Task	Improve recommendation efficiency on the website and make results tracking transparent
Solution	Optimize UTM tags on the website for correct order attribution to the interaction source Update the website recommendation system
Results	7x conversion rate +76% CTR Order Share from recommendation blocks grew from 2% to 20% in 14 months
Resources	Senior Performance Marketing Manager at PUMA Customer Success Manager at Yespo Yespo development team PUMA development team NetPeak team

About the Project

PUMA is a global sports and casual apparel brand represented in 60 countries worldwide. The company not only cares about product quality but also pays special attention to service, creating a positive experience for customers.

Since 2022, PUMA Ukraine has been communicating with customers and implementing marketing activities through Yespo CDP – using email, Viber, web push, App Inbox, as well as personalized recommendations on the website and in triggered email campaigns.

The retention strategy is developed and implemented by Netpeak agency, actively working to strengthen customer loyalty. The project is supported by a Customer Success Manager from Yespo: they help use the CDP capabilities to the fullest and participate in optimizing business processes to improve marketing campaign results.

Challenge

As part of a regular process review, the task arose to increase attribution transparency. One factor was insufficient accuracy in reporting metrics display. To correctly assess the contribution of recommendations and channels, it was necessary to align metric calculations between systems.

The task was to identify improvement points and coordinate processes and updates that would ensure unified attribution logic across all systems.

During the same period, Yespo developers were implementing a global platform update – new transformer-based recommendation models. The decision was made to test the new recommendations on the PUMA website to evaluate their performance.

Against the backdrop of these changes, it became even more important to improve tracking to accurately assess how much more effective the new algorithms were compared to the previous ones.

Solution

Features of Product Recommendations on the PUMA Website

Product recommendations on the website are a proven way to increase the conversion rate of existing traffic. This tool is a cornerstone of retention strategies for many major brands.

PUMA Ukraine uses several types of Yespo recommendation blocks on the website, including:

• “You may also like”, “People also buy”, and “Recently viewed” – on the product page
• “Bestsellers” (bestsellers from the category) – on the category page
• “Bestsellers” (personally selected products based on popular items) – on the homepage
• “Recommended for You” (based on popular items) – on the 404 page

Block settings and performance statistics can be managed directly in the Yespo account, in the “Website – Recommendations” section.

UTM Tag Optimization and Data Transfer

During the analysis of website recommendation performance, a need was identified to unify session data storage – attribution tags were not preserved when users navigated between website pages.

Specifically:

• If a user came to the website from advertising (PPC) and then clicked on a product from recommendations, the completed purchase was attributed to the recommendation block rather than the original channel.
• If a visitor clicked on a recommended product and then changed the color or size in its card, the tag also disappeared. Because of this, the sale might not be attributed to the block that actually prompted the customer to purchase.

To make analytics reflect the real customer journey, an update was implemented on the website: now the mechanism preserves initial UTM parameters throughout the entire session, even if the user switches between channels or changes product parameters.

This helped improve sales attribution and eliminate data discrepancies in Yespo and Google Analytics. Now teams could more accurately evaluate the effectiveness of recommendations and campaigns.

Implementing Next-Generation Recommendations

In April 2024, Yespo developers updated recommendation algorithms, transitioning to transformer architecture. This is the same technology that powers ChatGPT and other leading AI systems, and is used today by global market leaders such as Netflix, Amazon, and others.

What Changed in the Algorithms

Previous recommendation systems worked on the basis of collaborative filtering, analyzing similarity between users or products. Transformers work differently:

• Sequential modeling: the algorithm analyzes a user’s sequence of actions as a complete story, not separate events.
• Contextual understanding: the system considers not only what the user bought, but also when, in what order, with what intervals.
• Deep learning: models independently identify behavior patterns and make more accurate suggestions.

What Does This Mean for Ecommerce Business?

• More effective marketing campaigns and personalization
• Improved shopping experience, increased repeat purchases and loyalty
• Increased average order value
• Improved conversion in highly competitive conditions
• More accurate recommendations with limited user interaction history data

Learn more about transformer models in this article.

How the New Algorithms Integrated with PUMA Campaigns

The update was launched centrally – Yespo CDP clients didn’t need to change anything in their settings. For PUMA Ukraine, the launch of new models coincided with preparation for the summer sale season. The growth in traffic and interest during sales created ideal conditions to test the new algorithms in action.

Results

The new AI architecture, accurate purchase attribution, and organic traffic growth collectively produced a significant impact on PUMA’s recommendation system. From April 2024 to June 2025, the following results were achieved:

• CTR (Click-Through Rate) increased by 76%
• Conversion rate of recommendation blocks grew 7x
• Order Share from website recommendations increased 10x – from 2% to 20%. Now every fifth order on the PUMA Ukraine website comes from personalized recommendations.

Blocks placed on product and category pages generate the most sales. Here’s how sales shares were distributed among different types of recommendation blocks from April 2024 to June 2025:

• Product Page “You May Also Like” – 46.00% The most effective block, which suggests similar products based on viewing or purchase, provides almost half of all orders from recommendations.
• Category Page “Bestsellers” – 33.83% The block with category bestsellers remains a consistently powerful sales generator.
• Product Card “Recently Viewed” – 13.26% This block actively returns customers to products they were already interested in and encourages them to complete the purchase.
• Product Card “Complete the Look” – 5.03% The cross-selling mechanism based on complementary products brings noticeable additional revenue.
• Homepage “Bestsellers” – 1.47% Despite a relatively low share, the homepage remains an important platform for starting interaction with recommendations.
• 404 Page “Recommended for You” – 0.38% Even on a “dead-end” page, recommendations work and return users to relevant products.

These figures prove: recommendations work effectively throughout the user journey – from the homepage to the product card – and support different purchase scenarios.

Plans for the Future

The PUMA Ukraine team doesn’t plan to stop and continues to develop the recommendation system to help customers find “the right” products even better.

New blocks are a vivid example of Yespo CDP’s flexibility. The platform allows creating both basic algorithms like “Bestsellers” and complex custom solutions adapted to a specific niche.

Want artificial intelligence to study your audience’s behavior and suggest exactly those products that interest customers most? Schedule a consultation with our experts, and we’ll show you how personalized recommendations can increase your sales.

Get professional expertise