Back to Data | E-commerce
AI: how Einstein enhances online customer experience in Italy
November 15, 2023

Carrefour Italy has transformed its e-commerce experience through Salesforce Einstein. This isn’t about the famous physicist but a cutting-edge tech solution. The real magic? Salesforce Einstein’s ability to personalize the customer experience using data & AI. Discover the behind-the-scenes of Carrefour’s approach.

Three years ago, Carrefour Italy ecommerce website was experiencing a 10-fold increase in traffic during the first wave of the pandemic… that couldn’t be handled at that time. Three years later, Carrefour Italy is now announcing an improved online customer experience thanks to artificial intelligence powered by Einstein. It’s not the Albert we all know, but it’s a technology provided by our partner Salesforce.

What happened during the last three years is Carrefour teamed up with Salesforce to implement “Commerce cloud” in Italy. This new technology stack offers a scalable, flexible and secure ecommerce platform. More importantly, since then Carrefour Italy has offered online shoppers an intuitive user-interface, enhanced search, and new payment methods such as vouchers, gift cards, and subscription models… But what about going further and personalizing the customer experience? That’s precisely what Salesforce Einstein does.

Italian market insights and beyond

Since 2022, AI’s use has strongly accelerated in Italy. Nowadays, 61% of large Italian companies have initiated at least one project in this field, according to Among these companies, 42% have multiple operational AI projects, showcasing how AI is becoming an integral part of companies’ strategies and operational practices.

Let’s now go with an additional turn of the screw: With generative AI, new use cases are emerging everywhere at a mind-blowing pace. Let’s just pick a ‘random’ example with Carrefour’s GenAI initiatives 😉 In Carrefour France, we have for instance launched “Hopla”, our ChatGPT-based chatbot on Hosting friends this Friday, with one being vegetarian and another peanut-allergic? Hopla recommends a recipe and lets you add the ingredients to your cart. Also, there is the “Carrefour Marketing Studio”, powered by Google Vertex AI, for swift social media campaign generation.

Hopla, a service powered by GenAI on & Salesforce Einstein: powering data-driven e-commerce

What is Einstein? It is an integrated set of AI technologies, embedded in their customer management suite. Einstein includes a wide range of technologies, including machine learning, natural language processing (NLP), and data analysis. The primary goal of Einstein is to help companies like Carrefour make the most of available data and become data-driven, uncover hidden insights, provide predictions and recommendations. 

These predictive capabilities are integrated directly into Carrefour’s applications, allowing for personalized and anticipatory actions while automating tasks to offer Carrefour customers a better shopping experience and boost their engagement with more relevant content. Moreover, Einstein also enables up-selling and cross-selling opportunities by showcasing complementary items. 

That’s what we did on the website. AI assists Carrefour merchandisers in eliminating guesswork, providing constant commerce insights essential for maximizing conversions. We’ll come back to this, but first, let’s take a look under the hood and see how Einstein works.

Decoding Einstein: how personalized recommendations work

Recommendations and strategies

Einstein algorithm is capturing customer characteristics, learning from the actions they take, and ultimately suggesting products that they might be genuinely interested in. The algorithm takes into account the device, the operating system in use, and whether the customer is new or not. As a result, different customers may see equally diverse recommendations at the same time.

With Einstein, we can establish various product recommendations and strategies. Depending on the recommendation type, different strategies can be implemented. Currently, our most commonly used for our references are:

  • Customers who bought also bought
  • Real-time personalized
  • Customers who viewed ultimately bought
  • Recent most viewed
  • Recent top sellers

Recommendations’ rules

After determining the strategies and their importance, we establish the rules. For each recommendation, up to 30 rules can be formulated and applied. Every rule specifies an action that determines how to display, hide, or rearrange matching items within the recommendation list. Watch out not to set conflicting rules when configuring them. For instance, don’t have rules that both show and hide the same product items or that promote and “demote” the same products simultaneously. There are four available rule actions to select: 

  • show: Displays items matching the specified field values and conceals those that don’t.
  • hide: Conceals items that align with the specified field values.
  • promote: Elevates items for increased visibility in recommendations.
  • demote: Deprioritizes items, reducing their visibility in recommendations.

