Volume Demand Forecasting
Updated: Feb 6, 2022
Like any other planning process, the forecasting process for workforce planning is one part art and one part science. It is an art because your judgment and experience sometimes determine the accuracy of your forecast. It's a science because there are numerous step-by-step mathematical processes that can transform raw data into predictions of future events.
There are roughly speaking four approaches to forecasting:
First, using a tool in which advanced mathematical forecasting methods are implemented to assist with data-gathering, data exploration, incorporation of causal factors and automated best-fit modeling, and advanced machine learning techniques such as Facebook Prophet, Amazon forecast's DeepAR, and long short-term memory (LSTM) neural networks.
Using a simpler but systematic approach that the forecaster executes.
Using a non-systematic approach primarily based on human judgment.
Using a combination of the above three.
I would recommend using a strategy that combines all three of the first three approaches. However, this takes time to develop, and in general, simple mathematical methods frequently outperform both more sophisticated techniques and human judgment when compared on an individual like-for-like basis.
One of the issues with pure human judgment is that it lacks objectivity, which often leads to us being overly optimistic. It is also challenging to explain fluctuations in contact volume because they can have multiple causes, each with an unclear impact. Humans struggle to understand because there is no clear relationship between cause and effect. For these reasons, forecasting based solely on human judgments should be reserved for situations with no historical data (such as a new product) or avoided altogether.
On the other hand, relying entirely on a computerized system will rarely outperform human-led approaches because every forecasting method must be able to identify special events, interpret business changes, and so on, which are nearly impossible to achieve successfully and consistently without human intervention. Moreover, because of this human intervention, forecasters must understand the consequences of their interactions with the system, disqualifying some of the more advanced black box-type forecasting methods from practical use.
No matter what tool a forecaster has at their disposal, understanding these calculations is critical to quality check the inputs/outputs and have an informed discussion with business leaders. After all, how can you expect decision-makers to buy in and support your plan if you can't explain how you arrived at the forecast in the first place?
There are four main forecasting components typically used for Operations forecasting:
Time-Series (demand history)– a method based solely on history to extrapolate forward.
Cause & Effect (Causal) – a method best suited for situations with regularly characterized ups and downs due to causal factors or drivers.
Guesswork and manual correction tweaking for a special event – yes, sometimes when there is a lack of accurate data, gut feel is all you have. However, it would be best to look out for future events that might affect the level, trend, & seasonality of your forecast vs. history.
Leadership Review – incorporation of various process subject expert matter views to align the forecast variables to business direction.
Now let's look at the six typical forecasting process steps that usually occur.
1) Define the Characteristics
Before gathering data to feed a forecasting model, sit down and sketch out the characteristics of the object, you want to forecast. This is done to ensure that as you proceed with subsequent steps, you have something to guide you on what data you might need, how accurate a forecast of this object is likely to be, and which forecasting model/method is most appropriate.
There are many characteristics you might consider, but some common ones to are:
Determine the granularity – at what level of granularity do you intend this forecast to provide, e.g., monthly, weekly, daily, or intra-day down to a 15-minute level.
Determine the time frame - whether the forecast is made for a week, a month, three months, six months, one year, or more.
Growth – is this an established market and product, with a steady increase in forecast growth, or is it relatively new and thus likely to be openly volatile full of unpredictability.
Seasonality – are there certain times of the year you are likely to see a larger volume than others?
Customer sensitivity – when forecasting demand for, say customer contact for the national lottery, you are likely to see much larger surges in volatile and short-notice demand (say when there is a prize roll-over) than you would say for a Banking service.
Customer Self-Serve System/Process Generated Volatility – How reliable and mature are the systems and processes that support your customer's self-serve options.
Upcoming changes –will your business make adjustments to the product to improve the proposition or align with new regulations? Knowing these changes in advance allows you to adjust your forecasting process accordingly.
2) Data Gathering & Cleansing
Of course, the source and availability of your data are dependent on how much history exists (if it is a new product, no history may yet exist), your technological setup, and the type of customer channel you are forecasting, such as phone, live chat, email, back-office, retail footfall, manufacturing units, and so on.
If no data is currently available, the information must be derived from expert judgments. If the forecast is solely based on judgment and no actual data, we are dealing with qualitative forecasting – a topic for another day.
The data-gathering stage entails determining what data is required and already available. Furthermore, different patterns can be observed in the available data sets, and it is critical to identify them to select the appropriate forecasting model.
At this point, it is also critical to consider data quality and spend time cleaning up known outliers. An outlier is any data point that falls outside of the data's expected range. Ignore outliers at your peril; they will significantly reduce the accuracy of your forecast. As any stockbroker will tell you, history does not always accurately predict the future. So, keep an eye out for unusually low or high numbers, missing data, and trends you know will not repeat.
3) Select the forecasting model or combination of methods.
The forecaster must decide which forecasting method(s) or model(s) to use in this step. There are numerous forecasting methods, both qualitative and quantitative. In the quantitative section, these are typically classified as Time-Series of Casual Demand Drivers, whereas in the qualitative section, techniques such as Nominal Group and Delphi methods are available.
4) Build and test the forecasting model
In this step, the forecaster constructs a forecasting model using a subset of the available data. A statistical or mathematical formula is a model. Next, the forecaster puts the model to the test using the remaining data. That is, the forecaster will apply the formula and determine whether or not it produces an accurate result. If not, the forecaster will make the necessary changes to the formula until the forecaster achieves the desired results.
5) Refine for Special Events
An infinite number of potential events could disrupt a forecast pattern, especially as the granularity of the forecast interval decreases, for example, to a 15-minute interval. Therefore, picking your battles is critical, as is allowing for obvious changes in customer demand behavior. For example, a major event, such as the Super Bowl or the World Cup final, may result in lower customer demand during this time period. Still, more subtle special events, such as the changing of the clocks or student term time schedules, may also occur.
6) Compare events with the forecasts
At times, getting an accurate forecast can be difficult, especially when some people believe you have a crystal globe - so for those frustrated forecasters out there, the first rule of forecasting is that all forecasts are either wrong or lucky (it's impossible to be 100% correct 100% of the time) - forecasting is like trying to drive a car blindfolded and following directions given by someone looking out the back window. However, failing to learn when your forecasting is incorrect or lucky decreases the likelihood that forecasting accuracy will improve over time. After all, you can't manage what you don't measure, so the first and most important purpose of accuracy analysis is to learn from your mistakes.