Connexion Contact Us

Stratégie de pricing : Trouvez la meilleure technique d’optimisation des prix grâce au Machine Learning

by Emmanuel Kouratoras

#Soldes #automl #machine learning #Pricing

Find out how machine learning can help you determine the best pricing strategy for successful sales or promotions. Case study…

  1. Pricing strategy and Machine Learning
  2. How to develop a pricing strategy?
  3. What return on investment can be expected from the use of machine learning?
  4. Can we rely on the forecasting results obtained by the machine learning?

Retail companies have a lot of information about their customers and their consumption habits. Combined with machine learning, this data can become a source of useful predictions for the points of sale. Through a practical case study, we will see which data is essential to optimize prices and determine an effective pricing strategy.

Pricing strategy and Machine Learning

What is machine learning?

Machine learning is a technology that, thanks to algorithms, enables the processing of a large amount of data in order to make predictions. Based on statistics, machine learning algorithms can identify significant patterns in data. This processing is essential when dealing with a large volume of data. If an algorithm can process millions of data points, it can also cross-reference these data to identify trends.

Today's businesses collect a significant amount of information about their customers and their purchasing habits. While it is possible to extract some information from this data, it quickly becomes cumbersome to process it quickly and in its entirety. This is why more and more companies are integrating machine learning into their analysis tools. By working with retrained models, a machine will be able to integrate multiple factors (internal or external) and update its understanding of the data in a short period of time. In the retail sector, for example, machine learning can be used to quickly study consumer purchases, supply mechanisms, the effects of logistical constraints or the specificities of each point of sale.

Why use machine learning to define your pricing strategy?

The data collected to make the machine learning models work provides a detailed knowledge of its environment and can become a real strength for a company. This is only possible if they can be manipulated efficiently, which is not always easy for human beings. In addition to saving analysis time, machine learning makes it possible to understand how a customer acts with regard to a company's products. Knowing the habits of its customers will allow a point of sale, for example, to adjust its stocks, to enable it to find its product, or to offer promotions adapted to its basket.

In the retail sector, machine learning studies the information available on how customers behaved when looking at their receipts or shopping carts. By examining this data, the computer will be able to highlight patterns in consumer behavior. As a result, machine learning can help to understand which promotions generated the most margin or which promotion strategy was effective.

The same is true for inventory management. Using information from each point of sale in a store, the forecasting program will be able to determine where a product is selling best (based on region, competition, sales force, additional services, etc.) and thus adjust the product portfolio.

Finally, machine learning is a significant aid in price optimization and pricing strategy. All the prescriptive and predictive data available to the models make it possible to make quick and objective decisions. For example, a ready-to-wear retailer will be able to analyze the prices of its competitors to determine the price of its own products.

How to develop a pricing strategy?

In order to illustrate the steps involved in setting up a pricing strategy, we propose to follow the case of a major player in the ready-to-wear industry. This company called on Verteego to optimize its prices and thus determine the best sales scenarios.

This client structured its needs in 4 questions:

  • Which product(s) to promote?
  • At what price(s)?
  • What is the interval between each markdown?
  • What quantities should be allocated to the local stock and to the remote stock in the warehouses?

Determine the required data

In order to provide our client with better visibility on its future sales, Verteego's objective was to generate the most accurate forecasts possible based on the historical data our client entrusted us with. These were supplemented by external information, which is more difficult to master without machine learning technology. Verteego integrated all this data and then generated predictions that could be directly used in the platform's interfaces or in tools already used by our clients, such as their ordering systems, marketing software or Business Intelligence (BI) tools.

With regard to the internal data required for machine learning, Verteego takes into account information concerning :

  • the product (its category, brand, different types of packaging, ...).
  • the points of sale concerned by the promotional operation (their location, size, assortment, local events surrounding them, ...).
  • the price (the history of sales prices applied in previous operations, promotional reductions applied, price changes over time, promotional mechanisms, ...)
  • the sales force present at the point of sale (its level of remuneration, its level of qualification and the number of people present according to the schedule)
  • the different channels on which the promotions are applied (physical point of sale, web, drive or delivery)
  • Advertising (advertising budget, merchandising mode, promotion on social networks or in catalogs)

For the external data required for machine learning, information about :

  • the annual, monthly or weekly seasonality, which will help to better determine the periods of the month that can cause the curves to vary greatly: events, exceptional events, pay days, school vacation periods, etc.
  • the weather, which can also play a role when it comes to very short-term forecasts (< 3 days), particularly in ready-to-wear, specialized distribution or shopping centers
  • economic factors such as the exchange rate, the average wage in different regions, the inflation rate, etc.
  • information on the competition around a point of sale (its nature, its density, ...)

It should be noted that, since the COVID pandemic, other health factors may be taken into account in these external data.