Einstein Configurator

When setting rules, we pick values for specific fields using the Einstein Configurator. After a few adjustments, we preview the potential product recommendations. When a customer views a product on our site, details are sent to Commerce Cloud Einstein. Depending on the recommendation type, we might provide product details or category info. The final step is activating the recommendation to display the product suggestions in our modules.

Generally, most recommendation engines offer standard product suggestions (1:many suggestions), but some provide very personalized experiences (1:1 suggestions) based on a user’s individual browsing interactions.

Einstein’s effectiveness hinges on the quality and volume of data. Better predictions arise from accurate and varied data. Salesforce Einstein uses machine learning to analyze data, find patterns, and improve predictions over time as it ingests new data. Another crucial part of Einstein is natural language processing, enabling it to comprehend written text, extract insights from various documents, and offer relevant responses

Salesforce Einstein Configurator interface.

Leveraging AI to improve online experience

Now, let’s see how we’ve implemented Salesforce Einstein on First, a very concrete example with the “zero result page”. When a customer searches for a product that is currently unavailable, Einstein offers recommendations through two modules. The first displays the ‘most viewed products‘ in that category. The second module, labeled as ‘You may also be interested in’, displays suggestions in a carousel, which are based on cross-category recommendations. For instance, if you added a smartphone to your basket, the algorithm might suggest you purchase a charging cable.

Both of them can also be used to ‘promote’ private-label products that have been most visited by customers on in the last 7 days.

Suggesting products encourages customers to continue their online shopping, helping reduce the abandonment rate. Thus, this initial layer of AI is crucial for ensuring a quality experience for customers and generating business for us.

When no article is found Salesforce Einstein shows recommended products through two modules.

Crafting customized user journeys with AI

The Product Detail Page (PDP) is not just a place to showcase a specific product with its images, description, price, and purchase options. We’ve strategically incorporated the ‘You may also be interested in’ module there. However, this module operates under more sophisticated rules.

Firstly, it’s designed to suggest complementary product categories to the customer. For instance, if a customer is viewing pasta, the recommended products module might suggest sauces and pesto.

Secondly, it can recommend similar products. As an example, if a customer is looking for a pack of cookies, the module might suggest other types of cookies.

Lastly, it promotes seasonal products or items related to specific occasions. For example, if a customer is looking at sparkling wine, the recommended products module might highlight pandoro and panettone, popular treats during Christmas festivities.

Tuning Einstein to maximize e-commerce performance…

During our exploration of artificial intelligence, we faced many challenges. However, we’ve learned and made necessary adjustments. It’s clear that there isn’t a universal solution for every e-commerce platform. The combination of strategies, recommendations, rules, and other factors creates a unique dynamic. The key to success is closely monitoring specific Key Performance Indicators (KPIs), as these various components can lead to a myriad of outcomes.

Additionally, what may have worked well for a period might not be suitable in the following months due to seasonal changes, catalog updates, or shifts in sales strategies. Therefore, Salesforce Einstein is a tool that requires regular monitoring and updates.

Our progress has two parts: 1/ Setting up Einstein ; 2/ Reviewing and tweaking its  algorithm rules.

We faced issues during the review. One major problem was that AI product suggestions weren’t visible to all users due to do-not-track cookies. To solve this, we swapped the ‘Real-time personalized’ strategy for ‘Recent top sellers’, which doesn’t need user consent. We also shortened its timeframe from 30 days to 7.

… by tweaking its algorithm’s rules

The rules we set were too strict. We then shifted from simply “showing” products to actively “promoting” them. We decided to emphasize specific categories and products, as well as crafting tailored recommendations to highlight our private labels and ongoing promotions.

