Skill-Based Scheduling

How Much Precision Do You Really Need?

By Maggie Klenke, The Call Center School

Skill-based routing (SBR) has been available in Automatic Call Distributors (ACD) for over a decade and many contact centers have adopted the concepts to one degree or another. There is a capability to queue calls simultaneously to more than one group of agents for handling (or in steps based on  script). It is also possible for an agent to have more than one skill and be utilized for any combination of calls in those skills that are required (with priorities for certain types of calls if desired). No more moving an agent from one call type to another to follow the peaks in workload – the system will automatically find the resources in the order that is predefined. The benefits of SBR are significant. While having universal agents who can handle any type of call will achieve the most efficiency in terms of staffing, this is often not realistic. It may require weeks of training to achieve mastery of all the types of calls in a single center and some calls require a special personality (i.e., sales or collections) or special skills that are acquired over years, not weeks. It is also true that having a dedicated team of agents for every call type will require the  maximum number of total staff. SBR helps us to utilize our staff in a way that may not be fully universal but is better than individual dedicated teams, so we can gain some of the economies of scale. It also gives us a way to train new hires
on one of the call types and let them take just that type of call until they have demonstrated mastery and have gained the confidence needed to take on another skill. Customers are likely to talk to agents who have the requisite skills to solve their problems which increases customer satisfaction, reduces handle time and errors, and can increase first call resolution.

We all know there is “no free lunch” so there are some tradeoffs to get these benefits. The complexity of the workforce management (WFM) process is increased, often significantly. Agents who used to be able to trade shifts with anyone in the center now can only trade with agents who have the same skill combination. The same challenge exists in bidding for vacations, time off, and overtime. Senior agents (or high performing ones) who used to have the best shifts now have to bid against a smaller pool of similar-skilled agents and may find themselves in unattractive shifts.

One of the biggest challenges is knowing how many people are needed with each skill or combination. This used to be a simple question answered with Erlang C or a variant. However, when agents have a combination of skills, the specific combination makes a big difference. Hiring staff who will start training with Skill A may not seem to address a shortage of resources in Skill B, but might relieve those agents with both skills of taking Skill A calls so that they can concentrate on Skill B. On the other hand, what happens when an agent calls in sick? We used to know just what to do to backfill for that person, but now it really matters what set of skills that person has. If the missing agent has 4 skills, must we find a 4-skilled person to replace her or could we make it work if the substitute has only a subset of those skills? Might we find ourselves in a situation in a very low call volume queue with no one logged in during some period of the day? Analyzing these possibilities is very complex.

We could treat the efficiency gain somewhat like we handle shrinkage. In that case, we would establish a percentage increase in the utilization of the staff and apply this percentage as a reduction in the requirements across the board. Unfortunately, while this is a relatively easy approach, it is likely to be fairly inaccurate when applied equally to all time frames and skills.

Software tools available in the marketplace approach it in a variety of ways. Some use sophisticated simulation programs to try to anticipate the exact mix of calls in each period and the precise staffing compliment in each combination of skills. These processes can take hours to run even in a small to medium sized call center. They are based on assumptions about the mix of call types by period and shrinkage percentages that may be applied across the board or by skill. The process makes assumptions about which agents will show up and who will be absent which is likely to vary daily if not hourly. That leads us to the question of how much precision in the planning is really useful.

Generally speaking, the assumptions used by the simulations are based on the call volume and AHT forecasts along with shrinkage losses which are themselves based on historical patterns. How accurate is your forecast? May centers are delighted to achieve + or – 5% for the day or week, but may have variances at the half-hourly level nearer 10 or 20%. The smaller the workload for a specific call type, the more likely it is that the variances will be in the higher percentage ranges in any given period.

What about the assumptions about which staff will actually be logged in for any given period? How precise can the simulation be in choosing which agents (with specific skill combinations, preferences, and skill levels) will be logged in? If the schedules are run weeks ahead of time, some agents may have terminated and others trained on new skills. Even a schedule for the upcoming week may experience changes as agents request time off, team meetings are rescheduled, and the myriad of day-to-day challenges are managed. The patterns of agent availability are not easily predicted with much precision in many centers. At the start of the day, the assumption might be that you will experience 7% shrinkage within the day that is not reflected in the schedules (sick, tardy, long team meeting, etc). But which agents will it be? It definitely makes a difference.

What if you change the routing logic in the ACD? Many centers make frequent changes to routing rules or ‘vectors’ in response to changing business needs in an effort to get the calls to the right agents. Some change the agent skills and priorities to bring the agents to the calls. Such changes make the simulation results invalid and mean that the simulation must be reprogrammed and run again – a potentially labor-intensive and time-consuming process.

In the end, the simulation has made a number of assumptions that may or may not turn out to be accurate when creating the schedule. As the workday begins and the call arrival patterns show themselves to be somewhat different than expected, and/ or the staff that is in place has a different mix of skills than was assumed, it is necessary to rerun the simulation to adjust to the current conditions, and of course those conditions are constantly changing. If the re-simulation process takes a long time to complete, it may not be a practical option for intraday adjustment planning.

Another approach to developing schedules for a skill-based environment is to use a sophisticated algorithm (instead of simulation) that completes an acceptable level of precision for longterm planning and shift bids or rotation plans while allowing for frequent updates as the time to work the schedule draws nearer. Such algorithms can run in minutes rather than hours compared to the simulation process. Intraday adjustments can be calculated quickly so that staff can be more precisely matched to the needs when the workload and staff skill mix is better known.

While this approach may seem on the surface to be less precise than a simulation process, in the end the speed with which the recalculations can be done as needed will likely result in as good or even a better match of staff to actual workload. Since that is the goal of any workforce management process, it makes sense to do it as quickly and easily as possible if the results are as good or better than other methods.

Workforce planning for skill-based environments is often highly complex requiring the WFM team to respond to changes and requests quickly. When an agent asks for time off in the afternoon, the team needs to be able to determine what the impact is likely to be. If an unexpected surge of calls is received in one call type, planning for overtime or other adjustments will need to be done quickly to be effective. When computer problems extend the AHT or greatly shorten it, matching the staff to the new requirement is something that must be accomplished quickly. After all, the role of WFM is to get the right number of resources in place to match the workload as closely as possible in each period of the day. This is how the center maximizes the consistency of the customer experience, ensures a steady and reasonable workload for the staff and minimizes the expense. In a sense, the planning needs to be reasonably accurate allowing the ACD routing process to serve as a “cleanup batter” by moving the calls to the available staff in the most effective way in real-time.