Define the DNA of your point of sale

Each point of sale has what we at Verteego have called the "sales genome". It is in a way the unique DNA of each point of sale. This information can already be found in the data used daily by each point of sale. Verteego will simply identify them and extract their correlations to use them in its forecasts. The "sales genome" thus gathers all the information necessary for the algorithm to highlight the patterns we mentioned at the beginning of this article.

With the whole of this data chain, Verteego is able to determine who will buy a product, which product will be purchased and in what quantity. It also includes cannibalization effects (when one product is sold instead of another) or missed sales opportunities against competing products. By analyzing this data, the "sales genome" makes it possible to predict, for the next promotional operations, the purchasing practices of customers, which channel they will prefer and what are the motivations that triggered their trip to the store.

Once this analysis is done, internal and external data is calculated, integrated or connected from a cloud database and made available on the Verteego platform. Then begins the machine learning phase, during which Verteego trains the machine learning models, which will then be queried to generate predictions.

Choose a scenario to meet your business objective

The implementation of machine learning technologies requires a definition of the main business objectives. Most often, we can find objectives for optimizing margins, units sold and turnover. These objectives can be chosen either separately or in combination with a predefined weighting. Verteego will then be able to recommend scenarios capable of responding precisely to each of the previously defined objectives.

Each scenario is made up of variables that can be acted upon and that must be defined as a priority. For example, one can choose the communication budget, the markdown percentage, the duration of the promotion, etc. For each combination, Verteego can generate predictions, from which Verteego will recommend the best scenario that meets the set objective.

In the example below, if the objective is to optimize the margin, scenario number 5 will be chosen. This scenario will generate a margin of €22,000. Conversely, if the objective is to optimize units, scenario 4 will be preferable.

Analyze its results

Once the data has been processed by Verteego, the results can be distributed to the points of sale or studied via data visualization tools.

Data generated by Verteego can be automatically retrieved and transferred to data visualization tools such as the Verteego platform or, in our illustration, Google DataStudio. These tools will allow the formatting of information about products, stores, promotions, etc. These results can also be filtered to study the specifics of each scenario chosen. In our example of selected scenarios, we can see that the coats will sell very well during this sales phase.

The forecasts generated by Verteego also allow a detailed analysis of sales in terms of quantity, turnover or margin, on a daily basis. Our client was therefore able to determine the impact of sales on its sales by accurately measuring the success of markdowns according to identified objectives.

In our example, we can see, in the 3rd markdown, that a -70% promotion on the Stock Keeping Unit (SKU) will degrade the seller's margin more strongly than the 50% margin. However, this trend is not always verified. We can clearly notice it in our case, since the scenario that generates the most margin does not concern waves of reductions between 10 and 40%.

What return on investment can be expected from the use of machine learning?

The ROI of a machine learning project depends largely on the quality of the data collected. The more complete, informed, segmented and ordered the data, the easier it will be for the artificial intelligence to train its models. As we have seen, the higher the quality of the data sets, the more complete the scenarios will be and the clearer the results will be to analyze.

In the case of our ready-to-wear customer, the margin was increased by more than 32% over the previous year. In terms of sales, our client saw an increase of more than 7% during this sales period, while decreasing its excess inventory by 18%.

In addition to optimizing margins, improving sales and reducing fixed assets, our client significantly accelerated the productivity of its teams through the automated decision making provided by our technology. Finally, our forecasts have enabled our client to optimize its linear meter and reduce its inventory costs.

Can we rely on the forecasting results obtained by the machine learning?

Machine learning is a science that can be difficult to approach and often opaque for companies. During a forecast, the relevance of a variable can be questioned and thus influence the final result.

In order to determine the importance of each variable, Verteego provides a platform that allows the user to discover the specificities of each product. As a result, the user of the solution will be able to better predict whether a product is more sensitive to the size of the point of sale, the price or the packaging mode. All this information, used by Verteego, has an influence on the choice of the promotion scenario and can give a more global view on how products behave within an offer.

For example, we can study two types of products, each with a particular sensitivity to selling price and markdown. On the left, the product is only slightly sensitive to promotions compared to the product on the right.

Thus, price and markdown are in the minority in relation to "feature importances", unlike the right-hand product, which has a high sensitivity to price. It can therefore be deduced that a promotion on the left-hand product will have little impact on the volume of sales, whereas the right-hand product will see its volume strongly reduced with the implementation of a markdown.

Determining and optimizing your pricing strategy is a complicated task that requires increased customization at each point of sale to be effective. Thanks to its calculation algorithms, machine learning remains the most reliable and fastest tool for accurately determining the price of products.

Do you want to optimize your pricing strategy? Ask for a demo of Verteego platform now.

Be informed of the latest news