The concept of skill-based routing has been around for about 10 years, but call centers are still struggling with how to take advantage of the benefits without suffering from the negative implications.  One of the biggest challenges is figuring out how to schedule staff so that all skills are covered with the minimum number of agents.

The basic premise of skill-based scheduling is based on economies of scale.  That is, it takes fewer agents to handle a combined workload than it would to handle the same work in separate groups.  For example, let’s assume that the call center will get 120 calls of 300 seconds each in an hour, or 10 hours of workload (120 X 300 seconds / 3600 seconds per hour).  That is 10 hours of work to do in a one-hour window of time.  If we want to achieve an 80% in 20-second service goal, then we will need 14 bodies in chairs for that hour. If we had 5 such groups each handling a different kind of call, then the total staffing required would be 5 X 14 or 70 agents.

But if we could cross-train the agents so that they could all handle all of the calls, then the total staffing required is not 70 agents but 57 agents, a savings of 13 people.  The challenge will be the investment required to achieve that level of training and capability among the whole group of agents.

So skill-based routing allows us to gain some of the benefits of cross training without having to go to the extreme of creating the universal agents who can do anything.  Let’s assume that your call center trains new hires to handle one simple kind of call so everyone in the center is trained on this call type “A”.  Once the agents are proficient on A, then the center will train them on another call type, but they don’t all get the same training at this point.  Some will learn call type B and others will learn C depending on the needs of the business.  Now we have some new agents who only have skill A, some who have skills A and B and a third group who has skills A and C.  This is where the scheduling starts to get complicated.  Let’s make the following assumptions:

The workload for each of the three skills and the required agents are computed as if the groups were kept separate.  Added together that means we need 41 total agents, but if totally cross-trained, we could achieve the results with only 36 agents.  The question is, how many agents for each skill are needed to handle the workload if the agents have a mix of skills?  The answer is somewhere between 41 and 36, but can we determine the exact number for each skill combination so that we can create a schedule?

One thing is certain – we must have enough agents in the B and C skills to handle the total workload for each of them since no other agents can assist them.  So we can establish a minimum requirement of 8 agents with the A & B skill and 14 agents with the A & C skill.  The number required in the A skill is a function of the available time the dual-skilled agents will have left after they have answered their own calls.

The occupancy of the 8 agents in the B skill is fairly low at 62% handling their own calls, so there should be some time available to help out in the A skill.  Occupancy for the 14 agents in the C skill will be somewhat higher at 71%, but there will still be some time available to help out in A.  The balance of the calls will need to be handled by A-only skilled agents.  Let’s assume that we are willing for the total occupancy for all agents to reach 85% (since that is approximately what would result from 36 fully cross-trained agents and is within the burnout prevention range). 

In the chart above, we have assumed the current level of occupancy of the B & C groups and calculated how much additional occupancy could be available to fill with A calls.  Multiplying that percentage times the size of the agent group reveals the number of agent hours that could be filled with A calls.  With a total of 19 needed in the A group, subtracting the availability of the B and C teams reveals that we need 15.2 agents who have the A only skill for this hour.  This is a total of 37.2 agents for this hour rather than the 41 for separate skills or the 36 for universal agents.

Unfortunately, it is a lot more complicated than this suggests.  The assumptions in this exercise were that the AHT is the same for all call types, and that the agents all log in at the beginning of the hour and are there for the entire period.  It also assumes that the distribution pattern of calls within the hour will be such that peaks in B and C will not coincide with peaks in A, allowing the available occupancy to be used for A without affecting the service in B and C.  These are not very realistic assumptions.  But doing these calculations manually suffers from these kinds of limitations.

For centers with 10-20 skills and a wide variety of combinations among the agents, it is clear that this manual method will be nearly impossible.  You can estimate the total needed for separate skills and the total needed if all combined, and you will know the right answer is somewhere between them, but where it is depends on multiple variables.  And sure as you get it all figured out, an agent with the 5 skills will call in sick and throw off the entire plan.

To get a more precise forecast of the scheduling requirements, skill-based scheduling really requires a sophisticated algorithm or simulation package that can run multiple sets of assumptions to find the best results.  For those of you trying to do it manually, this method will get you in the ballpark, but its precision is very limited.