Agile forecasting

My previous article explained why the conventional monthly forecasts are sub-optimal for executing day-to-day supply functions, as they fail to recognize various recent changes in demand patterns. As a result, operating supply chains on the basis of the S&OP and IBP forecasts often leads to the twin problems of stockouts as well as excess inventory.

Companies working on Replenishment Supply Chains have circumvented this problem by delinking their daily supply actions from the monthly forecasts and instead using actual demand signals for replenishing stocks. However, an important parameter setting in such systems is the inventory buffer at each node, which is set as the maximum expected demand during the replenishment lead time. In absence of near-term demand forecasting at a granular level, companies resort to setting inventory buffers based on current level of actual demand (Average Daily Usage) and a variability factor to account for expected demand fluctuations.

The good news is that we don’t have to live with this limitation now. Recent advances in Artificial Intelligence and Deep Learning algorithms have made it possible to forecast demand in the near-term horizon at the most granular level for which regular data is captured. Their accuracy is amazingly good. Moreover, these forecasts can be refreshed everyday to capture the most recent changes in demand patterns.

These near-term demand forecasts for the next five to ten days at a granular level (sku-store-day) are, therefore, more accurate in refreshing inventory buffers on a dynamic basis. These buffers can be reset everyday as new forecast values become available.

Granular level daily forecasts are particularly useful in the following three cases.

1. Seasonal products

These products witness a sharp uptick in demand when the season sets in and a sharp decline at the end of the season. However, the timing of season setting in and season ending vary significantly across geographies and are difficult to predict. Granular forecasts updated on daily basis help in identifying these local seasonal inflexion points and work out much better than the conventional forecasts for the entire duration of the season.

2. Promotions

Brands often engage in consumer or trade promotions to spruce up their sales. Inventory buffers are reset in such cases to account for the expected uptick in sales as finalized in the S&OP discussions. However, the actual response to these promotions varies quite significantly across various geographies, leading to the twin problems of stockouts in some areas and excess stocks in others. Granular forecasts are quick to catch the actual response in each territory and make dynamic adjustments to near-term forecasts, thereby making the replenishment algorithm more accurate in correctly allocating the promotional stocks and avoiding stockouts as well as excess stocks.

3. Sales skew

Many companies witness a sales skew towards the end of the month. Although efforts to smoothen the skew are taken, we do nevertheless have to live with this pattern in most cases. As a result, inventory buffers get stressed towards the month-end and are found in excess at the month-beginning. Granular forecasts identify this recurring pattern for each sku-geography and adjust the near-term forecasts accordingly.

The granular forecasts described above come in handy in synchronizing supply with demand on a dynamic basis. Having a more reliable view of near-term demand at the most granular level helps in improving customer service level as well as product freshness. This was the missing piece in dynamic synchronization of supply with demand, which can now be plugged in by CPG and Pharma companies.

Would you like to incorporate it in managing your supply chain?