Considerations in Call Center Staffing

Workforce management (WFM) software and call center calculators are very useful tools for resource planning and management. They estimate resource requirements, recommend staffing and scheduling, and forecast performance. However, these tools possess a somewhat simplified view of the operational structure of the call center and special considerations must be made in order to apply them correctly in real-world situations. This article describes several situations that require special attention and proposes methods to address them:

- Multiple Queues
- Variable Load
- Shrinkage

Staffing for Multiple Queues

The majority of call center WFM tools are based on the Erlang model. This model assumes that all calls arrive into a single queue and are handled by a single pool of telephone agents. In reality, of course, many call centers are comprised of multiple queues, each with different call load and performance characteristics and potentially different service level expectations.

Calculating resource requirements for multi-queue situations depends on the type of multi-queue setup.

Disjoint Queues

Disjoint queues are separate call center queues, in which each queue has its own telephone resources and agents answer only calls arriving in their queue. Disjoint queues represent the simplest form of multiple queues in that each queue is independent and has no effect on the other queues; in essence, each queue behaves as a separate call center.

In a call center with disjoint queues, calculate the resources for each of the queues separately. In most cases, the resources and forecasted performance can be added or averaged, as necessary.

Joint Queues

The individual queues in many call centers are not truly disjointed. Instead, they operate separately until queue exceeds its capacity, and calls overflow to a less busy “backup” queue. From that point on, the two queues are joined and act like one large queue.

When calculating the resources and the performance characteristics of joint queues, the call volume is the sum of the call volumes of the individual queues. However, the average handling time (AHT) of calls in different queues is usually different and therefore must be recalculated. It is incorrect simply to average the AHT of the two (or more) queues. For example, if the AHT of Queue I is 100 seconds and the AHT of Queue II is 200 seconds, the AHT of the combined queues is not 150 seconds. Not only is it mathematically incorrect to average numbers that are themselves averages, but the number of calls of each type in the joint queue is also different. Therefore, the AHT should be a weighted average that is proportionally adjusted to the call volume of each AHT.

Queue	I	II
Calls	5000	3000
AHT	100	200

The output of the call center calculator – number of agents, utilization, abandonment rate and so forth – will be the same for both queues at peak times, when overflow occurs.

Because agents in the backup queue are sometimes less trained or equipped to handle overflow calls, their AHT is often longer than the AHT of the agents in the original queue. In this case, a better approach is to determine the maximum call handling capacity of each queue (EasyErlang calculates the maximum capacity and the percentage of additional capacity required to meet expected peak load) and the estimated number of overflowed calls. Use the same weighted average method to calculate the resources and performance of the backup queue during peak time.

Staffing for Variable Load

Daily Call Volume The call load in a typical call center varies from hour to hour and from day to day. The figure on the right shows changes in daily call load in a typical call center. The common sense approach that is usually recommended is to calculate staffing separately for each high load and low load period. A call center calculator can estimate the service level for each of these periods, but what is the overall service level of the entire call center?

To calculate the combined service level for several periods or for different shifts, one cannot average the individual service levels. For example, if the service levels for three shifts were 89%, 91% and 84% of calls answered in 20 seconds, the overall service level is not 88%. Rather, each service level has to be computed considering the total number of calls offered and the number of callers that have experienced each service level.

Table

The same method is used to convert daily service levels to weekly or monthly service level.

Let us consider two periods in the day, one averaging 500 calls per hour, and the other only 280 calls per hour. Staffing for the peak period, assuming a target service level of 80% in 20 seconds, will require 30 agents. Staffing for the low volume period will require only 18 agents. Because it is impossible to meet the exact target service level, WFM tools overestimate the required staff. As the table below shows (Row 1), the service level in each of the two periods, as well as the combined service level, exceed the required performance target.

Table

If we reduce the headcount in Period I by one, the service level in that period will drop to 78% (Row 2), which is not significantly worse than the target of 80% and is likely tolerable, especially if the duration of the peak is short. The overall service level still meets the target.

Row 3 shows what happens when we decrease the headcount of the low volume period (Period II). The impact on the service performance during that period is more significant, yet the overall service level is still above the target.

Interestingly enough, moving the headcount that was “saved” from one period to the other improves the service level during that period, as expected, but the overall service level is actually lower than the one achieved using the original allocation shown in Row 1 (see Rows 3 and 4).

We conclude that staffing for very dynamic load conditions while maintaining service level agreement and minimizing inefficiencies is difficult. As the examples show, this type of planning does not require overly precise measurement of call load in very small increments. Instead of adjusting staffing for each period in the day, call center managers should consider focusing on overall performance.

Shrinkage

The term shrinkage is used to identify the portion of the time an agent is unavailable to handle calls. This includes planned shrinkage such as breaks, meetings, training, research and special projects, vacation, and unscheduled shrinkage such as sick time.

Shrinkage often accounts for more than 20% agents paid time. Ignoring shrinkage will cause a significant understaffing, especially during peak load time. How can a call center manager adjust staffing to compensate for this productivity loss? The simplest answer appears to be to increase the headcount by the amount lost to shrinkage. If the staffing calculator determines that current service level and call load require 60 agents, and the anticipated shrinkage is 25%:

60 x 1.25 = 75

The staff of 30 calculated by the staffing calculator is available only at 75% capacity because of the 25% shrinkage. Therefore, in order to achieve 100% capacity the increase headcount should be:

60 / 0.75 = 80

The difference between the first calculation of 75 agents and this one is 5 FTE, which is quite significant. Moreover, the agents we add to account for the shrinkage would also be subject to the same shrinkage, accounting, in this case, to nearly 2 additional FTEs:

5 / 0.75 = 6.7 or 7 FTEs

Therefore, in large call centers, where the shrinkage compensation is significant, a more accurate adjustment would be to account for the additional shrinkage as well. EasyErlang does this automatically.

As in all other call center staffing calculations, the theoretical results should be monitored, evaluated and adjusted to fit the specific needs, behaviors and constraints the individual call center .

As a general guideline, call centers that monitor their service levels and staff utilization continuously and make frequent adjustments can use the first or second approach described above. Call centers that prefer a more conservative and safe approach and to err on the side of overstaffing, are better off using the third approach.

Back to Articles Page

Prices, Licensing Terms, and Free Support and Free Upgrade offers are subject to change without notice.

The information in this Website, screen captures of EasyErlang, and the use of EasyErlang software do not represent advice or recommendation. Any action taken based upon this information is the sole responsibility of the user. No liability for errors or omissions is implied or accepted by the owners of this site.