Harnessing the real-time web to provide better recommendations
November 06, 2015
Everyone who works in e-commerce should know who Greg Linden is. Not only should they know who he is, but they’ve probably devoured his “Early Amazon” blog posts. A lot of Greg’s work paved the way for modern e-commerce as we know it.
Greg was the Amazon engineer who built the original product recommendation engine. What originally started as a small side project quickly became a significant portion of Amazon’s revenue, making up 35% of their revenue in 2006. The recommendation engine originally only recommended items at checkout, but now the engine is integrated across the whole platform, from search, to product pages, to emails.
Amazon isn’t alone. Netflix is another pioneer of the recommendation system. In a blog post from 2012, Netflix revealed that about 75% of what people watch on Netflix is a result of their recommendation engine. This is a key differentiator for them, and is a large part of the reason why people spend so much time watching Netflix.
They work so well, in fact, that a number of companies are building recommendations-as-a-service products. The value proposition is clear: implement a good recommendation system and you will see revenue growth.
The key to a good recommendation system is good data. Not just any data will work; it’s incredibly difficult to find correlations in unstructured or poorly organised data. In many cases, cleaning up your data set is the most difficult and time-consuming part of building a recommendation engine.
But clean data isn’t enough. A truly effective recommendation system can only be built when you have enough data. A recommendation engine built using only first party data – typically, browsing behaviour and purchase data – is limited to a very narrow understanding of your customers. You might have the what, but you’re missing the who and why.
Your customer might proceed to checkout with only a New York Yankees t-shirt in their cart, but what if you also knew that your customer played football for the Florida Gators and was a huge fan of Seinfeld and Whose Line is it Anyway?
Without this sort of data, you’re missing out on the opportunity for even more tailored recommendations that could lead to increased revenue.
So how can you get this data? First-party doesn’t give you a holistic understanding of your customers, and asking customers outright is a great way to get them to leave your site. Fortunately, many of your customers are already generating this data somewhere else: on the real-time web.
The real-time web is an incredibly rich source of data. We live in age where 500 million tweets and 3.5 billion Instagram likes are generated every single day. Consumers are constantly telling the world who they are and what they like through the medium of social. Using these data sources, Macromeasures can build an incredibly deep and structured profile on who an individual consumer is and what they’re interested in. Their gender, their location, their hobbies, the publishers they read, the celebrities they like – all constantly evolving along with popular culture.
Using our profiles, engineering teams can add a layer of intent and context to recommendation systems, even when a registered consumer hasn’t browsed any products yet, because their demographics and psychographics are known ahead of time.
To put this into perspective: when a customer browses your store for the first time, and you know that this customer is a woman who likes hiking, tea, and John Green novels, you can show her a different set of products than a man who follows competitive gaming, watches Mr. Robot, and loves Taco Bell.
Interested in delivering more powerful personalisation to your customers? Get in touch with us at email@example.com.
Macromeasures is an audience intelligence platform that helps brands and marketers better understand their consumers. Our technology transforms the billions of pieces of otherwise unstructured public social data into deep profiles on the demographics and psychographics of a brand’s audience and customer base at the individual level.