August 18, 2016
Thanks to her gravity-defying routines at this year’s Olympics in Rio, Simone Biles has become a household name, walking away with four gold and one bronze medals. Could her success have been predicted? Absolutely. Even though this is Simone Biles first Olympics, she has been dominating gymnastics since 2013. She took the World Championships in the all around category for three consecutive years.
Although past performance is a pretty good predictor for success of an athlete, the same can’t be said for predicting product demand.
A lot factors have to be considered: trend, seasonality, new product launches and marketing campaigns just to name a few. Legacy prediction algorithms, while mathematically sound (Auto ETS and Auto ARIMA, etc.) rely on data scientists to manually build models to fit historical sales. This process takes a long time and even the best can only build a handful of models every week. With traditional approaches it’s easy to miss important patterns in the data that can point to a more accurate forecast.
Advancements in training, nutrition and knowledge have taken a new generation of gymnasts, like Simone Biles, far beyond what was possible decades ago. The same can be said when it comes to forecast accuracy. At FusionOps, we’re using machine learning algorithms to automatically and simultaneously create and test thousands of models, and then apply the most accurate. This method enables us to find important patterns in the data missed by traditional techniques.
Why are we even talking about this?
Because forecast accuracy is the single most important metric for driving supply chain performance. Predicting the future inventory and optimal demand you need to carry amounts for some companies as the equivalent of winning a gold medal. To win at forecast accuracy gives you a huge leg up on competitors, but more importantly, on customer satisfaction too.
What does gold medal forecast accuracy look like?
Let’s start with the basics: best in class forecasters in the pharma industry achieve 84% accuracy, as measured by mean absolute percent error (MAPE) on a 3-month lag. Industry average is measured at 75%. But we see many companies far below that. In our experience 50% or even 30% is quite common. At FusionOps, we’re using cloud, Big Data and machine learning to help companies dramatically improve forecast accuracy. Here are some things that we do to help you improve your chances of gold medal forecast accuracy:
- Forecast at a granular, product-location level and aggregate as needed
- Automatically forecast across all SKUs, not just those with the most volume or sales
- Use machine learning-based algorithms which can generate and combine thousands of models to create the most accurate picture of demand and reduce forecast error
- The automation of as much of the forecasting process as possible, to ensure greater accuracy and reduce human error.
Getting all of this in one solution may seem like a tall order, but it’s possible with advanced supply chain intelligence from FusionOps. According to industry analyst firm, Gartner, businesses can expect numerous benefits, including:
- Reduce costs and working capital thanks to 15 percent less inventory on average
- Increase efficiency leading to 35 percent shorter cash-to-cash cycle times
- Improve market share and revenue as a result of 17 percent stronger order fulfillment
For more Olympics reading, be sure to check out our blog on Katie Ledecky and service levels. Also, take a look at our webcast on machine learning and forecast accuracy. As long as you have the right tools and know-how in place, you can transform your supply chain into one that is truly worthy of a gold medal.