By Ger Koole
Last week we had an advanced WFM training in Amsterdam in which we discussed the various ways in which WFM can be improved. An example is shift inefficiency, to some known as schedule inefficiency. It measures to which extend the combination of available shifts hinders us in making an efficient schedule. The situation is as follows: We know for every time interval the required staffing level, for example by using one of the Erlang formulas (see my related post on the Erlang X). Now we try to cover the requirements by the available shifts. This will typically lead to overstaffing: even if the shifts are relatively short, because they consist of multiple periods in a row it is not possible to follow the changes in requirements all the time.
Especially in the middle of the day, when most shifts overlap, there is a lot of overcapacity. The overstaffing is 35%: the total required staffing (summed over all intervals) is 486, the optimal combination of shifts consists of 41 shifts of 8 hours, summed over the 30-minute intervals 656. Thus the overstaffing, or shift inefficiency, is (656-486)/486 = 0.35. (Using 4-hour shifts instead of 8-hour shifts the shift inefficiency is reduced to 6%.) Evidently, shift inefficiency is undesirable, but it is hard to reduce. This requires renegotiation of labor contract or hiring agents with different contracts.
Comparable forms of inefficiency exist in other steps of WFM. The example of shift inefficiency was given for a single-skill call center. There are good reasons to distinguish between different skills. However, treating all these skills by dedicated agent groups can be very inefficient. For this reason certain agents are multi-skilled, they get calls from different skills. The most efficient situation is when all agents are fully multi-skilled. This is often impossible (for example, in the case of a multi-language call center) and always expensive, because of training costs. And it is not desirable, most of the advantages of multi-skilled agents are obtained with the first multi-skilled agents. We see so-called diminishing returns. Consider an example to compute the multi-skill inefficiency.
To analyze the hypothetical situation where all agents are multi-skilled we can use the Erlang X formula. This shows the same performance for 27 agents, 1 agent less than in the simulated situation. The multi-skill inefficiency is therefore (28-27)/27 = 3.7%. Multi-skill inefficiency is undesirable, just as shift inefficiency. But it is equally hard and costly to reduce, leading to a trade-off between single and multi-skilled agents. Finding the optimal trade-off is a challenging task.
Let us finally consider routing inefficiency. While most call centers are aware of the loss of efficiency due to inflexible shifts and single-skilled agents, the effects of bad routing are rarely addressed. To define routing inefficiency, consider a particular situation in which the service level objectives are achieved. Very often this requires reducing skill sets, but rarely adapting routing rules to the current situation. This leads to the situation that routing is usually not appropriate, requiring overstaffing. Routing inefficiency is the overstaffing compared to the optimal routing rule. Determining routing inefficiency is not easy: there is no method to derive the optimal routing rule. But if we find routing rules that are better and require less agents, then we know that the routing inefficiency is at least as big as the difference of these policies.
It is my opinion, shared with Wout Bakker and other participants of the training, that routing is a largely overlooked way to improve call center efficiency. Moreover, it does not entail high training or labor costs. Thus, routing should be a focus area for call centers looking for cost reductions.
I would like to thank the participants of the training for the discussion that led to this blog.
I recently published a book on WFM that covers the subjects of this blog: www.gerkoole.com/CCO.