After reviewing the navigation categories for our “online grocery” and “electronics and home” sections, we revised the rules. Currently, there are 23 rules with a focus on food categories and Carrefour private labels. The remaining rules can be utilized for more seasonal references or products featured in ongoing campaigns on our e-commerce platform.

A few examples of rules we set in the Einstein configurator.

Monitoring Einstein’s performance: Gauging user interest

We evaluate Einstein recommendations on using two KPIs: Click Through Rate (CTR) and Add to Cart (ATC). CTR gauges the relevance of a suggestion and customer interest, while ATC measures how often suggested products are added to carts, indicating user interest even if it doesn’t always lead to a purchase.

When we compared the CTR from the first month of our new recommendations to the same month last year, the results varied. Only between May 9th and May 23rd did we see a positive change in the CTR. Did we get worried? No, because this KPI can have three limitations:

  1. CTR reflects user interest in the suggested product. However, other factors, like where the recommendation appears on the page, can impact it. For example, if the carousel displaying the recommendation is “below the fold” (a term meaning it’s not visible without scrolling), it might not get proper attention.
  2.  CTR calculates “clicks per view.” However, even if a user doesn’t see the recommendation, the view is still counted, regardless of whether the carousel of suggested products was actually visible to them.
  3. Moreover, the products shown might not be what the customer wants right now since they’re based on what items are frequently purchased together. In groceries, most visitors already know what they want, which can affect the results.

In the graph below we can see the evolution of the 2023 CTR (blue line) compared to 2022 (orange line):

Monitoring Einstein’s performance: Assessing purchase intent

The Add to Cart (ATC) metric tells a different story. Since making our changes, this KPI has improved from the previous year and keeps rising. Specifically, the ATC on product detail pages (PDP) hit 26%, and for pages with zero results, it soared to an impressive 35%! This is notably above the average for other e-grocery platforms, which lags by 12 percentage points.

The chart displays average ATC values for product detail pages (PDP) and zero result pages over time. The upward trend since implementing the modifications is evident. This marked rise in ATC shows that our changes to the recommendation system effectively encouraged more users to add suggested items to their carts, especially on product detail pages and even when no search results appeared.

Overall, the data indicates that our changes positively influenced user engagement and shopping habits, resulting in a notable rise in the Add to Cart rate, surpassing other e-grocery platforms by a wide margin.

Wrapping up our journey with Einstein

In a nutshell, the journey of integrating and optimizing Salesforce Einstein into Carrefour Italy’s e-commerce platform has been both challenging and rewarding. Over the past three years, Carrefour has enhanced the onsite & online shopping experience for its customers, with AI and data proving to be a game-changer (ie: Optimizing bakery production & sales with AI). 

The results underscore the significant impact of personalized, AI-driven product recommendations. Carrefour’s ability to adapt, learn from challenges, and implement necessary tweaks has been crucial in not only meeting but also exceeding customer expectations. As AI continues to evolve and transform the digital landscape, Carrefour remains at the forefront, poised to harness new innovations for the benefit of its customers.

About the Author

Senior Success Manager, Salesforce

As Success Manager, Rossella is responsible for driving the adoption of the Salesforce B2C Commerce solution across a portfolio of Italian retail customers to unlock opportunities for growth. In close alignment with the customer’s goals and objectives, she shapes a path to value, by leveraging the full Salesforce Success Ecosystem resources.

E-commerce Specialist

Elisa currently holds the role of e-commerce specialist in Carrefour Italia and is responsible for the organization of the information architecture of the e-commerce, the optimization of the internal search engine, as well as the coordination of projects related to contents and reviews online. Before joining Carrefour she worked in the technology division of PwC. She graduated in business management from the School of Management and Economics of Turin and obtained a master's degree from the Polytechnic of Milan in Digital Innovation Management.

What’s new?