Demand Forecasting is an important principle in WorkForce Management. It gives visibility to the nature and volume of the work required in the future, serves as an analytical tool to plan resources and, enables understanding of different metrics such as labor shortages. In this post, I will discuss general considerations to forecasting when using time series and in a later post, I will show how to use that input to plan the capacity required to meet such demand.
The demand for better forecasting has been rising over the last few years with the introduction of data-driven approaches to business operations. The fact that businesses can now easily capture, store and process large volumes of data has created opportunities to improve legacy processes by introducing novel methods to drive operations based on different data driven techniques. Over the last few years we have heard the term “Data Science” and how it has (and will) change the way we do business and interact with day to day operations. Many field service businesses are already engaged in some type of forecast and capacity planning task which serves as input for the rest of the workforce management operations.
The Forecasting Task
Businesses should give careful thought about the forecasting task, the context in which it is done and the importance it carries in the workforce management cycle. Firstly, whoever is responsible for the job should be able to determine what problem the forecasting is trying to solve and, equally importantly, what problem it isn´t trying to solve. These aspects are vital since everyone involved should have the same expectations about the outcome.
Defining the Problem
Time series forecasting is a powerful technique to make predictions in the future. It is well established and there are multiple methods available for analysts and scientists to use. However, there are many considerations that need to be well-thought- out before using any available model. The first question that needs to be solved is: What are we trying to forecast? It may sound simple and obvious but having this question answered correctly is crucial to the success of the forecasting task.
This question carries many elements that may seem trivial but has an impact on the correctness and accuracy of the model selected. The business unit needs to determine if the forecast is targeting tasks, work orders or inbound calls. Once the output data has been determined, the business unit must review the different subtasks that may arise and if these should be included in the model.
Additionally, there should be an agreement about what the forecast is not; this will guide the forecaster in understanding whether to use an explanatory (why), anomaly or any other modeling technique (ensuring alignment to business goals).
Adjustments and Cleaning Considerations in Time Series
Once the people involved have a clear picture of the outcome of the process and assuming the raw data is readily available, there are a series of adjustments and cleaning tasks that need to be performed. Missing data is a common issue that comes up when dealing with complex datasets and can arise due to multiple reasons such changes in the organization’s mobility systems or low engagement from field personnel. If missing data is unaccounted for, it can significantly impact the outcome of the forecast and the operations of the business thereafter.
Depending on the type of data issue faced and the intended outcome, the process undertaken by the forecaster may be:
- Quantify the effect, remove the data, create the forecasting model and then re-apply the effect.
- Remove data only.
A different adjustment that should be considered is the level of granularity required by the time series forecast and the level of granularity provided by the data at hand. This may lead to the following questions:
- Does the business require a monthly or weekly estimate?
- Does the data provide the information at a daily/weekly or monthly level or is data transformation in order?
It may seem minor to aggregate data from a daily to a monthly level, however, depending on the business this may have a significant impact. Many geographies and businesses react differently to holidays, requiring data manipulation to ensure these are treated appropriately. For example, the Easter break does not fall on the same week on a yearly basis, thus needing to be adjusted to reflect the relevant period. These differences make it important to precisely define the business rules that will govern the granularity of the data presented.
Choosing the model
Now that the data is clean and all the assumptions and inputs are clear, the next step is to create and validate the time series model. The autoregressive integrated moving average (ARIMA) is a form of regression analysis used to make predictions based on historical patterns which contain data with seasonality features and increasing or decreasing trends. Nevertheless, there are many different models are available ranging from ARIMA based models, Logistic models and much more.
Wrapping it up
At WorkForce Delta, we have extensive experience in helping business forecast their demand, as well as understand and make sense of their data to ensure that decisions made reflect not only the needs of the organization but industry wide best practices.