How Many Staff Do You Need, And How Do You Know?

By Ric Kosiba, Vice President, Bay Bridge Decisions Group, Interactive Intelligence

[dropcap2 variation=”purple”]L[/dropcap2]
ast month, we were discussing staff planning with the Vice President of Customer Care at a well-known financial institution. We talked about simulation modeling of her contact center operation and ways to make sure that their week over week staffing was “just right” — not too many agents and not too few. We covered just-in-time hiring, math models for finding the best balance between hiring, overtime, and controllable shrink, and how to manage the blend of activities associated with multi-channel agents. We also discussed forecasting through seasonal peaks and valleys, and calculating the weekly staffing requirement. And then we were asked a great question:

“But how do you know?” This is one of those questions that goes directly to the core of the workforce management problem. How do we know that our staffing requirement is correct?

Why this is a great question?

There is a truth in contact centers: they are all completely different. There are multi-channel centers with agents handling phone, email, and chat in different combinations across different times of the day. Many networks are multi-skilled with different agents handling different mixes of types of customer interactions. Workflows are likely different from one company to another.

Each contact center network handles different sorts of customers, even in the same industry. The purposes of the contact are different, and hence so are the handle times, the distribution of contact arrivals and handle times, as well as the customer patience and their willingness to abandon a call. Also, and this is not at all trivial, both inter- and intra-company, the value of the customer interaction can be very different.

Heck, even the same contact center network behaves differently at different times of the year.

“How do you know?” is a great question. It goes straight to the core, to the purpose of the workforce management discipline. We need to be prepared to answer this.

How Do You Know?

So how do we know we have the right number of staff and the best plans? Before we all turn to our capacity planning spreadsheets and attempt to answer this, let’s think about what goes into explaining this seemingly simple question.

Knowing that we are staffed correctly requires that we have the appropriate service standards for each of our many contact types. And therein is the rub — what is “appropriate”? Should we run at 80 percent of all calls handled in 20 seconds for our customer service call center? Should it be longer or shorter? Should it be different by season?

There is so much that goes into this decision — the corporate mission and brand identity (premium brand or low cost provider), customer expectations, the availability of alternative channels, the availability of competitive alternatives, the cost of servicing, and the revenues (or perceived value) of each contact.

But determining the appropriate service goals for your contact centers is critical to being staffed correctly. This (usually) single metric, the service goal, drives costs and revenues for sales centers.

Once we have a goal that we have analyzed and proven, the next step is to develop a process that performs four important steps:

  • First, gather appropriate and clean ACD, payroll, and workforce management historical time-series performance data and populate a contact center history database. This data will be used to produce forecasts and validate the accuracy of our analytic processes.
  • Second, use this database to develop forecasts of all important contact center metrics, such as volumes, handle times, agent sick time and absence, customer experience scores (i.e., Net promoter Score, agent quality, or first call resolution) and agent attrition. It is important that we forecast at the appropriate level of detail, by center and staff group because each contact center will likely have a different seasonality associated with items like shrink and attrition. Note also that different metrics, with their different trend and seasonality may require completely different forecasting methods.
  • Third, apply a mathematical model to transform your volumes, handle times, and shrinkage forecasts into an interval by interval, and week over week staffing requirement that exactly hits the ideal service goals. Many planners utilize an Erlang C equation or an assumed occupancy calculation to determine this staffing requirement. Other companies are beginning to also use discrete-event simulation to determine their staffing needs.
  • Finally, we must look at the staffing requirement and compare it to the expected agent staff availability week over week (after agent attrition and uncontrollable agent shrink), and determine when we are over or under staffed, and determine a hiring, overtime, under-time, and controllable shrinkage plan. Many capacity planners perform this step by hand, looking at an over/understaffed chart and “guessing and testing” when to hire. There are, however terrific technologies, like integer programming that that will develop very efficient “just-intime” plans.

So, how do we determine with confidence that each step we are taking will yield the right plan? Let’s look at each of the four steps.

First, it is very important to validate and prove that our database has clean and consistent source data. While this sounds pretty simple, it really is not. This step requires that definitions of data — most often from different data sources — are understood and consistent, and that in the end there is a method for validating the accuracy of contact center data.

Here is an “easy” check. Since our database will include staffing, shrinkage, and ACD performance data, we can check to make sure that the topline staffing, minus agent shrinkage, totals to the bottom line staffed hours that our ACD records. We know how many hours our agents were on a contact or in a ready state, we keep track of shrink, and we know how many agents we have on our payrolls, so we should be able to reconcile these.

Any difference, we call “lost time” and it represents unaccounted agent time. If lost time for your contact centers is relatively low, then your data is relatively clean. In other words, you know your data is consistent.

The second step is to develop forecasts of all important plan drivers. There are standard forecast “goodness” processes and metrics that can check the expected validity of a forecasting technique. Typically, they go like this: the analyst holds out data from their time series history and applies a forecasting methodology (like Holt-Winters) to the historical time series data to develop a forecast. This forecast is compared to the “hold out” actual performance data. There are a series of metrics (like root-squared mean error) that can judge the accuracy of the forecast on this hold out data.

The third step involves converting volumes, handle times, and service goals into a week over week staffing requirement. How do we know that this step is right? It is actually a very simple process — simply take several week’s work of ACD data, plug the actual calls offered and handle times into the staffing methodology and plug the historical service level achieved as the method’s goal. If the staff requirement predicted by the model is the same as the staff on hand for that historical time period, then the model is “validated”.

But here is an observation: the Erlang equation always overstaffs; its requirement is always too high, when compared to actual performance. This is why many contact center planners have switched to discrete-event simulation as their requirements generator of choice. Simulation does not usually have an overstaffing bias.

The final step is determining a hiring and overtime plan. This step is difficult to validate, without developing a perfect plan to compare. The good news is that there are systems that will develop perfect just-in-time hiring plans for you using mathematical technologies such as integer programming. But if you are using an integer program to determine how good your planning process is, then why not use the perfect plan in your operation?

By checking and validating your planning process you will know that your forecasts are right and your staffing and hiring models are valid and optimal. When your senior manager asks “How do you know?” you’ll be able to answer confidently.

Ric Kosiba, PhD is a charter member of SWPP and vice president and founder of the Bay Bridge Decisions Group at Interactive Intelligence.