I read The New York Times every weekend. I wake up, retrieve it from my doorstep, make coffee and I read it, glancing up once in a while at the Manhattan skyline. It's a great experience for me. It's not that I can't get similar content on the web instead, or read it comfortably on my iPad, it's just that heft of the print edition reconfirms to me -- in a familiar way established over years of habit -- that someone I trust curated what I read, and by positioning where each story runs, real people are making recommendations on what they think I might find interesting. Moreover if I have someone over, we can share by trading sections of the paper, something that can't happen on my iPad or PC.
As much as I appreciate the depth of experience behind each "recommendation" I get from the editors -- as I move away from the homepage or front page of a website where I totally get the value -- and I dive into the various articles -- the recommendations I get are often broad, category-based, and not necessarily targeted to me. That's because they were never meant to be for me to begin with, but rather for the majority of their audience .Not an easy task for anyone. Even for machines.
And therein lies the challenge. How can a handful of editors "curate" to satisfy millions of very different readers? Now, on the web as opposed to the print version, editors could potentially go article by article, and match them with links to videos and other articles based on some parameters, and keep fine- tuning those every day. However, that might be costly. Mind you, this is essentially matching what millions of different people want to watch or read after reading every single story. As an example with video recommendations -- on a site with 10 million users, where the average user reads 3 stories, with 1,000 available videos to be recommended per story -- we're talking about a range of 30 billion options of recommendations during users' sessions. A lot.
Machines on the other hand, are suppose to be able to make those decisions by looking at what readers are spending time on, what they've read and watched in the past, and instantly recommend other stories or videos they may like. And machines can do this for a hundred readers, a thousand, a million or tens of millions, each individualized to the reader's behavioral patterns, predicting what will most likely engage them to read or watch further. Think of it like a smart TV network, that instead of showing you the same prime-time lineup of shows, serves you up programs it knows you like based on your past viewing history. With the right data, the network could even introduce you to new shows that it has "learned" over time you will probably like.
There is big value in having editors put items such as videos in front of you that you never knew you might like. It broadens your horizons and gives the news dimension. It also helps you relate to the brand and the people behind it. So, in a perfect world, the editors and the machine coexist on every site.
If you're a publisher and you wonder how to combine the two, here are some tips:
(1) You can combine human and machine intelligence in a way that helps add revenue. A nice example is how the New York Times is presenting other videos you may like, and right below it a similar UI of featured content curated by editors, maximizing users' experience. (Disclosure: The New York Times is a customer of my company.)
(2) Make sure you're labeling your recommendations in a way that works with users' expectations. If a recommendation is machine-generated, let the user know.
(3) Much like the car you drive, aim to get the best machine-learning engine you can for your site. Your users are important. It's easy to disappoint, and it's hard to rebuild trust that is broken by underpowered recommendation functionality.
(4) Some machine-learning vendors offer capabilities to bias their recommendations, and input some guidelines. Those can make your editorial team feel more comfortable. In some cases, publishers can allocate certain recommendation-slots and promote videos or categories of their likings. This is usually used if a publisher needs to promote a video they've sponsored, etc.
We are at an inflection point in the media world where machines can take over much of the back end tasks that maximize the user experience and result in more revenue for the publisher. But, as I am affirmed every Sunday morning with my coffee and print edition of the Times, there will always be a place for human editorial judgment. Also, on the Internet, editors and machines can indeed coexist.