Product recommendations for e-commerce sites are not new in concept, but the suggestions they present to shoppers are increasingly getting smarter thanks to the algorithms behind them.
And the result of delivering more relevant product ideas? Higher spend of course. When Jewelry.com partnered with omnichannel personalisation technology firm, Dynamic Yield, to integrate personalised product recommendations on its website, for instance, it saw revenue increases per visitor of 39% from the homepage, 13% from product pages, and 18% from cart pages.
The key, according to the team, was not just to focus on the usual ‘most popular’ or ‘similar to current item’ suggestions, but instead to turn to machine learning to automatically select the most effective strategy for each user.
That meant finding a personalisation strategy that would work for both visitors with a rich history of behavioural interactions, and those who are new to the site, thus for whom minimal information is known. Doing so is about capturing signals from shoppers about their buying intentions and preferences for specific products as they move through the sales funnel, the Dynamic Yield team explained, and then providing upsell and cross-sell opportunities throughout.
“Traditional retail is beginning to have what we like to call a ‘moneyball moment’ where the old way of simply making gut decisions on which experience to serve your customers is being challenged. As machine learning technology becomes more advanced, algorithms will outperform humans in recommending products that users are more likely to show an affinity for, and ultimately buy,” Mukund Ramachandran, CMO of Dynamic Yield, notes.
“With Dynamic Yield, we can use machine learning to make data-driven recommendations based on where visitors are in the sales funnel. The ability to assess the level of valuable information about each visitor and automatically serve the most effective strategy has empowered us to increase revenue across our site,” said Jon Azrielant, director of marketing at Jewelry.com.
On the homepage, for instance, the Dynamic Yield widget leveraged affinity-based recommendations, recommending products according to a weighted score of what returning users had added-to-cart, viewed, or purchased in the past. To induce engagement among new visitors, the widget presented products with the highest amount of page views and click-through-rate on the site.
“While ‘point solutions’ for deploying product recommendations have existed in the market for decades, these solutions are limited by data silos that restrict their algorithms to only making decisions based on a user’s interactions with product recommendation widgets. With Dynamic Yield’s unified data stack, information onboarded from all onsite interactions, third party data, CRM data and loyalty data can be ingested,” Ramachandran explains.
On the product pages, the team has been running A/B tests, comparing 45% of users who were recommended products ‘similar to the current item’, 45% who were recommended ‘top-selling’ products, and 10% who received?a control variation. As a result, Jewelry.com revealed that recommending ‘top-selling’ products provided a 10% uplift compared to the other variations.
Finally, an additional widget was introduced on the bottom of the cart page to showcase items frequently bought together with the current item.
“These results are very strong compared to industry benchmarks. We think this is the case because with Dynamic Yield product recommendations are only part of the puzzle. The entire site starts working better for you – the homepage engagement is higher which leads more people to discover the most relevant products as they browse,” Ramachandran adds.
This post first appeared on Forbes.com.