Machine learning in inventory forecasting and recommendation systems: Dodo Pizza’s case study 12.03.2019
Machine learning in inventory forecasting and recommendation systems: Dodo Pizza’s case study

Due to a data-driven approach, retailers no longer have to risk, relying on expert analyses and previous experience in their decisions. A data-based management strategy allows to examine consumers’ behavior, realize their needs and preferences as precisely as possible. But what if we use an unconventional approach to processing this information? Will such experiments be justified?

Andrey Filipyev, Senior Developer at Dodo Pizza and speaker at AI Conference, has proven that the development of new areas is profitable and efficient, even when their benefits are not obvious at once. The article reveals how Dodo Pizza has integrated machine learning and what it has resulted in.

Concept of recommendation systems

The recommendation system allows retailers to increase sales and create trends in certain products. They refine user data left on the website (date of birth, order history, device used to enter the website, etc.) andadapt to customers’ preferences.

Based on algorithms passed by these data, one can build associative relations between filtered information and website content. Thus, recommendation mechanisms determine a similarity coefficient in clients’ preferences and offer corresponding goods and services.

Dodo Pizza’s advancement of recommendation system using machine learning

Dodo Pizza has more than 230 pizzerias. It means that a single recommendation system is not enough at such a level. To make customer data processing more accurate, Andrey Filipyev offered to integrate machine learning. The company applied artificial intelligence to solve two issues.

Case study 1. Clusterization of recipes for inventory optimization

The clusterization of recipes allowed Andrey to define six sets of ingredients chosen by clients most frequently. It has shown that one can predict certain recipes based on the data analysis. Stocks can be optimized by determining the most and less popular ingredients.

Case study 2. Machine learning in additional sales

Machine learning in the recommendation system is an efficient way to boost sales.

“Clustering customers in terms of their taste preferences, we loaded recommendations for each cluster. People started buying more diversified products than we offered earlier using the algorithm based on expert judgment,”Andrey Filipyev explains.

However, AI recommendations help to sell products to users who have made at least one purchase and signed in. Well, what to do with clients who have entered a website/mobile app for the first time or who do not leave their data?

Therefore, Andrey developed a recommendation system based on products in the Shopping Cart.

“We have a lot of new customers, and we do not know their purchase history and preferences. That is why, by December, I designed a simple model recommending goods driven by previously selected products. Here is an excellent example: the system has recommended the Greek salad instead of the most popular snack, Dodster, to the ‘Vegetables and mushrooms’ pizza,”Andrey says.


As the result, Dodo Pizza’s sales grew after the launch of machine learning in the recommendation system.

AI Conference: Machine learning in inventory forecasting and recommendation systems: Dodo Pizza’s case study 1
Dodo Pizza’s sales growth per weeks: the chart from the Habr article by Andrey Filipyev

Andrey Filipyev will discuss more machine learning benefits for retailers on April 9 at AI Conference. The expert will share best practices of ML development and integration into company’s systems as well as will stress the results that can be achieved using the technology.

About Andrey Filipyev ►►►

Register to the event ►►►

AI-industry news and cases for you