One of the biggest challenges for physical and online retailers haa always been correctly predicting the demand for products and properly setting all the other business activities accordingly.
This every-day matter becomes much more important in some specific periods of the year: Black Friday is considered the starting date of the astonishing rise in sales that every year shows up when approaching the Christmas period. Until a few years ago, Black Friday (and also Cyber Monday) was mainly an American recurrence, but recently it started spreading out also in other part of the world at an extremely fast pace.
Untrustworthy forecasts - and subsequently wrong ordered quantities - can for sure jeopardise companies’ profitability, especially when coming to technological or complex products which are very frequently characterised by longer lead times hence making impossible an “in-progress” replenishment. This article by Michele Russo wants to show an integrated model that merge together sales forecasts and optimisation of the ordered quantities.
Forecasting the future
Perhaps forecasting sales is the toughest job for a company for two reasons: the predictive analytical models implied are usually quite complex and future events could easily deviate from what is forecasted. For both these reasons, it becomes crucial to keep in mind how to approach the forecasting phase; here follow three macro suggestions to ensure your predictions to be reliable.
Look at the past
In order to have a good final result, the starting point is always to look at past data and to try to figure out which are their main features. Each time series can be usually decomposed into a trend component, a seasonal component, and a cyclical one. More specifically for our problem, big emphasis should be put on trend and seasonality. Understanding the trend of a time series could be extremely useful in ascertaining whether the interest for a specific product is increasing or not: this could help in setting future marketing policies for the company. Secondly, it is also fundamental to spot seasonal patterns and to map those with the different holidays’ periods during the year; this is key to predict the moments in which the demand for a certain product could rise sharply.
Based on these points it is therefore possible to choose the best forecasting model for our sales, which will help in predicting the future figures. In the picture below it is possible to see how a forecasting tool can model the future behaviour of a time series, the green part shows the forecasted data points. Data implied are real data coming from a very popular product category of a big American retail chain.
Plan for uncertainty
Having a good punctual estimate is for sure a big achievement but more important is taking into account uncertainty. The best strategy when forecasting should be always considering which could ideally be the interval in which our forecasts could reasonably lay hence analyses could be enriched adding maximum or minimum values for the predicted data. This will be extremely helpful in conducing some “what-if” scenarios and make a more informative choice. In the previous picture it is possible to see what just explained by looking at the light blue areas that exactly account for uncertain part of the estimates.
Trust data (but not too much)
This very final point could be in contrast with what previously explain but eventually it is not. Data and forecasts are for sure an extremely helpful instrument but in most of the cases it is crucial to integrate the model with qualitative considerations. Time series predictive models could perfectly capture past behaviour, but this won’t ensure that the future will be exactly the same therefore make sure to supplement and adjust the quantitative analysis with a bunch of qualitative insights.
Optimising the strategy
Once sales forecasts are done, the next step is using these data to make choices that in most of the cases are cross functional.
What is the optimal quantity that we should buy, given these forecasts? Which markdown is the best? At what level should we advertise these discounts?
These are only few of the possible questions that a company should consider before moving to the actual implementation of the strategy. As is it possible to understand, analytics are extremely helpful in coordinating a great variety of activities which span from marketing and sales departments to procurement and accounting ones.
There are two main rules that should be followed when planning all these actions, details follow below.
Catch the overall picture
Analytics in this changeable setting are very powerful because can take into account multiple parameters and scenarios: values as unit price, variable costs, fixed costs, promotion costs, markdown levels and inventories should always be part of the optimisation model.
More specifically, companies should pay particular attention to two parameters in these types of problems: markdown levels and promotion costs, in the best scenario indeed sales would rise thanks to the introduction of discounts or targeted advertisement. It is therefore important how to estimate these parameters; lift in sales for previous years could be used to predict the future increase of revenues for each level of promotions, whereas advertisement costs could be easily taken from accounting documents.
Once all the setting is done, it is possible to build a model that shows the level of forecasted quantities and profits, an example of this could be found below.
Finally, the model will propose an optimal quantity to order optimising the profit function, which already includes all the varying parameters, in a “news vendor model” fashion.
Choose the best strategy
Before moving to the final choice, the company should decide the strategy to follow, its timing and intensity. For instance, by looking at the previous analysis it could make sense to not waste money in advertisement in the first weeks of the holiday season but only to further boost the effect of the markdowns. This part of the analysis is very company specific and should be carried out pay a lot of attention.
To conclude, all the companies – especially if active in the e-commerce and retail industry – should seriously think to follow this analytical framework to unleash the power of forecasting, always keeping in mind procedural rigor and managerial implications at the same time
Comments