September 2009 – Cover Story: Fulfilling the Product Recommendations Promise

Proven best practices to help retailers get more out of their online product
recommendations, driving a higher and more lasting increase in conversion
and deeper customer engagement

By Darren Erik Vengroff, Ph.D.

Online product recommendations have been around for almost as long as e-commerce itself. Most online retailers today offer some type of product recommendations on their sites, with research proving that presenting shoppers with items that fit their interests boosts sales and increases customer satisfaction. But the truth is, many retailers are not doing all they can with recommendations; they plunk them down on product pages, do little to monitor performance, cross their fingers and just hope the recommendations work to boost sales a bit. Often, there is an initial spike in conversion, but performance tends to fizzle out after a few months.

However, there are simple ways to make recommendations work better–proven tactics that not only drive a greater initial lift than standard recommendations, but also create a lasting and sustainable lift over time. For years, I’ve been working hand-in-hand with retailers to devise the best personalized product recommendations, starting when I was a principal engineer at Amazon.com and now as chief scientist at next-generation recommendations provider richrelevance. I’ve learned a lot about what works and what doesn’t in the quest to deliver on the ultimate promise of personalized product recommendations: a faster, easier and more enjoyable shopping experience. When done right, relevant product recommendations can deliver a five- to 30-percent lift in conversion right away and sustain or even increase these levels over the long term. On the contrary, if your recommendations aren’t relevant and timely, you risk alienating site visitors so much they go elsewhere to make today’s purchase and never return to your site to make tomorrow’s.

Retailers that want to take their product recommendations systems to the next level of relevance can follow a few best practices that have proven effective for a variety of large and small online retailers. The following techniques will help you get more out of recommendations, so that you can start delivering on the promise to your customers of the best possible shopping experience (and, in turn, benefit from a sustainable lift in repeat visits and sales.)

Use a variety of competing recommendation techniques
There are a number of different concepts people reference when they’re talking about recommendations: the wisdom of crowds, wisdom of communities and “behavior banks” to name a few. Cut through the jargon, though, and you’ll find there are really four main ways to deliver recommendations. Segmentation divides users into groups based on characteristics like age, gender and geographic location. Collaboration, or collaborative filtering, starts with an individual and attempts to locate others like him. Personalization relies on a user’s prior actions to determine what they are likely to do next. Similarity starts with products, rather than users, and models relationships between them to serve recommendations.

Most recommendations platforms use one or two of these techniques. However, in today’s high-stakes market, that’s simply not enough. To deliver hyper-relevant recommendations, you need to use a recommendations platform that leverages all four at once. Retailers like Sears and Burton Snowboards, which use more than one recommendation technique, have been able to boost site-wide sales by up to 25 percent over the long term and drive significant increases in user loyalty and repeat purchase rates.

So how do you combine the techniques to create the best overall shopping experience? Make sure the recommendations provider you work with offers all of the recommendations types and that their system continually tests the performance of all recommendations, optimizing and automating the combination of techniques to deliver the most relevant, impactful recommendations.

Give merchandisers control
One mistake many retailers often make is adopting a recommendations system that takes merchandisers completely out of the equation. That will never work. Merchandisers know things that no recommendations “black box” is ever going to know–like the fact a particular camera is going to be replaced by a new model three weeks from now, that you have a specific promotional agreement with a manufacturer that precludes recommending another brand against their products or that a backlog of summer merchandise is larger than expected and needs to be presented to the customers most likely to be interested in it.

You need a recommendations system that allows merchandisers to manually override certain recommendations and fine-tune recommendations based on their inside knowledge. For example, a merchandiser may want to increase the rate at which they recommend overstocked products, but still only recommend them to those people who are most likely to be interested in them, as opposed to simply pushing them out to everyone whether or not there is any reason to believe they are interested. Make sure the recommendations provider you work with offers merchandiser controls, because a great recommendations system doesn’t try to replace the merchandiser, but instead gives them the tools to do “bionic” merchandising online.

Put recommendations everywhere
Your on-the-go, price-comparing customers don’t shop in only one place, so you need to recommend products to them everywhere they are: online at your site, online at an affiliate site, looking around a store or reading an e-mail on their iPhone. When implementing a recommendations system, make sure the provider allows you to automatically extend the use of personalized recommendations across all customer touchpoints, both online and off, and throughout the customer lifecycle. You need to recommend relevant products to potential customers from the minute you intercept them (via your online ads), when they land on your homepage or product page, while they’re browsing around your site, after they leave your site for an affiliate or competitor’s site, a few days later when they get a promotional e-mail from you and when they come back to your site for a second look. If you constantly recommend relevant products to shoppers, no matter where or how they are engaging with your brand, you’ll deliver an extremely useful service to your potential customers–and we all know that happy customers are paying customers.

Get the message right
Over the years, I’ve seen many retailers make mistakes with how they “message” the recommendations to site visitors, presenting products in a way that turns off potential buyers. You should always explain clearly where the recommendation came from and why you recommended it to this particular person. Instead of “May we recommend”–which sounds as if you’re trying to unload excess inventory–communicate something like: “People who purchased this DVD player also purchased this HDMI cable,” or “63 percent of people who looked at this shirt eventually bought this other shirt instead.”

Be as specific as possible.
Say, “Because you purchased three Bruce Springsteen CDs in the past, you might like this new Bruce Springsteen CD.” Research has proven that when you add clear, understandable language to each recommendation, conversion not only increases more significantly, but customer service contacts decrease. Make sure your recommendations platform allows you to add your own messages to recommendation types so that your site visitors know exactly why those products are being recommended to them.

Avoid classic mistakes
Make sure to avoid recommendations pitfalls that can damage your brand and negatively impact sales. Some of the most common–and avoidable–mistakes I’ve seen include: recommending the product a customer is currently looking at; recommending controversial or inappropriate products that may damage your brand (such as R-rated films in a children’s film category); over-recommending products to users for whom they are not relevant; and not giving customers control over their recommendations (for example, not allowing them to tell you to no longer recommend products based on the present they bought for their niece two years ago). These mistakes, and others like them, are avoidable if you use a next-generation recommendations platform that leverages all four types of recommendations, because most of the above problems stem from using just one recommendation technique.

Most importantly, remember that personalized product recommendations are the start of an ongoing dialogue with your customers. The end goal of product recommendations may be more sales, but to get there, you need to nurture the customer relationship through thoughtful, relevant recommendations that are truly useful to your customers.

Darren Erik Vengroff is chief scientist at richrelevance (www.richrelevance.com). He is a renowned computer scientist and technology industry veteran. For both start-ups to industry leaders Amazon.com and Goldman Sachs, Vengroff has designed and built a variety of world-class, customer-focused, scalable applications.