The ecommerce landscape has fundamentally transformed through the application of artificial intelligence to product recommendation and personalisation. Where online retailers once presented the same product catalogue to every visitor, sophisticated personalisation engines now ensure that each customer sees a unique assortment of products tailored to their individual preferences, behaviour, and predicted propensity to purchase. The ecommerce personalisation market is valued at approximately $11.4 billion in 2025, with personalised experiences now driving more than 35 percent of Amazon’s total revenue. Product recommendation engines represent the most visible application of personalisation technology in ecommerce, but personalisation extends far beyond recommendations to encompass homepage customisation, search personalisation, email product suggestions, and real-time behavioural triggers. For marketing and ecommerce professionals, understanding the technical foundations of personalisation, the platforms implementing these capabilities, and the methods for measuring personalisation impact has become essential for driving revenue and customer experience improvement.
At the foundation of personalisation lies the recommendation engine, a software system that predicts which products a customer would find most relevant or most likely to purchase. Three broad classes of recommendation approaches have emerged, each with distinct strengths and limitations. Collaborative filtering represents the oldest and most theoretically well-established approach. The core insight behind collaborative filtering is that customers who have similar purchase histories or ratings patterns are likely to find similar products appealing. If customer A and customer B have both purchased and rated products in similar ways, and customer A has purchased product X which customer B has not yet seen, then customer X is likely to appeal to customer B. Collaborative filtering is powerful because it requires no explicit information about product attributes or categories: the algorithm learns purely from patterns in customer behaviour. However, collaborative filtering struggles with new customers who have no history, and with new products that have no purchase history. This is known as the cold start problem.
Content-based recommendation engines take a fundamentally different approach, focusing on product attributes rather than customer behaviour patterns. These engines build detailed profiles of products including attributes such as category, brand, price range, colour, materials, and other descriptive features. They simultaneously build customer profiles representing the attributes and features that customer has historically preferred. The recommendation engine then matches customer preference profiles to product attribute profiles, suggesting products where the attributes align with historical preferences. Content-based approaches handle cold start more gracefully than collaborative filtering, because even brand new products can be described through attributes and matched to customer preferences. However, content-based approaches are limited by how well product attributes can be defined and whether customer preferences remain stable. Content-based engines also tend toward recommending similar products to what customers have already purchased, potentially creating filter bubbles where customers never discover genuinely novel offerings.
Hybrid recommendation engines combine collaborative filtering and content-based approaches to capture the strengths of both. Hybrid engines might use collaborative filtering to identify customers with similar tastes, then use content-based matching to find products with attributes that align with established group preferences. Alternatively, they might use collaborative filtering to generate initial recommendations, then filter through content-based rules such as ensuring diversity across product categories. The most sophisticated modern recommendation engines employ machine learning models that process dozens or hundreds of signals: past purchases, browsing behaviour, ratings given, time spent viewing products, demographic characteristics, session context, inventory levels, margin levels, and many other features. These models are trained to predict the probability that a customer will purchase a particular product, or alternatively to predict the revenue that would be generated by recommending a particular product. The model outputs then drive the personalised recommendations shown to each customer.
Personalisation Platforms and Ecommerce Applications
Specialised ecommerce personalisation platforms have emerged to implement these recommendation engines and broader personalisation capabilities. Dynamic Yield, acquired by Mastercard, provides a comprehensive personalisation platform serving major retailers with recommendation engines, homepage personalisation, email personalisation, and content personalisation. Barilliance focuses specifically on ecommerce retailers, offering product recommendation engines optimised for different use cases including homepage recommendations, email recommendations, and post-purchase upsell recommendations. RichRelevance similarly serves ecommerce retailers with powerful recommendation engines and attribution reporting. These platforms integrate directly into ecommerce platforms like Shopify, Magento, BigCommerce, and custom implementations, capturing data about customer behaviour and serving personalised experiences across all customer touchpoints.
Personalisation in ecommerce extends far beyond simple product recommendations. Homepage personalisation customises the content and layout of the homepage for each visitor, emphasising product categories or specific products that are most relevant to that customer based on their history or profile. Search personalisation modifies search results rankings so that different customers see different products in different orders when searching for the same query: a customer who has previously purchased luxury brands might see premium products ranked higher in search results, whilst a price-sensitive customer might see discounted products ranked higher. Email product recommendations automatically generate product suggestions in marketing emails based on browsing behaviour, purchase history, or abandonment behaviour. Real-time behavioural triggers automatically show product suggestions or special offers based on the customer’s immediate actions: if a customer is viewing a product category and hasn’t yet added anything to cart, a personalisation rule might trigger a “top sellers in this category” recommendation section. These various applications of personalisation work together to create a highly customised experience where each customer sees an ecommerce environment tailored to their specific preferences and predicted purchase propensity.
