Even Netflix's early algorithms — well before the streamer had its own original TV shows — recommended movies and TV shows at the sweet spot between what customers wanted to watch and what Netflix wanted them to watch.
The company's cofounder and first CEO described Netflix's early DVD-matching system that launched in 2000, called Cinematch, in his new book, "That Will Never Work: The Birth of Netflix and the Amazing Life of an Idea."
That recommendation engine helped shape how Netflix suggests titles to viewers, Randolph told Business Insider in an interview in advance of the book's publication on Sept. 17.
Cinematch matched titles based on viewers' tastes and the company's DVD inventory.
Cinematch, Netflix's first recommendation system, was developed to make it easier for people to find movies they would like. But it didn't do anyone any good to recommend titles that Netflix didn't have in stock.
From the very beginning, Netflix's algorithms were designed to make suggestions that optimized for the company's inventory of DVDs, as well as viewers' tastes.
"We had to find a way to recommend items that weren't the easy ones that everyone thought about, which was new releases, but ones that they might like even better that we also happen to have in stock or have better economic availability," Randolph said. "I think — I'm not part of the company now — that's never changed."
To gauge viewers' tastes, Netflix asked people to review and rate titles. It "clustered" users based on their overlap of positive and negative reviews to better understand what they might like to watch next, according to the book.
A five-star ratings system was also introduced after much debate on how it should work. "That stupid star rating system was the source of hundreds of hours of argument," Randolph wrote in the book. "More battles over fewer pixels have never been waged."
Netflix's algorithms are much more advanced today, but likely built on the same principle.
Today, Netflix's personalized recommendation system is much more complex and nuanced than it was in 2000.
Netflix has more than 2,000 taste communities that it uses to make recommendations based on what individual viewers, and viewers like them, have enjoyed. The company predicts to a percentage point how likely you are to enjoy a title. It customizes the homepages for each user, down to the title art. It's constantly testing new ways of enticing audiences, like autoplay video and curated collections of titles organized by theme on mobile. The hotly debated star-ratings system is gone, replaced with a simpler, "thumbs-up, thumbs-down" ratings model.
Randolph thinks the principles remain the same. Netflix's primary offering is now streaming video, not DVDs. The system could optimize for other business drivers, like Netflix's original TV shows and movies.
"It was all tied together, but it all came back to that single point of wanting to create a place that helped match you with entertainment that you liked," Randolph told Business Insider. "That still is what Netflix stands for. It's what led toward, I believe, doing their own content."