Pricing strategies: optimizations using predictive models

Optimizing pricing helps companies to achieve their business objectives. Different parameters are taken into account when defining the prices of products to be sold: understanding customer behavior, logistics costs, seasonality, competitor behavior, etc. Therefore, the analysis of all these dimensions requires the use of tools to help understand the different dynamics that can influence the prices of the products sold.

Machine learning helps companies to implement their pricing policy by exploiting different types of data and by maximizing the potential of the companies' historical data. What results can artificial intelligence achieve? What data must be provided to generate high quality predictions? What are the benefits of using predictive models? We take a look at the solutions that data science can provide to optimize the pricing of products offered by retailers, e-tailers and mass marketers.

Pricing strategy and business objectives

Projects based on machine learning require the definition of a business objective guiding the actions to be undertaken. The business use case allows defining a set of variables to be analyzed and optimized. These target variables will guide the training of machine learning algorithms. The commercial objectives vary from the need to optimize the company's margins, to increase sales, to optimize product assortments, to improve promotional or sales forecasts, etc. Verteego's role after defining the use case business is to recommend the scenarios to be implemented in order to guarantee the success of the project.

The project starts with the launch of a POC (proof of concept) targeting a precise set of points of sale in order to test the results before the industrialization phase. The validation of this first phase is based on the analysis of the performances induced following the optimization of the target variables thanks to the predictive models obtained. These variables can be: logistics costs, duration of promotions, types of promotions, product assortments, etc.

Specificities to take into account when defining prices

In the case of pricing optimization, companies need to take into account the specificities of points of sale and items simultaneously to improve the pricing of their products. Indeed, predictions generated at the human level cannot encompass the correlations between products and point of sale specificities (sales force, geolocation, etc.). Moreover, when data becomes voluminous, trend and correlation analyses become difficult to highlight. As a result, different data cross-referencing must be considered in order to understand the factors that require a product's price to evolve, the promotions to be granted, the duration of promotional sales to be dedicated to it and the communication channels to be invested in to sell it better.

The data needed for machine learning technologies

Machine learning algorithms go through a training phase where they take different variables and data files as input. In the context of setting up pricing choices, and to ensure successful machine learning, different data sources must be exploited beyond the historical data of the company concerned.

Descriptive data of the product

The company's sales history

The sales history includes both transactions made at the checkout, promotions used by the company and orders placed for this product. The transactions related to the analyzed product are indicated in the customers' sales receipts. Each receipt matches a customer's shopping cart and thus provides information on the recurrent assortments of products purchased together. The role of predictive models is to highlight such patterns.

Product-related promotions can be expressed through data such as past discount rates, promotional mix and advertising budget. Finally, information on the sales channels used will enrich the picture drawn up for the item studied: drive, click and collect, online ordering, delivery, sales in physical stores, self-service sales, etc.

Data related to the supply chain

Information recovered from all supply chain actors: raw material prices, supplier delivery times, previous supplier delays experienced, transport and storage costs, etc., provide insight on the availability of the product over time, the factors that can affect its stock shortage for example, its speed of flow, etc.

The characteristics of the point of sale

The required exogenous data

Data from sources external to the company and its supply chain are combined with the data described above. This can include information about the climate and weather in the regions where the retailer's customers live, strikes, holidays, festivals, average customer salaries, etc. Data describing the target market and the competition is also desirable.

The importance of data in machine learning

The choice of data allows to optimize the quality of predictions generated by predictive algorithms. Also, their quality can have an impact on the return on investment of the project. The data extracted from the company's data lake will require upstream processing in order to fill in incomplete information, eliminate duplicates, and segment and structure the data collected.

The choice of data is crucial to improve the return on investment of the project. This can be in the form of increased margins and sales, but also indirect gains such as the avoidance of overstocking situations and unnecessary capital tied up in the form of stocked goods. Finally, the automation of analysis and prediction tasks frees operational and marketing teams from time-consuming tasks and allows them to focus on high value-added aspects of their business.

Analysis of predictions generated by Verteego

Once the learning has been completed, the results are deployed within data visualization tools. These dashboards summarize the information needed to understand the scenarios predicted by the machine learning models. Each scenario can be analyzed in detail. The user can understand variables such as:

• proportions of products sold by color, size ;
• quantities of products sold per day, week or month;
• sales generated by time and product;
• results by type of markdown or promotion.