Ever since Amazon first deployed its recommendations system based on human curation and best-seller lists across its ecommerce platform two decades ago, product recommendations have been an important part of the broader retail shopping experience. Today, product recommendations are ubiquitous across most ecommerce sites — improving visitor journeys and enabling better product discovery as brands constantly push to drive up conversion rates and increase online revenues. Just how important are product recommendations to the top line?
Like most technology innovations over the past two decades, recommendations have evolved (albeit only recently) and have become much more effective at matching shopper interests with the right products at the right time, thereby genuinely changing the experience for the end customer in a dramatic way. However, despite recent developments, many brands still rely on machine learning-powered recommendations engines, the most prominent being collaborative filtering, and because of that, miss the mark in actually creating dynamic and personalized experiences for their customers. A far more effective approach to recommendations is those driven by sequential deep learning technology.
So how do machine learning and collaborative filtering techniques fall short in an effective revenue-generating ecommerce strategy that meets the demands of today’s customers?
Firstly, to be effective, collaborative filtering models require loads of data, largely reliant on looking for patterns in this data to findproducts that are similar to each other in terms of “viewed together,” “bought together,” and “bought after viewing a different product.” These data-hungry models, in their product-to-product nature, forget the most important element — the customer.
Further, machine learning-powered recommendations engines tend to suggest only the top 30% of the product catalog. However, this top 30% — comprised mostly of “hero” products — are actually only relevant to 16% of visitors, according to Qubit’s own research.
Lastly, these data-hungry engines with a limited understanding of the full catalog typically run in batch methods every few hours, or in some cases daily. The stark reality is that most consumers see the same recommendations on a given product, an experience far from the real-time needs of today’s customer. The mismatch between what the customer is looking for and what is recommended is detrimental to the quality of the shopping experience, resulting in frustrating shoppers who may quickly bounce to another brand to more easily find what they’re looking for.
Take this scenario for example: shoppers browsing for tan purses may see a “You may also like” carousel of selected similar products based on the collective browsing behavior of other customers. The products selected in this carousel using conventional recommendations are typically limited to those items that are most viewed or purchased within the catalog, but they may not be the most relevant to a particular customer.
Conversely, customers are 40% more likely to spend more than originally planned, and at least twice as likely to add items to the basket, when the delivered experiences are highly personalized, according to a Google/BCG study.
To be truly effective and successful, recommendations should be pulling the best products from the entire breadth of the catalog in order to effectively inspire customers with truly personalized product recommendations.
Conventional product recommendations are ineffective at creating a personalized 1:1 shopping experience because they are product-to-product. Since deep learning-driven recommendations are customer-to-product versus product-to-product, the entire experience is highly personalized based on the individual shopping journey, taking into context actions occurring as recently as a last-second interaction.
Highly Evolved, Accurate and Personal
Deep learning is much more effective at powering product recommendations because it uses layered sets of algorithms that mimic a neural network inspired by the makeup of the human brain. By mimicking the brain, deep learning is able to create important connections between different items, such as products, with much less consumer data than machine learning-based collaborative filtering.
These connections can also be made with a variety of different interconnecting “signals” such as:
- Every aspect of the product, including product metadata (i.e. color, size, category)
- Understanding how the customer landed on the product (i.e. referral source)
- Time spent on viewing a product
- The browsing order of products
The other key benefit of using deep learning recommendations over the conventional approach is the speed at which deep learning can make connections between customers and products, using very little data. This is where the neural network really comes into its own, enabling significant exposure of much more of the product catalog to visitors and adapting truly to the individual shopping journey in real time. The result from our analysis? Deep learning recommendations tend to show 300X more products than conventional systems.
As the ecommerce “universe” has grown more complex over the past two decades, consumers have also become much more savvy about shopping online. Amidst rising expectations and demands, consumers have an array of preferences, category affinities and price limits. They also browse at different times of the days, in various orders and with varying levels of purchase intent. Conventional product recommendations cannot consume and utilize this amount of contextual data.
Lastly, and perhaps most noteworthy to retail brands, while conventional recommendations engines typically increase revenue by up to 3%, deep learning recommendations can increase overall ecommerce revenue by up to 8%.
Brands should be on guard about letting their ecommerce stack get stale and outdated. One of the most important competitive weapons in the conversion toolbox isproduct recommendations. By delivering true 1:1 personalized recommendations to customers, brands can expose more of their product catalog, including “long tail” products, to provide shoppers with the most relevant items and more personalized shopping experiences.
The result is 300X more products shown, as customers find products 3X faster, and driving 3%-5% more revenue over standard recommendations. Deep learning recommendations can be a true competitive advantage in an increasingly crowded retail marketplace.
Sergio Iacobucci is Director of Marketing at Qubit.