How to Get the Most from Your WFM System
1 May, 2004
By: Mark StanleyGetting the most out of any workforce management (WFM) acquisition involves thinking through the reasons why your contact center made the purchase in the first place. Even if you were still doing things without a fully automated system (let’s see a show of hands – how many of you are still using some form of an Excel spreadsheet for scheduling?), there would be several ways to optimize the process of workforce management.
First, let’s define what we’re talking about. To make sense of the entire WFM process we need to break it down into three main areas: forecasting, scheduling and tracking.
• Forecasting is the process of predicting the probability of an event (most often telephone calls, but it could also represent other channels such as e-mail or web chat) happening at some point in the future. The best indicator of a future event comes from a “seasoned” look at past events. That is, the more we know about the nature of your business and the type of customer you interact with, the more accurate we can be in anticipating what is likely to happen. Typically, the information we are looking at comes from reports generated by your phone switch because the “events” we are speaking of usually are telephone calls.
• Scheduling is the process of finding the right mix of agent skills that correspond to anticipated caller needs. That means finding the right agent for the job at the right time of day to handle a call. On the face of it this ought to be pretty straightforward. After all, the forecast indicates when the calls will come in. But it turns out that the forecast is the easy part!
• Tracking compares what actually happened to what you thought would happen so that you can evaluate the effectiveness of the process. Here we’re interested in analyzing caller behavior to see how accurate the forecast was. We’re also interested in agent behavior to evaluate the schedule. The result of the tracking process loops back into the forecasting and scheduling process as the part that adds the “seasoning” mentioned earlier.
Now that we understand what the WFM process looks like, let’s further define the optimal state for each component.
Optimal forecasting happens when we utilize highly accurate information to predict the future. The level of accuracy is measured by the size of the delta when comparing forecast to actual events, with an optimal delta of less than 1 percent.
Optimal scheduling is observed by comparing service level objectives to staffing to ensure that overstaffing and understaffing have been minimized. For example, if your service level objective is to answer 80 percent of the calls within 20 seconds, a result of 75 percent in a given half-hour would indicate understaffing, while 85 percent would indicate overstaffing—neither of which is desirable. The closer you can come to 80 percent in all half-hour periods, the higher the standard of service that will be delivered to your customer at the most economical cost to you.
Optimal tracking is found in an operation that tracks deviations from scheduled events in a real-time manner and uses that information to adjust staffing levels throughout the day. This type of intra-day management helps you to keep on track as the day unfolds.
So, how do you achieve the optimal WFM environment? Here are some specific ideas for the three key areas:
Forecasting
Think of calls handled as just another type of work. Let’s imagine that there was some way to symbolically represent every call and that we could pile all of the calls on the in-box on our desk. The size of the pile of calls tells us how much work we have to do today.
Reports that come from your switch will show the number of calls offered and abandoned; the difference is the number of calls actually handled. These are usually grouped by call type (e.g., sales, service, etc.). In addition to the number of calls, we are also interested in the average handle time (talk time and after call work time) for each call type. When we multiply the number of calls handled by the average handled time we have the amount of work that needs to be done—the size of the work pile.
We can create one of these models for any period that we might want to look at—hours, days, weeks, months or seasons. What we will find when we plot the result is a pattern of work that has some regularity to it. That pattern has to be “seasoned” with other information we have about the future, such as changes we plan to make in operating hours, new lines of business, and so on. The key in developing accurate forecasting is ensuring that the source information (i.e., call switch reports) are accurate and that the “seasoning” applied realistically depicts what is known about your future business changes. How many times have we had a manager tells us to “use last year’s numbers and add 10 percent” as a way of predicting future volumes? That kind of methodology is useless.
Again, regardless of whether you are using WFM software or an Excel spreadsheet, the assumptions you make are critical to the accuracy of the result. So spend a fair amount of time analyzing key business drivers to really understand those things that will impact call volumes and then quantify their influence. Additionally, it’s a good idea to periodically look at call switch reports and compare them to the invoice you receive from the phone company. For lots of reasons they will never match exactly, but they ought to be in the same ballpark. If your switch indicates you are handling 100,000 calls per month and you are being billed for 200,000 calls, something is terribly wrong. You won’t be able to do any realistic forecasting until you get this issue resolved.
Assuming you have a high comfort level with the number of calls and average handle time, and you have accurately identified upcoming business changes, you are now ready to do some forecasting. Although it is good to look at a full year to see how seasonality can affect call volumes, the most accurate forecasting is going to be limited to about 90 days. You should track forecast vs. actual results on a daily basis and calculate the variance. As mentioned earlier, well-tuned systems should have a variance of less than 1 percent. If the variance is over 5 percent you should check your assumptions and re-forecast.
The output of the forecast should show the number of calls (or amount of work) that is anticipated. You can use a basic Erlang C calculator to determine the number of agents you will need to handle the volume. Most workforce management companies have Erlang C tools available for download from their Web sites. Note that the result is only going to calculate the net number of agents needed—in other words, bodies in chairs. They won’t indicate how many people you actually need on the payroll because the amount of shrinkage (e.g., sick calls, vacation) varies at every company.
