Deriving Meaning through Machine Learning: The Next Chapter in Retail
Three slides from Benedict Evans’ brilliant talk, The End of the Beginning, really caught my attention.
Across industries, machine learning is helping us get to successive levels of meaning of what a thing is. It’s helping us explicitly understand what things are, a leap forward from the existing state of affairs, which rely on extrapolation through indirect inference. In the context of retail, the implications of this are significant. Here’s why.
For years, companies have been trying to target consumers to push personalized recommendations for product discovery — be it through targeted ads served via social networks like Facebook/Instagram/Google or through personalized pages on e-commerce sites like Amazon/Walmart/eBay. Yet, the dominant mode of discovery continues to be the humble pull centric keyword search, which is restricted by users’ ability to translate intent into words.
For e-commerce sites, the handicap is that it’s difficult to tell who the user is, since they lack demographic information, preferences, friends, and so on. The workaround is to use recommender systems to infer what other users with similar on-site behavior prefer.
For social ad platforms, the handicap is in knowing what a thing really is. Retail ads are usually tagged with catalog feed data, which is restricted in the number of structured fields and the degree of normalization — poor in comparison to the rich catalog data that e-commerce sites carry.
Between the two groups, social ad platforms have the bigger handicap; it’s tougher to get by without knowing what you’re selling, than it is to get by without knowing who you’re selling to. Which is why these companies have had limited influence on retail behavior. But all of this is changing as machine learning becomes increasingly mainstream.
By developing a semantic understanding of what retail ads are actually referring to, social ad platforms can overcome their product data handicap. And what’s more, with algorithms that can draw in information from the public web, they can augment their understanding with external knowledge. Borrowing a mental construct that Benedict Evans invokes, the best way to think about this is to imagine how an army of interns annotating each retail ad with their semantic and cultural inputs might help improve the platform’s understanding. Think of the number of additional attributes that each ad can now carry, and how developing a real understanding of what each ad represents could help offer more relevant and potent personalized recommendations.
Excited yet? I know, it’s tough to get pumped about “better ads”? But what if I put it this way: think about what happens when these platforms begin to integrate this awareness into first order experiences. Forget ads, think Facebook posts, Instagram stories, Google search results and Pinterest pins. Imagine a world in which your shopping recommendations are as relevant as your social newsfeed/timeline recommendations are? Wouldn’t that change the game!
This trend can have meaningful value-add. Consider an example from your author’s personal experiences. As a vegan who looks out for ethically sourced products, shopping is inevitably an exercise in enervating research, making my relationship with e-commerce restricted and utilitarian. But if an army of interns were to have done this research for me, my shopping experience might actually be pleasant and my average spend is likely to increase.
The opportunity for supercharging personalization is massive here. E-commerce spending in the US is still only 8% of total US retail and a far tinier fraction of total consumer spending. The Beginning is only just Ending, and most of the experiences in the chapters of retail and advertising remain to be written. It’s going to be very interesting to see how the various players in this game acquire data and use machine learning to shape this future.
This article was originally published on the Semantics3 Blog
