Why Retail Recommendation Engines Need More Than Buzzword Tech
Pairing customers with the right products isn’t as easy as it might seem. Recommendation engines have come a long way over the years, and most deliver satisfactory upsells. The perfect method of steering paying customers toward additional complementary products they actually want has proven difficult to nail down. This is critical to help ecommerce retailers to increase their profit.
Most companies depend on collecting personal user data through cookies, browsing history, and past customer purchases. And even then the recommendations are not guaranteed to convert—most retailers have poor cross-sell performance, leaving money on the table. And retailers can see this when they compare their cross-sell performance in-store versus online.
The next wave of tech behemoths are turning to highbrow technologies like AI and machine learning to improve their success rate. However, incorporating buzzword tech is no surefire solution to perfect cross-sell product recommendations either.
The good news is that there are easier ways for e-commerce retailers to see results today. Let’s take a deep dive into the often the world of recommendation engines, and discuss how smart approaches can be a better choice than high-tech ones.
The thinking behind cross-sell product recommendations
Bundles are proven to be the best method of cross-selling online; when done right it helps both the customer and the company. The most popular example of bundling is McDonald's: a single hamburger is often not enough to warrant a meal, so cross-selling with fries and a drink to complete the trio is presented in the interests of both parties. The customer gets more food, and the company sells more, higher-margin products. This is how recommendations should be considered in the ecommerce age. There are bundles, which are compatible, relevant for the customer’s purchase, and easy for them to buy in one click, while helping the retailer sell more to maximize their margin and revenue.
Most companies today try to figure out what complementary products to recommend based on big data. It focuses on comparing and contrasting prior transactions and what previous customers have appeared to be interested in. Armed with all this data, and product inventories to clear, ecommerce retailers try to deduce the most likely combination to result in an extra sale.
However, recommendation engines are not perfect entities, and the bulk run into three main problems: irrelevant recommendations that look like they don’t make sense,; missing recommendations on products, or incompatible products that cause unhappy customers to return products, hurting the retailer’s bottom line. This means the resulting recommendations might not make sense in comparison to the product being bought, because it fails to understand the “why” of each product sold and its relationship to other products.
Why most product recommendations leave money on the table
Recommendation engines often fail because there are so many variables and so much data required to deliver accurate recommendations. Retailers are turning to hyped technologies like machine learning and Artificial Intelligence in an attempt to fill the gaps.
But these have their flaws and continue to fall short, particularly with bundle recommendations, because they need to understand the products being recommended and have enough data on them. While 35 percent of Amazon’s revenue comes from recommendation engines, Amazon still does not have enough data in 57 percent of cases to make any suggestion at all—and that is a massive missed opportunity. If one of the world’s biggest retailers does not have sufficient data to make bundle recommendations, there is little hope for others to be able to collate and unlock customer behavior to their benefit. And there’s an even bigger problem, just looking at past customer purchases, doesn’t guarantee the products will work together. For example, on Amazon, you’ll often find that recommended bundles are incompatible, such as the laptop bag being too small for the laptop being sold. This will cause unhappy customers to return products, and hurt their bottom line.
And this is why many try new and old methods to improve their cross-sell performance. Retailers often fall back to manual recommendations to associate one SKU to another or set up simple rules. However, this is yet another imperfect system because it’s not scalable across thousands of products, and constant new product releases.
Back to the (bundling) drawing board
Bundling strategies have been successful for years. As we shift to more and more online purchases, we’re losing the expertise of the sales assistant in store who understands the products being sold. This is the foundation to effective cross-sell promotion, understanding the product being sold to recommend the exact complementary product to maximize margins.
This works best when demand is high, with the aim to bundle products together for a higher total sale price. And where possible, cross-selling items from the same distribution centers is an effective method to maximize margins by further reducing costs on delivery.
The three pillars of cross-selling that each retailer must focus on are compatibility, relevance, and optimization. This requires retailers to understand their products before leveraging customer behavior in order to make the most relevant cross-sell recommendations. An additional product sale per transaction is big business—with huge margins at stake—this means it is integral to bottom lines to get recommendation engines right.
Anthony Ng Monica is the CEO of Swogo. Hundreds of retail leaders in over 30 countries around the world drive profitable growth with Swogo. Swogo takes a unique approach that focuses on understanding a retailer’s product assortment - Swogo Product Graph combined with machine learning and AI algorithms surpassing billions of recommendations per year.