Has Amazon just single handedly initiated the commoditization of the recommendation engine? Amazon Personalize, the company’s machine learning-powered product recommendations technology, has this week been made generally available to AWS customers – popping the lid on technology which has long been an exclusive luxury relished only by the retail platform itself.
Amazon’s reasoning for expanding Personalize to AWS customers essentially boils down to its claim that recommendation and personalization software can be challenging to deliver across a variety of use cases because there isn’t what it describes as a single, master personalization algorithm.
But do AWS customers want a master personalization algorithm? And will it ultimately improve the consumer experience? In fact, does anyone, even pay TV operators or OTT video service providers, big or small, have any desire for a master personalization algorithm and in doing so remove any edge they might have on competitors? We can’t speak for service providers, but we know a few vendors in the recommendation software space which would fight to the grave against such an insult to their years of pioneering R&D work.
Yet Amazon might have a point. Perhaps the time has come for recommendations to become ubiquitous in a sense, rather than siloed into one engine for video, another for music, another for news articles and another for product sales. AWS might have sowed the seed for cross-data applications and our guess is that doing so will remove some of the fundamental value in each recommendation served.
Of course, each use case comes with its own nuances and specificities, each requiring a unique mix of data, algorithms and optimization to generate a result – i.e. a recommendation.
The exclusive nature of Amazon Personalize technology to the ecommerce world of Amazon.com is why readers of Faultline Online Reporter may be unfamiliar with the technology, and indeed why the term has never appeared in our archive. So, let’s take a quick look at how it works.
Amazon Personalize essentially simplifies the development of applications with personalization use cases, including product recommendations, individualized search results and customized direct marketing. Personalize trains, tunes, and deploys custom machine learning models, while provisioning the necessary infrastructure and managing the full machine learning pipeline. An API provides customers with results on a pay-as-you-use basis – claiming no machine learning experienced is required.
It’s so easy, Amazon claims, that AWS customers just have to provide Personalize with an application’s activity stream, be it video views or number of signups, for example, along with an inventory of the items to recommend, such as videos, products, or music – and then receive recommendations via an API. Personalize processes and analyzes this data, identifying anything meaningful and useful via multiple algorithms which it says have been built and refined over years for the retail business.
The immediate applicability and efficiency of Personalize in serving recommendations outside of the retail sector could therefore be thrown into question, and no doubt rival vendors will be hammering this point home to customers should any deployment strongholds come under threat from Personalize. AWS coming after your lunch is a daunting prospect, nonetheless, which is why recommendation software vendors like ThinkAnalytics have been smart and diversified their product lines in recent years to encompass marketing, big data and more.
Indeed, AWS has driven Elemental to new heights since taking over the encoding business in 2015, yet rival encoding vendors have nit picked at AWS Elemental enough to make bold claims about the technology appearing to be much cheaper on the surface than in reality. Its cloud computing clout and myriad of services surely warrants a higher price tag, however, while still undercutting some of the best on the market.
A somewhat ominous customer case study comment from Japanese electronics manufacturer Yamaha said, “We expect Amazon Personalize to be more accurate than other recommender systems, allowing us to delight our customers with highly personalized product suggestions during their shopping experience, which we believe will increase our average order value and the total number of orders.” Of course, the highly personalized product suggestions Yamaha is referencing are relevant to ecommerce, not necessarily content.
AWS has a formidable user base in a variety of verticals as we know. In our universe, it plays host to content from Netflix and Hulu, to cite just two, which just so happen to be two of the three top SVoD platforms in the US. The third is Prime Video, so you get the picture. The significance of Amazon Personalize becoming generally available does not, however, lie with the tech giants relying on AWS, but in readily making machine learning techniques available to application developers and data scientists at businesses of all sizes across all industries.