Though Pikis represent an interface to product search that you may not have seen before, you’ve encountered modern product recommendation if you’ve ever made a buying decision online (and almost everyone has). If you’ve browsed books on Amazon, rented a DVD suggested by Netflix, or clicked on an ad in Google, you’ve taken part in product recommendation systems in their current incarnations.
Though product recommendation systems used by the corporate giants you know and love operate by bombarding you with products you’re apt to buy based on your previous selections, the products they recommend aren’t necessarily the products you need most or even want. Corporations who rely on product recommendation to sell products collect and analyze volumes of data about their customers and products, but give very little transparency about how they make their suggestions. Ultimately, consumers know almost nothing about why they are being recommended certain products. A closer look at the patents held by the goliaths reveal that most modern product recommendation systems are designed with one priority: the profits of the company they benefit. The idea of optimizing recommendations for consumer benefit fell by the wayside not long after it first appeared.
To identify where product search went wrong, we delved into the trail of patents back to the early product search engines. The concept of the product search engine is older than the Epic of Gilgamesh, relative to the history of the internet, and we traced Pikimal’s lineage as far back as the late ‘80s when the original recommendation engine was first patented.
The Early Years: Consumer Benefit Ruled Product Recommendation
Feb. 12, 1991 – A Pennsylvania company called H-Renee Inc. was awarded a patent for a straightforward product search engine. Much like Pikimal, this engine accepted user input about desired specifications, and then retrieved a list of products from a database. H-Renee’s system relied entirely on filters, allowing the user to select a range of desired specifications, and possessed no means for ordering or ranking the products.
H-Renee’s engine was devoted entirely to narrowing down lists of products for users based on user input, and was constructed entirely with the idea of alleviating the hassle of shopping for consumers. The paragraph featured here is a long list of complaints the patent holder had about finding the right computer. Sound familiar?
Feb. 26, 1991 – The first recommendation engine earned a patent. The inventor names John Hey filed the patent for a company called Neonics, a company about which an internet search reveals little. This engine was modest in scope: Hey anticipated recommending movies, music and books to individuals based on how similar users rated them. The ratings table featured here is copied from the actual patent, which suggests, if nothing else, that Hey had a sense of humor about his work (who would rank The Untouchables lower than Beverly Hills Cop?). Which movie would you recommend to Jones, based on Smith and Wesson’s ratings?
The Mid 90s – Consumer Benefit Takes a Back Seat to Corporate Interests
Oct. 29, 1996 – Kevin O’Connor and Dwight Merriman of internet advertising company DoubleClick applied for a patent for targeting and tracking ads online. The system operated by building up online profiles of users based on what ads they clicked, using the information to determine individuals’ interests and preferences. The algorithm would then selectively serve up ads to meet those interests, encouraging consumers to purchase more goods from their affiliate advertisers in their areas of interest. The question of how the ad was determined remains a mystery to the user, and whether or not the product offered was the best for the consumer was not a concern of the patent. Unlike its predecessors, DoubleClick went on to survive the dot-com bust, acquired several other advertising companies in the oughts, and was finally purchased by Google in 2008 in a whopping $3.1 billion deal.
The Oughts through Modern Day: Corporations Looking out for Corporations
Long since the early days, tailoring recommender systems to maximize profitability has become a massive industry. 2007 saw Netflix offering a $1 million reward for anyone who could improve their recommendation algorithm by 10%. In the meantime, studies of Netflix’s recommendations have revealed that Netflix ultimately encourages users to buy niche products over blockbusters, as these movies are less expensive for them to supply. Thus, if you’re taking recommendations from Netflix, you’re ultimately watching the movies that are cheapest for them to provide to you, not the ones they think you’ll enjoy the most.
Meanwhile, the reigning overlord of the e-commerce world, Amazon, has published a patent for their product recommendation engine, a more highly developed version of the first recommender system, which is aimed at encouraging you to purchase as many products as possible. Finally, researchers at Brown University have directly published research on maximizing profit using recommender systems, in which is stated outright that companies have to by sly and incorporate targeting with profitability to achieve the best results.
Ultimately, in order to be a savvy consumer, you must pay attention to what companies stand to profit when you’re taking their advice on products. If you want objectivity, make sure to go to the recommenders who have stated that your interests are their mission, companies who give you complete transparency about your choices and ultimately companies whose profits don’t increase or decrease depending on the item you choose. Although the concept of a consumer-oriented product search engine has been 20 years in the making, Pikimal is attempting to get back to the roots of recommendation algorithms: by basing rankings on facts and user preferences. With a completely transparent interface, you can dependably find the best product for you each and every time.