More than a decade ago, when I worked for a supply chain planning software vendor, I was entrenched in demand forecasting and struggled to explain to customers that although the latest software package they purchased was very good, it still couldn’t predict the “perfect” forecast.
In those days, demand forecasting essentially examined past sales and used statistical forecasts to predict future demand. Companies already knew that a history-based forecast was inherently constrained by past promotions, and didn’t provide sufficient information for new product introductions. However, there weren’t many alternatives.
The best that companies could do to mitigate the forecast error was to rely on demand planners with decades of experience, who basically relied on gut instinct to cleanse historical demand from promotions and smooth the forecast from the most obvious spikes.
Today – as I pen my next report to investigate the SCM World Matrix’s demand-sensing quadrant – I find myself deeply entrenched in demand forecasting once again. And realizing that nothing has really changed!
Forecast Error Stuck at 50 Percent
Many supply chain executives were not shy in admitting that 50%, 80% or even 100% forecast error is the norm for many of their most difficult to forecast products. E2Open’s 2016 Forecasting Benchmark Study confirms that, over the past five years, the average forecast error is stuck at 50%.
It is clear that traditional forecasting processes based on past sales analysis and gut feeling have reached their limits. Clearly, companies haven’t found the magic sauce yet. However, many companies that have struggled with high forecast error rates have spent the last 10 years testing and trying a new path towards the next generation forecasting. The most interesting route points to an ability to harness the power of big demand data.
Big Demand Data
With the cost of data storage becoming less expensive, there’s plenty of supply chain data available – from point-of-sale (POS) data to social media – that can inform demand planning processes better, reflecting current market realities instead of historical estimates. Among the companies leading the way toward big demand data:
- Campbell – One of the most promising ways to reduce forecast error on the short-term is demand sensing that leverages data available along the supply chain, including POS scans, customer forecast and shipments. Campbell invested in Terra Technology’s Demand Sensing (now part of e2Open) to produce a daily statistical forecast at the SKU/location level, which is recalculated every week. The company measured a 45% reduction in forecast error at a weekly SKU/location level, and benefited from reduced customer service issues, stabilized inventory and improved manufacturing scheduling.
- David’s Bridal – Wedding dress retailer David’s Bridal is harnessing social media to improve demand sensing and forecasting today. Using First Insight’s product ranking and recommendation solution, David’s Bridal planners can test product concepts in a virtual environment. Based on consumer reactions and commentary to the virtual product offering, product designs can be adjusted, price elasticity can be gauged and unpopular models can be identified before a single raw material has been purchased. Jeff Warzel, Senior Vice President of Supply Chain at David’s Bridal, estimates that forecast accuracy has improved by about 20%, while product testing times have dropped from eight months to just two weeks.
- Merck – Forecast errors not only impact customer fulfilment, but also impact upstream on inventory, manufacturing and sourcing. Germany-based Merck KGaA (unrelated to U.S. based Merck and Co.) is in a multi-year project to improve demand forecasting and plans to deploy sensors to generate real-time data throughout its supply chain and algorithms to analyze it continuously. The system would rely on machine learning to predict spikes and lulls in demand for products and suggest ways to re-route raw materials accordingly. Using dashboards built on software from Aera Technology (formerly FusionOps), Merck plans to display real-time measures of supply chain performance.
Demand planning is the first, utmost critical step in running a highly efficient supply chain. Digitalization offers new opportunities to improve the supply chain by harnessing the power of big demand data. When done correctly, demand forecasting inaccuracies might be a thing of the past.