top of page
Search

SKU-Level Demand Forecasting During COVID-19



"A prudent man foresees evil and hides himself, But the simple pass on and are punished." - Benjamin Franklin


This proverb is more apt in today’s Covid-19 impacted businesses. As of now the whole world has come to a standstill but like all storms-this time too will pass. What will come after is a lot of inventory which might not move quickly and a lot more demand for which there will simply be no supply. How should the businesses plan their way if they wish to survive the post catastrophe competition and the need to stay cost effective?


Forecasting demand at a SKU level in times of a Pandemic like Covid19 is not an easy deal. Generally, the retail supply chain works on a variety of forecasting methods – mostly on simple excel based moving average and trend analyses and there are some sophisticated models utilizing the power of Machine learning and AI. Even the best forecasting models cannot predict the impact of a catastrophic event like COVID-19 because all the previous trends and forecasting patterns go for a toss because of external and unexpected shock.


At this point however it is possible to assess the impact of a shock like COVID-19 over the next few months. It is important to know that even though no one could have predicted an event like COVID-19 but what is important for the businesses is to be able to forecast what will happen in the future and enable them to take control of the future demand surge or the plateaued demand that they may see for short/longer terms. This will allow the businesses to meet up the pent up demand once the market open up again and the situation returns to normalcy and at the same time allow to stop the hoarding mentality that can lead to overpriced inventory ordering to meet the expected demand surge which may not come within the expected time.


In data science we call it impulse analysis. Here, we have a sophisticated machine learning platform called SKUCaster which can incorporate a shock on demand & supply over the expected sales in the coming months. The platform can incorporate the impulse due to the temporary halt in the supply and demand cycles. Impulse analytics looks to make up the difference between what was expected and what the actual demand is and then tries to make up for the pent up demand that will inevitably come back when normalcy returns. However this is not to say that the future demand is simply going to be the difference between the projected demand and the actual demand ruled over the future periods it is much more complicated than that.


In statistical terms the impulse response is the derivative with respect to the shocks. SKUCaster product uses the power of machine learning and AI models and employs methods like vector autoregressive models, LSTM models and neural network models. The main purpose is to describe the distribution of the explanatory variables which ultimately impact the demand in reaction to a shock in one or more variables. In this case it will be literally be all the variables which are related to the end customer demand and the incoming supply variables. What we’re essentially looking at is to be able to trace the shock and it’s transmission overtime. Even in this time of unpredictable demand the businesses still have to plan about the future and protect their customer base so that they don’t have to face the lost sales due to miscalculated inventory or carry huge inventory costs because they did not assess the impact of the shock and expected the demand to return to normalcy faster than it did.


Please note that the impact of Covid-19 is not the same in all countries. While the US hasn’t yet reached the peak; China has already started to open up. That means that the supply lines are opening up. Even in the normal circumstances, it takes 3-6 months from order to supply from China to Canada or USA. So if we plan ahead then we should have already ordered and planned on capturing the market from weeks back. Its still not late and businesses must order now to meet the pent up demand surge by Q4.


Just plain inventory management in the warehouses is one thing but we also need to plan and book trucking companies, manpower etc. so that we can supply the goods to the end customer. That means we forecast now and order as of yesterday if we do not want to survive now but go out of business when the things are returning to normal.


SKUCaster can actually forecast the demand by accounting for the blip in demand now and the expected surge in the demand later that too within a reasonable confidence interval.


Machine learning and AI models allow SKUCaster to look for both the Generalized impulse responses as well as the temporal impulse responses. As we can see there is also the need to separate out the impulse by SKU categories and geographies as the demand for necessary parts and items will be more compared to optional parts and items. SKUCaster first segments the SKUs into different groups and one of the most important variables as of now is the sensitivity of the demand to the shock which can be measured and given a weightage in the segmentation. This means that the SKUCaster will give a higher forecast to a highly sensitive and necessary SKU compared to the SKUs which will move slowly.


The next challenge is to include the seasonality into the picture because COvid-19 or not; people will not buy winter tires in summer. The seasonality as well as the impulse analysis makes the forecasts of the SKUCaster much closer to reality and above all the product allows for manual inputs from the business owners to overwrite the forecasts based on specific inputs for each SKU or fineline.

3 views0 comments

Comentarios


bottom of page