Product Recommendations with Artificial Intelligence

CWI has been a fundamental partner for Arezzo, contributing to implementing artificial intelligence in the brand’s e-commerce, optimizing processes and reducing the time needed to update product recommendations. This collaboration allows Arezzo to instantly offer tailored product suggestions to its customers, improving the purchasing journey and boosting sales.

The largest retailer of women’s fashion footwear in Latin America is using artificial intelligence (AI) to improve product recommendations in its e-commerce sites. Previously, using a manual tool, the company faced challenges in creating and maintaining product recommendation rules. Whenever a new collection was launched, professionals needed to update all the rules, registering new information.

This approach required a lot of effort and limited the ability to adapt to customer preferences, requiring frequent adjustments and generating operational difficulties. Sometimes the team created so many specific rules that it became difficult to maintain them, modify them, and understand exactly what was being recommended.

In 2021, Arezzo conducted an internal proof of concept with AWS Personalize, an AI-based product recommendation tool. After the success of POC, the system was transformed into a viable product, initially being implemented in the Arezzo, Schutz and ZZMall brands. Later, product recommendations were expanded to the Anacapri and Vans brands, with the aim of further improving the customer experience. AI has been integrated into brands’ apps, providing personalized product recommendations on mobile devices as well.

Real-time Suggestions

One of the most striking features of this implementation lies in its ability to offer personalized product recommendations in real time, adapted to customers’ behavior while browsing websites. The AI system covers several points of customer interaction, from automatic events generated while browsing the website to product registration data provided by the brand itself.

Arezzo can even suggest products even when the user is not logged in, using the device’s interaction history. Furthermore, the company uses customer purchase data in physical stores to personalize product recommendations in e-commerce.

Arezzo recommends products based on customer interaction on the website, their behavior and product characteristics, using a classification tree. They consider the customer’s context, such as promotions, without changing their preferences. When there are performance issues, they perform analysis, A/B and functional tests, and can retrain the AI model or reconsider the type of recommendation.

Additionally, rules external to artificial intelligence are applied, such as the exclusion of products already purchased or out of stock, allowing additional adjustments and customizations in the recommendation process.

White woman with blond hair holding a cell phone.

In the image we have two smartphones. On their screens there are images from the Arezzo website. Next to them is a notebook. On its screen there are images from the Arezzo website. Next to it is an image of a woman's legs. They are white and on her feet are black pumps.

Schutz: Tailor-made Catalogs

In 2022, Schutz expanded its catalog to include apparel, previously focused primarily on bags, shoes and accessories. This strategic change was accompanied by significant innovations in its online platform, aiming to offer a more complete and personalized shopping experience to its customers.

The company invested in technology to better understand its customers’ preferences and behaviors. Using artificial intelligence algorithms and data analysis, the company was able to capture and interpret user behavior in real time. Each customer interaction with the website – from viewing a product to completing a purchase – is recorded and analyzed, feeding a robust and dynamic database.

When visiting the website, customers are presented with dynamic product recommendations that adapt as they navigate through the pages. For example, if someone views a leather boot, the system can suggest accessories or clothing items that complement the style of that boot. All with the aim of making the shopping experience more engaging and relevant for the user.

This data-driven and technologically advanced approach has not only strengthened Schutz’s position as a leading brand in the fashion market, but has also significantly improved customer satisfaction and brand loyalty. By integrating digital innovation with a deep understanding of consumer needs, Schutz has established a new standard of excellence in fashion e-commerce, demonstrating a continued commitment to evolving and improving its services.

Algorithms and Techniques

HRNN (Hierarchical Recurrent Neural Networks) is an artificial intelligence algorithm used by Arezzo to recommend products to customers based on their interactions on the website. It uses neural networks to understand temporal patterns in user activities, such as browsing sessions. This allows product recommendations to be adjusted in real time, taking into account customers’ most recent interactions. This ability to adapt is crucial due to constant changes in consumer habits and fashion trends.

In the image there are several abstract elements representing artificial intelligence.

Tangible and Measurable Benefits

To measure results, A/B tests are used, which provide metrics such as conversion rate and revenue generated by recommendation carousels. Monthly analyzes are also carried out on millions of storefronts for each brand, seeking to understand the performance of recommendations and their impact on sales.

This project has the collaboration of several CWI professionals. Among them are data scientists, developers, testers and DevOps. They provide a recommendation solution through an endpoint developed by CWI, which is accessed by the teams responsible for the visual development of the Arezzo website. This approach allows for efficient integration of AI recommendations into the Arezzo website, optimizing the process and reducing the time required to implement and update recommendations.

Challenges of AI-Powered Product Recommendation

One of the main challenges faced is understanding customer behavior to determine the ideal time to recommend certain products. This is because the same model does not always perform well throughout the year. During extended periods of sales and promotions, or before the launch of a new collection, it is common for performance to be lower. This suggests the existence of questions of interest that were not adequately captured by artificial intelligence.

Every week, the Arezzo team holds a performance monitoring ceremony, where the results of A/B tests and performance trends are analyzed. At that time, actions are defined to adjust the models, if necessary.

In the image there is a white person, with short dark blond hair and a dark blond beard looking at a notebook screen.

Future Projections

CWI is planning to develop an internal product recommendation platform for Arezzo. This will allow the company to have more control and flexibility over the recommendation process, being able to adjust and customize different aspects as necessary. Transitioning AWS Personalizer to an in-house platform will not only provide more control and autonomy, but also cost savings by not having to rebuild the entire system every time a modification is needed.

In the image there are nine cwisers inside a room sitting in front of notebooks.

In the image there are eleven cwisers inside a room sitting next to each other.

The Value of Partnering with CWI

At CWI, we approach the topic of Artificial Intelligence on two main fronts. At first, as a means to create new solutions or functionalities. Secondly, as a tool to increase productivity and improve the quality of our deliveries.

This culture within software engineering allows AI to be explored in diverse contexts and partners. This is possible due to the constant exchange of knowledge between cwisers and customers. Thus, experiences acquired in a particular project or solution can benefit other partners, even if in different sectors or areas of activity. This expertise in various verticals is an important aspect that contributes to CWI’s competitive advantage as an AI partner.

See how we are accelerating development and delivering value to customers through the use of artificial intelligence!