Plagued By Scheduling Woes? Boost the Accuracy of
Call Volume Predictions
By Bob Webb, Pipkins Inc.
The usefulness of the staff schedule created by workforce management
software will rise or fall on the reliability of the predictions it makes
about the expected volume of incoming work. Without the right assumptions
about the workload to be handled, the software's ability to accurately
calculate staffing needs is doomed. The effect on the bottom line can be as
chilling as a bad day on Wall Street on an investor's stock portfolio.
If the forecasted work volume is too high, too many agents will be
assigned, leading to boredom, a customer-irritating 'don't-bother-me'
attitude and wasted labor expense. If the forecast is too low, the call
center will be understaffed, resulting in longer average speed of answer, a
high number of abandoned calls and a proportionate level of lost sales.
Either way, staffing mistakes can drive customers away and cause lasting
business damage.
The problem: not all forecasting tools are created equal. Only the most
sophisticated systems can perform correlated forecasting; that is,
forecasting for specific events such as catalog drops that cause wide
fluctuations in the volume of calls (and e-mails, fax and/or online chat
sessions where applicable) that must be processed. For this reason, the
forecasting capabilities of a workforce management software package should
be a central consideration in any purchasing decision. Here are some factors
to consider, the steps involved in maximizing the accuracy of the forecast
and the impact the right forecasting tool can have on call center revenues.
The Importance Of Pattern Recognition
There are two basic methodologies used to forecast workload in a call
center: Exponential Weighted Moving Average and Historical Trend Analysis.
Both use historical data collected from the call center's ACD and both
take growth trends into account in their calculations.
The Exponential Weighted Moving Average calculates the average call
volume over a specific time period and then bases its projections on a
formula that assigns more weight to recent activity. This technique is
effective for contact centers where there is little fluctuation in call
volume and patterns, such as help desks and technical support organizations,
but it has shortcomings when trends change. It is unable to predict a
continuation of trends during periods of generally increasing or decreasing
volume, or to associate changes in volume and/or call arrival patterns with
specific events (pattern recognition).
Historical Trend Analysis not only accurately predicts the continuation
of trends, but the more advanced algorithms also incorporate pattern
recognition to fine-tune forecasts for special events like promotional
mailings or national holidays. Each time a particular event recurs, the
forecasted call volume is automatically adjusted to reflect the increase or
decline in incoming work caused by comparable occurrences in the past, such
as a historical 40 percent drop in volume on the Fourth of July.
In environments where workloads regularly ebb and flow due to marketing
activities and other definable variables, Historical Trend Analysis is the
only way to ensure proper staffing because it is the only methodology that
can incorporate complex historical trends in its calculations. Without
pattern matching to predict different customer behavior for different
events, the risk of over- or understaffing increases dramatically.
Mapping Historical Data To Special Events
A key step in using a workforce management program that employs pattern
recognition is regular data validation. Analysts must review the data
collected by the ACD, preferably on a daily basis and not less frequently
than weekly, to determine if there is an identifiable cause for all spikes
and drops in call volume.
Most unusual patterns will be related to recognizable events such as
direct mail campaigns, catalog drops, TV advertorials, discount offers,
competitors' promotions, pay periods, billing cycles or holidays. Some may
even be traceable to external factors such as the Super Bowl, the Olympics
or a snowstorm.
If a given fluctuation was triggered by a recurring special event,
analysts instruct the system to interpret that data set accordingly when
producing a forecast. Conversely, if a given deviation was the result of a
one-time anomaly like a product mention on the Oprah show, analysts can tell
the system to ignore that data set when forecasting. These instructions are
vital in producing the most accurate forecast possible.
Assigning Attributes To Specific Events
To further enhance accuracy, some forecasting tools also make it
possible to describe each event in detail through the use of attributes. One
catalog drop might consist of 10,000 pieces sent to women between the ages
of 20 and 35 in Southern California, for example, while another might
involve 5,000 pieces directed at older women in the Midwest. By logging
these characteristics into the system, analysts ensure that the differing
call patterns produced by each drop will be 'remembered' and used in
forecasting call volumes the next time similar mailings go out.
The most advanced systems can search for historic trends that parallel
upcoming events both by specific match (e.g., the specific guest host on a
TV shopping channel) and by a range of values (e.g., products between $50
and $100). This aids in correlating past and future events. There will be a
substantial difference in response to a piece of jewelry that sells for $200
and one that sells for $2,000, for example, and only a tool that allows this
information to be recorded can factor in that difference when creating a
forecast.
The Impact On Call Center Revenue
With historical patterns identified, attributes assigned and upcoming
special events entered into the system, call volumes can be forecast with a
far greater degree of accuracy than with tools lacking these capabilities.
Staffing requirements, in turn, can be predicted far more precisely. The
significance can be seen by considering the consequences of a poor forecast.
Let's say that a workforce management package has underestimated call
volume and therefore staffing needs so substantially that 100 callers out of
1,000 hang up before they speak to an agent. In a sales environment in which
the average order is just $50, that means $5,000 in lost revenues per day,
$150,000 per month, or a staggering $1.8 million per year. At best, these
lost sales cut into a call center's profits; at worst, they can ruin a
business.
There are, of course, many other components in the equation that dictate
the effectiveness of a given workforce management software package. These
include the software's inherent sensitivity to agent skill sets and work
rules, its real-time adherence capabilities, and its ability to calculate
staffing requirements based on highly specific user-defined service levels
ranging from mean time to answer to the percentage of busies and abandoned
calls that will be tolerated.
But all that is moot if the software's forecasting tool doesn't meet
the call center's needs. Since all agent assignments are based on
anticipated call volumes, a package with inadequate forecasting capabilities
is like a weatherman with old technology. Both will issue a disproportionate
number of wrong predictions. For call centers that rely on proper staffing
to do their work, choosing the right forecasting solution can make all the
difference ' and ward off an almanac's worth of rainy days.
Bob Webb is vice president of sales of Pipkins, Inc. (www.pipkins.com), a worldwide supplier of
workforce management software and services to the call center industry.
[ Return
To The June 2002 Table Of Contents ]
|