Forecasting gets a bad rap in some circles.
"We have been forecasting for ages and the only thing we have learned is that forecasts are wrong," to paraphrase the complaint of a Fortune 500 CFO. The youthful optimism inherent in early-stage forecasting projects leads planners to assume accuracy will forever improve. But the reality of a mature planning process is just the opposite. Accuracy plateaus; things can only be "dialed-in" so much. And so we often see the early excitement of project sponsors give way to a hint of disillusionment.
Machine learning is changing all of that.
In traditional demand planning, even the most well considered process relies on programmed instructions that cannot fully cover the numerous complex variables and relationships that will eventually affect demand. It's a sad story, but true. We humans are the limiting factor.
Since the days when John Galt began work on its ProCast™ Expert selection technology in the mid-1990s, machine learning has become quite the buzzword. ProCast grew into a robust technology that tests outcomes and selects a best fit model from a wide selection of forecasting techniques. Overnight the task of arriving at top-tier forecasting accuracy became a much less tedious affair. Individual planners could deliver better results in less time across a wider range of industries.
While that change was a dramatic improvement, ProCast is just the beginning. Recent years have shown exponential advancements in machine learning. All the buzzwords, including AI, neural networks, big data, etc, are coming to demand planning. These technologies will prove especially helpful in complex scenarios. They leverage our knowledge, experience, and skills in highly efficient and effective ways across a broad range of data, especially in the following five areas.
1. Trade promotions and media events
Have you struggled to predict promotional "lift"? Promotions, advertising, and other “demand shaping” are expensive. However, evaluating their impact is challenging. There are a large number of variables with complex interactions to consider. Having vast experience is generally not enough to understand correlations among variables.
Machine learning is being applied to this problem. Multi-dimensional modeling that considers both qualitative and quantitative variables is particularly well suited to describe and predict the non-linear demand driven by promotional activity.
2. New Product Introductions
Have you faced the challenge to forecast demand for a product without a sales history? Machine learning is suited to this area as well. Models can include early indicators such as web analytics, product attributes and even social media data to predict the performance of a product launch.
3. Social Listening
How do you capture social listening so you can correlate social sentiment with demand signals? Traditional demand planning relies mostly on transactional data, creating latency between customer needs and supplier reactions, but machine learning takes into account social listening by providing enhanced forecasting models.
4. Extreme or complex seasonality
How are you managing seasonality? Traditional demand planning can consider seasonality, but sometimes seasonality becomes so extreme or complex that it is not well suited to normal regression analysis-based techniques. Machine learning helps you to identify “clusters” with similar seasonality profiles and track seasonality patterns and trends through the SKU-locations.
5. Weather Data
And while we're on seasonality, how about the weather? We need to account for geographic areas, products, and demand lags. Certainly, it’s complex to model statistically with so many variables. Nonetheless, machine learning will do the work for you.
What's Next
While forecasts may always be "wrong" in an absolute sense, the steady growth of machine learning techniques in our supply chain planning tools will deliver higher plateaus of performance in all areas of demand planning. With them your planning team will deliver increasingly relevant and useful information across the organization, ultimately leading to better efficiency and higher profits. That's something to stay excited about.
