March 1999
Workforce Management With Skills-Based Call Routing:
The New Challenge
BY PAUL H. LEAMON, IEX CORPORATION
In today's competitive environment, many call centers are being asked to provide better
service with fewer people. Skills-based call routing promises to meet these needs. As a
result of this promise, many call centers are currently using skills-based call routing or
will be in the near future.
Skills-based call routing substantially complicates the management tasks of creating
forecasts and schedules. The strategies currently being used for these tasks will be
described and their strengths and weaknesses analyzed. Recommendations will then be made
for accommodating the complexities of skills-based call routing.
Forecasting Using Erlang C
In a call center that is not using skills-based call routing, Erlang C can be used to
calculate the number of agents needed from call volumes, average handling time (AHT) and
service level goals.
Erlang C assumes each agent handles a single call type on a first-come, first-served
basis. This assumption is not valid for skills-based call routing because:
- All agents do not have the same skills,
- Call types may have different priorities, so all calls are not handled on a first-come,
first-served basis, and
- Skills-based call routing delivers greater economies of scale, thereby reducing the
number of agents needed.
Using skills-based call routing, the percentage of time an agent will handle each call
type depends on call routing rules (such as conditional queuing), changing call priority,
queuing to backup skills and time of day and day of week. Availability also depends on
other agents' skills and schedules. Since scheduled agents may be used for a variety of
call types, the number of agents needed for each call type depends on which agents are
scheduled. This revelation is very significant since it identifies a circular problem.
That is, the exact number of agents can only be determined after schedules are created;
however, the schedules needed depend on the number of agents required.
Erlang C can be used as a single step of the process of calculating agent requirements
for skills-based call routing. One option is to calculate agent requirements for each call
type independently, then apply an efficiency factor to lower the requirements, since
multiskilled agents provide greater efficiency. Another approach is to add the call volume
of all call types and calculate the weighted average handling time (AHT) for each
interval. Since the workload was combined, the Erlang C calculation incorrectly assumes
that all agents are cross-trained. Therefore, the number of agents must be increased by a
factor to account for lower efficiency.
The strength of these approaches is simplicity. Their weakness is accuracy. The degree
of accuracy is limited by the method used to adjust requirements so enough agents are
scheduled for each call type without overstaffing or understaffing. To be accurate, this
adjustment must consider the true ACD call routing and the impact of which agents are
scheduled and which skills and skill levels agents possess.
Forecasting Using Multiserver Queuing
Another forecasting method uses multiserver queuing formulas to calculate agent
requirements. This approach is used for call centers with ACDs that assume agents within
agent groups possess identical skills and provide an approximation of the multiskill
efficiency gained by skills-based call routing. However, these formulas include
assumptions that may not be true. Multiserver queuing formulas assume that calls are
routed to separate queues for each agent group or a common queue for all agent groups.
These assumptions are incorrect if calls are simultaneously or conditionally queued to
agent groups or if agents possess different skill levels. The accuracy of calculated agent
requirements would be negatively affected if these assumptions are not valid.
This approach has other limitations, since agent requirements are calculated for skill
sets rather than call types. For example, if a call center handles English and Spanish
callers and agents have the English, Spanish or bilingual skills, then these formulas
would calculate requirements for English-speaking agents, Spanish-speaking agents and
bilingual agents (instead of just the requirements needed for English calls and Spanish
calls). This approach removes the flexibility to determine the best set of schedules
(i.e., using the best mix of English-, Spanish- and bilingual-skilled agents). Another
difficulty is determining how many total agents are needed for the English and Spanish
calls, since the percentage of time each bilingual agent will spend handling English and
Spanish calls is not known.
Multiserver queuing formulas cannot accurately calculate multiskill efficiency.
Additionally, these formulas cannot be used to forecast agent requirements when agents are
assigned to individual skills and skill levels.
The forecasting problem is not "How many agents are needed per agent group?"
but "How many agents are required per call type considering the multiskill efficiency
gained through skills-based call routing?" The multi-server queuing approach does not
solve the problem.
Skill Scheduling
The difficulty of scheduling for skills-based call routing is determining how each
multiskilled agent will be utilized for each call type throughout the day in order to
determine if enough agents have been scheduled for each call type. There are a few
scheduling techniques that attempt to solve this problem.
