In my last post, I discussed the importance of leveraging data science to develop the most accurate forecasts possible. However, an assumption many companies make when building forecasting models is that historical sales are an accurate estimate of future demand. There are two risks in doing this:
1) Historical sales are a function of both primary demand and substitution effects due to stockouts.
2) Even if you could estimate both accurately, future primary demand will likely differ from historical primary demand.
In this post I will discuss an approach to solve the first problem.
Manufacturers and retailers must consider substitution effects in addition to the tradeoffs between overstocking and understocking to determine proper stocking levels. While many papers in the inventory literature have considered substitution effects, many are very complex, highly theoretical, or have unrealistic data requirements. The result is that many of these approaches are difficult to execute in practice. A simpler, yet effective, method to make inventory decisions on substitutable products could be useful to retailers of staple categories such as personal care, grocery items, and office products.
MonarchFx uses a model with both realistic data requirements and intuitive results that decision makers can easily understand. For example, armed with only daily sales and daily out-of-stock records, we can derive product or SKU-level estimates for:
• True primary demand
• Substitute demand
• Recapture rate
• Total sales lost (primary demand – recapture rate)
• Implicit preference rates
When equipped with these demand estimates, model builders will have a more accurate profile of true customer demand, allowing them to create more accurate forecasting models. Thus, when building your next forecast just remember that a chain is only as strong as its weakest link. Do not let unknown substitution effects be your weakest link.
David Lengacher is head of Data Science for MonarchFx.