Measuring Personalisation Impact and Privacy Considerations
Measuring the return on investment from personalisation implementations requires careful experimental design and attribution methodology. The fundamental challenge is that personalisation is continuous: customers are always receiving personalised experiences, so there is no simple “before and after” state to compare. The most rigorous approach to measuring personalisation impact involves running A/B testing experiments where some percentage of traffic receives the personalised experience whilst a control group receives a non-personalised version. By comparing conversion rates, average order value, and customer lifetime value between personalised and control groups, retailers can quantify the incremental revenue impact of personalisation. However, A/B testing personalisation is complex because depriving control group customers of personalisation creates a worse experience than they could otherwise receive. As a result, many retailers employ multivariate testing where multiple personalisation approaches are tested simultaneously, or they measure impact through cohort analysis by comparing personalised customers with historical non-personalised baselines.
Privacy regulations and third-party data restrictions are reshaping personalisation strategies. Historically, personalisation engines relied on third-party data about customers: cookies tracking browsing behaviour across websites, data partnerships providing demographic or behavioural enrichment, and lookalike modelling using third-party audiences. The deprecation of third-party cookies and privacy regulations like GDPR and CCPA are forcing a shift toward first-party data and privacy-compliant personalisation. First-party data consists of information collected directly from customers through their interactions with the retailer’s own properties: purchase history, browsing behaviour on the retailer’s website, email engagement, customer service interactions, and customer-provided preferences. This first-party data is actually more valuable for personalisation than third-party data because it is specific to the retailer’s products and customers. Privacy-compliant personalisation using first-party data also reduces privacy risks and builds customer trust. Forward-thinking retailers are building robust first-party data collection strategies, investing in customer data platforms to unify first-party data, and developing personalisation approaches that deliver strong results using only privacy-compliant data.
| Type | How It Works | Strengths | Limitations |
|---|---|---|---|
| Collaborative Filtering | Recommends products based on behaviour patterns of similar customers | Discovers non-obvious recommendations, scales well with large datasets | Cold start problem for new customers and new products |
| Content-Based | Matches customer preference profiles to product attribute profiles | Handles cold start for new products, interpretable recommendations | Limited to known attributes, tends toward similar products |
| Hybrid | Combines collaborative and content-based approaches with machine learning | Captures strengths of both approaches, better cold start handling | More complex to implement and maintain, requires substantial data |
| Machine Learning | Trains models on purchase probability using hundreds of customer and product signals | Highest accuracy, handles complex patterns, optimises for business metrics | Requires significant data and engineering expertise, harder to interpret |
| Use Case | Technology Required | Revenue Impact | Implementation Complexity |
|---|---|---|---|
| Homepage Personalisation | Customer data platform, personalisation engine, real-time decisioning | 5-15 percent lift in ecommerce conversion rate | Medium: requires customer data integration and creative management |
| Product Recommendations | Recommendation engine, product data, user behaviour data | 10-30 percent of incremental revenue in many verticals | Medium: platform integration straightforward, optimisation ongoing |
| Search Personalisation | Search platform with ranking customisation, customer segment data | 2-8 percent lift in search conversion rate | High: requires search platform customisation and extensive testing |
| Email Product Recommendations | Email platform with dynamic content, recommendation engine integration | 15-25 percent increase in email click-through and conversion rates | Low: many email platforms support recommendation blocks natively |
| Real-Time Behavioural Triggers | Personalisation platform with real-time decisioning and rule engine | 3-10 percent conversion rate improvement on triggered recommendations | Medium: requires ongoing rule configuration and testing |
The maturation of ecommerce personalisation technology has created significant competitive advantages for retailers who implement it effectively. The practical challenge lies not in the technology itself, which is now available from numerous platforms and is relatively straightforward to implement, but rather in the data foundation, experimentation rigour, and ongoing optimisation discipline required to extract maximum value. Retailers that invest in clean first-party customer data, that run rigorous A/B testing of personalisation approaches, that continuously refine recommendation algorithms based on performance data, and that evolve personalisation strategies as privacy regulations tighten, will continue to drive substantial revenue premiums from personalisation. As third-party data becomes less viable and privacy-first personalisation becomes the norm, first-party data and first-party data-based personalisation will become even more valuable sources of competitive differentiation.