Skill scheduling for a single call type at a time - Beware of skill scheduling
solutions that assign multiskilled agents to one call type per scheduling interval. For
example, agents with sales and service skills might be scheduled for sales calls from 8:00
a.m. to noon and service calls from 1:00 p.m. to 5:00 p.m. These methods do not match ACD
skills-based call routing, which would route sales and service calls to these agents
throughout the day.
Schedule from most to least skilled agents - If agents are trained first for Skill 1,
then Skill 2 and then Skill 3, the following approach may be used for scheduling:
- First, schedule the most skilled agents (agents with Skills 1, 2 and 3) against the call
type forecast requirements that use Skill 3.
- Next, add the requirements of call types that use Skill 2 to those that use Skill 3 (and
apply a factor to decrease the combined requirements) and schedule agents with Skills 1
and 2.
- Finally, add the requirements of call types that use Skill 1 to those that use Skills 2
and 3 (again decreasing the combined requirements) and schedule agents with Skill 1.
This approach does not consider routing rules that affect agent availability such as
conditional queuing or queuing to backup skills (i.e., the approach assumes that if an
agent with Skill 1 or 2 is available, then the ACD will deliver calls unconditionally to
either Skill 1 or 2). This method cannot be used when agents have individualized skills,
such as language, and are not uniformly trained to be in one of a few different skill
sets.
Schedule All Agents, Then Simulate
If the above scenario will not work in your call center, then it is possible to schedule
all agents against the combined requirements for all call types, then iteratively
simulate, analyze and update schedules.
In order to do this accurately, schedules must be simulated with the same routing rules
that will be executed in the ACD. After simulation, changes will need to be made to the
schedules, since they were initially created without regard to individual call type
requirements or agent skills. This process of adjusting schedules and simulating must be
repeated until an effective set of schedules is created.
Once schedules are finalized, simulation must be used again to provide the data needed
to accurately adjust agent requirements based on who is scheduled. This adjustment is
essential to provide accurate agent requirements for management of schedules.
The only weakness in this approach is that the iterative process of manually simulating
and then adjusting schedules is very time-consuming and subject to human error.
The Preferred Solution For Skills-Based Call Routing
When mathematical formulas alone cannot be used to model a complex system, simulation
techniques can provide accurate modeling tools. Therefore, the preferred solution must
include a simulator to accurately analyze a set of schedules. The simulator would
calculate agent requirements by call type, including the economies of scale gained by
multiskilled agents and would calculate agent availability by call type.
By itself, a simulator is impractical because of the time needed to analyze results,
adjust schedules for a better solution and repeat the process multiple times until
schedules are acceptable. Therefore, the preferred solution embeds a simulator into the
scheduling program. This allows the scheduling program to automatically generate
schedules, simulate network and ACD call routing, analyze the results, determine changes
to schedules and adjust schedules to determine the best answer. The resulting output of
this preferred solution is:
- Agent requirements that account for economies of scale gained by using multiskilled
agents,
- Number of agents available by call type, and
- Schedules refined to meet call center goals for service level and maximum occupancy.
This solution accommodates the many variables and the complexity associated with a
multiskill environment and automatically creates accurate forecasts and effective working
schedules.
Recommendations
Accuracy is the key to successfully forecasting and scheduling for skills-based call
routing. Accurate forecasting and scheduling is needed to consistently meet and exceed
service level goals without significantly overstaffing. Without accurate scheduling for
skills-based routing, call centers will consistently miss service-level goals because of
understaffing or exceed labor costs due to overstaffing.
To overcome these problems, call center managers can create accurate forecasts and
schedules by including simulation in the workforce management process. There are varying
levels of integration of simulation software into workforce management products. The most
integrated and automated solution combines simulation within the scheduling software and
allows the circular nature of this problem to be handled by automatically creating
accurate forecasts and schedules.
Paul Leamon is a senior systems engineer with IEX Corporation. Paul joined IEX in
1990 and has more than seven years' experience guiding the TOTALVIEW Workforce Management
product design at IEX. He manages the system engineering group that designs the new
features in TOTALVIEW. His experience includes extensive work with customers to identify
the system requirements needed to provide solutions that match customers' needs. |