aside for it, be sure to either include it as scheduled work or consider it is as shrinkage from the incoming call availability.
• How much time is lost to non-adherence to the schedule?  Be sure to use actual losses here rather than the goals.  Even in a center that achieves 96% adherence in a 7.5-hour day, 18 minutes on average is lost per person per day.  Failure to build in added staff to cover for these losses would result in a lower than planned service result and the bodies-in-chairs calculations assumes everyone in their seats for the entire period.

Matching up accurate historical data for the “normal period” base with the known adjustments to be made for the specific upcoming period leads to a more accurate forecast.  While the sophistication of the latest WFM tools for forecasting is amazing, they cannot do it without the best possible data and a knowledgeable WFM professional at the helm.  As we all know, a solid forecast is essential to an accurate schedule and minimizing the intraday adjustments, so taking the time to get it right has a substantial return on investment.

Maggie Klenke was a Founding Partner of The Call Center School.  She may be reached at maggie.klenke@mindspring.com.