Scheduling
In order for the scheduling portion of the WFM software to function it really needs to know more than just calls offered and average handle time. Part of this includes basic shrinkage elements (time agents are paid to take calls but are not available) such as breaks, meetings and training, so be sure to plug those numbers in to the appropriate part of the software to be sure it is accounted for.
Another key element that must be accounted for in the software is known as “occupancy”. This is the amount of time you expect agents to be in the chair talking to customers. There is a natural tendency to expect the answer to be 100 percent, but in reality that is not such a good idea. The reason is that occupancy is inversely related to service level. That means that if agents were indeed busy 100 percent of the time, every call would need to queue. If that happened, service levels would plummet. Besides, your agents would suffer from fatigue, which would be reflected in high agent churn rates.
Finally, no matter how well the schedules generated by the software meet the needs of the caller, you must not underestimate the need for looking at the overall process from the agent’s point of view. The “human element” is difficult to predict accurately because it’s largely intangible. So, unless you have an operation where everyone works the same shift, eventually someone is going to get stuck with undesirable hours. This is where the art of scheduling comes to play: finding a good balance between optimal schedules produced by WFM software and schedules that an agent will actually work that produce a service level that customers can live with.
Tracking
The tracking process captures real-time events and uses the information to adjust scheduling for the remainder of the day. Assume that the schedules have already been posted when you come to work in the morning and that all agents understand when they are expected to report for work. In a normal day there will be some number of agents who will not report for work on time for circumstances that are usually not within your control.
For example, there will be sick calls. Some agents will encounter heavy traffic on the way to work. There could be a weather problem. There always will be some deviations to the schedule. If you have anticipated this and factored it in to the scheduling process you will have some overstaffing to handle the shortfall.
So, the first thing is to ensure that you have a process in place to capture schedule deviations as they occur. This might involve designating a specific phone as the “agent hot line” where agents can call to report when they will not be reporting on time. During normal business hours the phone would be answered by the scheduling agent, but there should be a process for after hours as well because many agents will be reporting changes before the start of the next work day.
As schedule deviations become known they should be loaded in to the WFM software. Throughout the day the daily work schedule should be “re-optimized” by comparing current incoming call volumes and schedule deviations and then producing an output that modifies activities such as breaks and lunches to help meet service level for the remainder of the day.
Understand that there are two components to workforce management. One relies on the accuracy of the tool used to manage the process as described above, while the other relies on the skill of the management team. In other words, you could produce the most perfect forecast and schedule in the world, but if agents don’t show up for work it’s all useless. Software alone won’t fix the problem.
Effective workforce management starts with setting expectations of the staff in a clear, unambiguous way: agents are expected to report on time each day that they are scheduled to work. Compliance with the expectation will be rewarded. Failure to meet the expectation will result in consequences. As long as agents understand that this is the policy and that you intend to enforce it, half the battle is won.
When you combine a good policy with well-tuned workforce management software, the results will be impressive.
___________________________________________________________
Root Cause of the Problem
A large healthcare provider in Southern California operates several call centers that provide customer service and claims resolution, ranging in size from 75 to 150 agents. Although workforce management software had been installed and basic training provided for the end-users, the overall results were unchanged.
An assessment of the situation revealed three basic issues that had not been addressed when the WFM solution had been installed:
1. The parameters used for staffing were incorrect.
2. The data feed from the ACD was not configured properly, so call volumes were inaccurate.
3. No methods were in place to continually evaluate the workforce management process and identify shortcomings.
The first order of business was to correct the ACD output, to accurately reflect the queues and to replace all inaccurate old data. Once completed, new models would be developed to reflect the true picture of call volume and handle times.
Next, every single parameter in the WFM system was analyzed to ensure consistency with workplace rules and desired management outcomes. For example, occupancy (the percent of time agents handle calls vs. wait for calls to arrive) had been set at 100 percent, which is unrealistic and thus unattainable. This parameter was changed to 85 percent, which would produce a service level of approximately 83 percent. These modifications produced a better picture of the real staffing requirements necessary to attain the desired service level objectives. Here you can see the effect of the change:
Agents: Before 66; After 87
Occupancy: Before 100%; After 85%
ASA: Before 6.96; After 0.39
Abandons: Before 5.80%; After 0.33%
Finally, the entire team collaborated to create procedures that would identify shortcomings in a timely manner so that corrective action could be taken quickly. One example involved comparing call handling reports generated by the ACD with corresponding information logged by the WFM software. As long as the two were identical, there was no problem. But when a discrepancy was found, a full investigation would be launched to understand why. This would be done before any more data was used for modeling. This simple procedure ensured a high level of data integrity, which translated to more accurate forecasting.
As a result of the changes the overall service level improved dramatically, with a corresponding rise in customer and agent satisfaction—a win-win situation!
