Before we had skill-based routing (SBR) in our center, I could easily answer the question, “how many agents do we need?”. But now it doesn’t seem like there is a reliable answer to that question. Is there any way to address this?


You have raised one of the biggest challenges workforce planners have with the SBR configuration. When we had single-skilled agents or even universal agents, the number required to meet a defined service goal was a simple calculation using the Erlang C model with adjustments for shrinkage and scheduling inflexibility added on. It was easy to know what the impact was of missing agents, what overtime schedules should look like or which agents we could offer time off. We can still figure it out in an SBR environment if all agents have the same configuration of skills but just different priorities in handling calls. Most centers have configurations in which some agents may have only one skill, others two or three skills, and maybe a few who can do any type of work. This may be a function of different call types or varying media such as calls, emails, and web chats. Now the answer to the “how many agents” question is much more difficult. The best answer (but unacceptable to operations managers) is “it depends.”

Tell me what time frame to use for the forecast, identify the agents who will log in and the skill mix that each has, and establish a reasonable shrinkage assumption and we can identify the total number of staff needed. But change the workload by skill, the shrinkage assumptions or the skill mix of the staff and that number can change. Essentially there is a continuum of possibilities. At one end, we have the single-skilled agent configuration that would require the greatest number of personnel. At the other end, the universal agents who can handle anything would require the smallest number of staff. SBR’s requirement is somewhere along the continuum depending on how many skills each person adds to the mix. It is different on Monday than on Saturday due to the schedule of agents who should be working, and different still between planned and actual intraday due to absences, tardies, long team meetings, etc. If Mary who has one skill reports out sick it will have quite a different impact than if Lucy who has five skills is sick. While sophisticated WFM programs with simulators or SBR algorithms can provide a reasonable estimate of the staffing requirements, the reality is likely to be somewhat different if the assumptions built into the plan are not accurate. But this is as close as we are likely to get in answering the question.

Have a tough question?
Send it to and we’ll try to find an answer!