Artificial intelligence (AI) or subfields like machine learning (ML) are mentioned in just about every technology-related article these days. More than tech publications are covering the topic. Just last week, major media outlets picked up a story about Facebook’s plans to use ML to fact check videos and photos posted by users on its social media platform.
_______________________________________________________________________
The difference between Artificial Intelligence and Machine Learning
According to Merriam-Webster, Artificial Intelligence (AI) is “a branch of computer science dealing with the simulation of intelligent behavior in computers.” In other words, AI is focused on having computers emulate human intelligence.Machine Learning (ML) is a subset of Artificial Intelligence in which computers are trained to make decisions by data rather than by program.
_______________________________________________________________________
AI and ML have become part of our everyday lives. These technologies reroute our daily commute due to traffic or an accident ahead. They clear inbound spam email from our inboxes. They recognize the faces of our friends and family when taking pictures with a smart phone. They recommend products based on our browsing and purchase history. And these examples only scratch the surface of how AI and ML are being used today.
It should not be a surprise that enterprise software vendors are incorporating AI and ML into their products. If you have attended a workforce management (WFM) vendor’s user conference recently, the vendor probably discussed plans to incorporate AI or ML into their platforms, if they have not done so already. There are many applications for this technology in WFM. One of the first to market – and one that has the potential to benefit WFM users immediately – is using ML for forecasting.
Forecasting business activity, often referred to as driver or volume forecasting in a WFM system, is due for innovation. This part of WFM has remained largely unchanged since it was introduced to them more than 30 years ago. The typical forecasting process starts by selecting and configuring a forecasting algorithm. This is often done when the WFM system is initially setup. Next, historic data is loaded into the WFM system. The forecasting algorithm determines how much data is required, with most commonly used algorithms requiring between eight weeks and two years of data to work effectively. Data is typically loaded weekly so that the most current historic information is available. Finally, the system runs the data through the algorithm to produce the forecast. And, generally, the results are okay for most users.
Still, the traditional forecasting process has its problems. Let me outline the three biggest. First, the algorithms in most WFM systems are linear. They expect data to be trending up or down. If the source data is non-linear or “noisy” (e.g. random up-and-down movement in the data), these algorithms produce poor results. Second, these algorithms typically trade off accuracy for granularity. In other words, forecasting a week or day’s sales for the enterprise will produce an accurate result, but accuracy starts to suffer when forecasting for a day or within the day. Third, statistical expertise is required to over the previous two challenges to improve accuracy. While many organizations have given someone this responsibility, that person often lacks the analytical and mathematical skill required to improve the results, and often inadvertently makes them worse.
ML overcomes these problems by flipping the old forecasting process on its head. Rather than start with the forecasting algorithm, ML starts with training data. This data represents at least two business cycles of historic data. For this discussion, let’s assume a business cycle is a year. The training data is split into two sets. One set – called the “input” – contains the first year of data. The second set – called the “output” – contains the most recent year of data. In this context, ML views the output as a forecast and analyzes the two sets of data to learn (hence, the name “machine learning”) about the relationships between them. Based on this knowledge, ML determines the best algorithm and/or parameters for creating a forecast like the output when it is fed a new set of data.
While ML forecasting is not available in WFM systems today, it is coming, and companies have been experimenting with ML outside of their WFM systems. These companies have seen a 15 to 25 percent improvement with ML compared with the traditional forecasting process. Such improvements can have a real and material impact by better aligning staff with demand, eliminating waste and improving efficiencies.
