Correlating and Aggregating Demand Drivers
Carefully tuned forecasts have had their Achilles heel exposed during the pandemic. Not that we need to reinvent forecasting, rather there's been a rebalancing towards taking account of regional and external drivers. We held a Virtual Boardroom on 16th November to look at how manufacturers can get closer to consumer behaviour and patterns. It was a common pain point that specialist data providers only provide you with a top line number, and retailers' forecasts are wildly inaccurate, and in any case, too short term. How do you get closer to market data, to pinpoint which driver categories are heavily influencing the demand you're seeing? With thanks to experts at PredictHQ who joined the call, here are some of they key points raised, and the common challenges!
Planners are doing manual work that in future is not expected to be done manually
SAP APO support ending in 2025 will see some looking for new tools, not just SAP IBP
PredictHQ provides Demand Intelligence - providing global data, both historically and in the future, around events, and customers across CPG, retail, accommodation use demand services to exploit data to better understand what's happening around them, whether it's very specific locations, like a retail store, or regionally, at country level.
Participants in our discussion all forecast bottom up.
PredictHQ’s Beam engine will aggregate events into categories - eg events or calendar observances, so that you’re not trying to pinpoint numbers of individual data points. Decomposition allows for extraction of a baseline or residual forecast.
Since the pandemic many organisation have had to reweight forecasting away from established models and more towards being driven by external factors.
Most not conducting any sophisticated analysis of past demand to identify causation for spikes
Short shelf life products - you’re on a knife edge with forecasting
Big onus on manufacturers to ‘clean the history’ - otherwise mistakes are repeated.
As a manufacturer some use Nielsen data to try to correlate that back via the customer to understand how would they be servicing that market. How do you integrate that data?
Customer forecasts are 10x or more inaccurate that inhouse.
Retailer’s forecasts are too short term to be usable.
Focus on a small number of SKUs and conduct an analysis to identify which categories of driver are influencing demand. Machine learning can give you answers and ultimately provider greater forecast accuracy.