Assortment forecasting: the necessary tools for optimization
The assortment forecasting must take into account all the operations implemented to anticipate and select the offer of the products proposed to the buyers in physical or electronic point of sale. The main objectives to be reached when implementing a product assortment forecasting policy are the improvement of sales and the optimization of the stock disposal of the products targeted by the assortment.
In addition, the optimization of product assortment forecasting can improve other aspects of the operation of a given point of sale. Once established, it affects both the forecasting of orders, promotional policies and the operation of forecasting teams. Which dimensions should be taken into account to improve product assortment forecasts? What tools can help improve the quality of assortment predictions by store? Find here all the aspects to take into account in order to succeed in assortment forecasting.
Forecasting of inventory, promotions and product assortments
Inventory forecasting and product assortments are linked. The objective behind the optimization of these two actions is to reduce storage costs and to limit the immobilization of cash in reserve. In addition, optimizing inventory forecasting and product assortments is also:
- reduce losses related to unsold products;
- limit the waste of perishable products (fresh, food for example);
- avoid cannibalization effects between products;
- reduce inventory costs;
- free up operational teams from time-consuming or non-value added tasks.
In addition, assortment forecasts are also linked to the promotion policy implemented by the sales outlet. This policy is itself linked to the optimization of margins and customer needs. Therefore, it is necessary to choose a working methodology and a decision support tool able to make the analysis of the problem of the optimization of the assortments of articles linked to the other operational and logistic aspects of the sales outlet.
The importance of data in assortment forecasting
To optimize assortment forecasting, the historical data of the retailer or business in question must be valorized. This data include customer baskets (invoices and tickets), sales and promotions history, etc. The choice of tools capable of taking into account all the data is crucial for improving the predictions generated. The forecasting and BI (business intelligence) methods used by companies until now suffer from shortcomings regarding:
- Lack of personalization based on points of sale;
- the inability to manage large volumes of data to gain accuracy;
- The implication of human bias due to the limited choice of data;
- The impossibility of combining internal company data with data from competitors, the target market, exogenous data such as meteorological and econometric data, and data related to the supply chain (data from suppliers, goods transportation, etc.);
- the inability to produce anticipatory results over several months or in real time.
The performance and quality of predictions depend directly on the data used to generate them. Therefore, in order to gain in accuracy, the cross-referencing of different data sources that can enrich the analysis elaborated by predictive tools is becoming an unavoidable imperative. Currently, data science tools meet the challenge by their ability to ingest large volumes of data during their learning process. Predictive machine learning models take into account data extracted from companies' data lakes, but also from exogenous sources (competing companies, climatic, regional, seasonal, econometric data, etc.). Finally, machine learning and forecasting algorithms ensure the production of predictive lines in real time, thus allowing to gain both temporal and physical granularity (for each point of sale).
Customization of the forecasts according to the points of sale
The optimization of assortment forecasts cannot take place without taking into account the specificities of each point of sale. Indeed, each store or virtual sales unit (website, mobile application, marketplace, etc.) has its own characteristics. Predictions based on manual tools or macroscopic data limit the quality and reliability of the predictions generated.
Mapping the retailer's points of sale (POS) must be followed by a rigorous analysis of each store. This second step will serve to establish a profile or identity card of the analyzed sales unit. To do this, the data to be recorded must answer a few questions:
- What items or products are sold within the analyzed POS?
- How long have these SKUs been stored and sold?
- What are the characteristics of the customer base for this POS?
- What is the lifespan of this customer base?
- When do they visit the physical or virtual store?
- How often do they shop?
- What are the purchasing channels offered and preferred by customers (click and collect, drive, self-service, web, mobile, etc.)?
- What are the motivations of shoppers within the POS in question?
The portrait drawn up for each sales unit should make it possible to highlight the frequency of purchases, the seasonality of purchases and customer behavior. In addition, exogenous data taking into account regional and local specificities (festivals, holidays, strikes, etc.) must be cross-referenced with the profile thus obtained in order to better understand purchasing motivations and better anticipate the product assortments to be offered.
Genericity and customization of predictions and machine learning tools
Although capable of addressing the above issues, machine learning-based predictive tools ensure the generation of predictions that are certainly customizable and tailored to all outlets in a given company. However, machine learning algorithms have the ability to be used in different production and sales domains. Thus, the following can benefit from it:
- actors of the mass distribution;
- pharmaceutical industry, automotive industry, etc.
To conclude, the optimization of product and item assortments in points of sale requires the improvement of forecasting policies for stocks, promotions and merchandising. This optimization is therefore carried out at different levels: marketing, sales and logistics. It induces a complexity of processing requiring the use of tools capable of handling these different dimensions. The advent and maturation of predictive tools from the field of artificial intelligence has made it possible to respond to all these constraints. Thus, the optimization of a single aspect of the point of sale operation benefits all of its operational levels.
We recommend these other pages: