The Real Me
The algorithm wars started in Search. One of the key takeaways from that time was made by the great technologist Vint Cerf who famously said Google didn’t have better algorithms, they just had more content.
Algorithms becoming ingrained and habit forming in our content discovery process may have started with Search but they have moved into other content experiences, most notably video. And while the resources needed for video cognition are much greater than textual, the value of being able to “see” is also greater.
In the consumer space YouTube “Suggested Videos” has sat for years on the right rail as the most powerful and valuable content algorithm. It kept people watching YouTube (aka viewing ads). While YouTube rabbit holes are fun and powerful, they still rely primarily on text and broad content classifications as well as view history. Those content graphs have limited amounts of nodes and relationships are very hard to discern. For example, YouTube is not smart enough to suggest I watch Big Audio Dynamite after The Clash even though they have the same lead singer.
While the suggested video algo of YT has not advanced much, other image and video classification and deep learning tools/technologies have progressed rapidly in the past five years. OpenAI is of course a big part of this. Technophiles are very excited about the image classification and content delivery aspects of DALL.E but that is just one use case. Across sight, sound and motion, recognition and delivery are gaining intelligence.
The Punk Meets the Godfather
In the past couple years the world of algorithmic content has had a sea change event. As most everyone knows by now it is called FYP or For You Page on TikTok. And darn if it isn’t leveraging Vince Cerf’s ideal learning scenario of massive amounts of content.
The TikTok algo makes the YouTube’s algo look out of the dark ages. It captures hundreds if not thousands of features. The difference between YouTube and TikTok is the difference between having a map and a self-driving car. Powered by the advances in image and video recognition TikTok is an inflection point for media.
But what is the inflection?
You may love it or hate it but TT’s algo is tuned to details in your interests and tastes far beyond any other content algorithm ever created. It learns what you like from your first time on the app and will exploit this learning.
Meta with Instagram Reels (the fastest growing part of IG) and YouTube with Shorts (50% of the top channels across YouTube) and Snap with Stories have no choice but to compete with TikTok now for consumer attention. The first net effect of this competition is the standardization of short form video. Short-video is no longer a “TikTok thing.” Scrollable short video feeds on our phone have become a format. This is an underrated event in media that many will be playing catch-up on.
It is also worth noting the huge advantage the short video feed has in machine learning vs other content delivery formats. A short video feed is a perfect training environment for a content algorithm. It can measure time spent as a proxy for interest. It can deliver a massive training sets of videos quickly because of the swipe and the multiple events it captures within each object. Put another way, it can easily explore your tastes, understand your elasticity, tempt your media palate and exploit this learning.
Is It in My Head?
Using feature extraction at scale on a per user basis has proven so effective that I believe these type of content delivery algorithms along with short-video feeds will soon be extend to all media delivery and consumption. There’s no reason why other apps and content providers with decent volumes can’t do the same.
Let’s take one big use case for TikTokization that is already losing the algo war. TV viewing.
When we turn on the TV why does it go to the last channel we had on? Why is it scheduled? Why is it programmed? Even “on-demand” is reliant on scrolling through lists of shows and then consecutive calls to action just to preview a show (and if you don’t like it you have to labor through multiple back-button taps to get to another selection). This is an antiquated and poor user experience for content discovery. My Verizon Fios is dumb!
Netflix has started to realize this. They have begun to auto-play shows in the background when you navigate. But this is still less than ideal and certainly few GenZ people have the attention span for this.
I’m reminded Spotify caught on not because of its early catalog but because it never buffered its stream. You could change songs and it was immediate. The immediacy of relevant short video is powerful and underrated consumer preference. As media consumers we’ve always demanded immediacy and control. Latency and irrelevance are a media killers.
The TikTokization use case that interests me most of all in media is vertical content.
While vertical content providers may not have the scale of data they can compete and win for relevance with better data models. Nobody knows their consumers better than vertical media.
As with all new media formats, differentiation will evolve by unbundling / verticalization. We can begin to see the early stages of this. The new IG algorithms are being auto-tuned to the content you already consume on Instagram outside your social graph. Looking at top YT Shorts channels you will see them made up of humor and kid related content (aka media training the next generation).
If short video does become vertical powered with better data modeling it means we are sitting at the cusp of brand building opportunity the likes of which we haven’t seen since the advent of cable TV. The same media opportunity that was seized by brands like HBO, CNN, Discovery that are still relevant 40 years later can be captured in vertical short video.
Just like cable there would be first mover advantage. This might explain why YouTube, IG and Snap pushed right into it. There are probably not going to be two short video apps where people share their skateboarding tricks. There are probably not going to be two short video apps for sharing live music. But I think we will see these and more unbundled from the current short-video platforms over the next 5 years.
This is because people desire to relevance whether they are opening an app or turning on a channel. We want what we want when we want it. It’s no coincidence that reinforcement is both a critical part of consumer media conversion and deep learning models.
We watch what we want to be. And our attention spans are shrinking. The future of content is then algorithms that understand us and formats that show us who we are.