How Awdisea predicts video views
Walt Disney

There was a time when Walt Disney would have a released an aniimated film only to watch it bomb. Creating animation is a very expensive undertaking. Disney was able to significantly reduce the cost of producing an animated film by only proceeding with the animation once a rough mockup of the film was given positive reviews by an internal team. The mockup would require scanning a lot of hand skteches and adding narration and music. You can watch this video of Steve Jobs describing this during an interview:

https://www.youtube.com/watch?v=BiXK11naSfY

The process of testing your idea during pre-production seems to have been something forgotten by Hollywood over time, as can be seen with the countless poor quality movies they've been producing over the past 20 years.

YouTube videos may not be on the same level as full Hollywood production movies, but even a high quality 15 minute YouTube video often requires higher production costs. Many YouTubers have been successful with virtually no production costs and simply record a video of themselves doing something.

But with nearly 4 million videos uploaded every day to YouTube, it becomes increasingly more difficult to become noticed and retain subscribers. If the YouTube algorithm does decide to promote your video, it could be a make or break deal based on nothing more than your video thumbnail and title with the thumbnail taking precedence. Many viewers won't even bother to read the video title if the thumbnail doesn't connect with them.

Zach King

And yet there are YouTubers that are able to produce consistently better quality videos over time and grow their channel. While there are many factors that go into producing a succcessful video, the primary reason why they succeed is because they produce highly engaging content that the majority of their subscribers enjoy watching. Behind these videos are creative people who have developed a mindset to recognize what viewers find interesting in their work and focus on that type of content. Zach King is one of those gifted creators. While he may have 40 million subscribers as of 2024, much of his earlier work bombed. The same could be said of MrBallen - a channel that specializes in "the strange, dark and mysterious".

Many YouTubers who aspire to become as creative as their role models, wish they could spend a day with them, get to know them, see how they work and peer into their way of thinking. For most people, this is never going to happen. The next best thing to getting inside of their heads is using Artificial Intelligence, and this is where Awdisea comes in.

Awdisea is based on an AI that tries to get as close to recreating the elements that go into the thought process of coming up with a great idea before even beginning any production on the video. As humans, we are subconscious about those elements. For example, you don't really think about the age group or gender you are targeting. If you're a 25 year old woman doing makeup videos, you already know who the vast majority of your viewers are. But the YouTube algorithm doesn't. It does try to figure it out using AI. For the most part it gets it right. While there are many parameters that the algorithm uses to figure out what to recommend to viewers, the title of your video and the similar videos that your subscribers are watching are the two biggest parameters. YouTube doesn't even use the thumbnail. This is because thumbnails are so vastly complex that even AI can't tell what makes a great thumbnail. But humans on the other hand are instantly captivated by a thumbnail while scrolling through suggested videos. Most people will only read the title if the thumbnail draws their attention and they want to have more context about the video.

Because the YouTube algorithm cannot use the thumbnail when it comes to recommending your video to others, it depends much on the title to find viewers who have already watched videos with similar titles. This is why your YouTube home feed always gets a lot of suggestions of similar videos whevener you watch just a few videos that have related titles. The algorithm also relies heavily on your existing subscribers. If a large portion of your subscribers watch a new video you post, it will recommend the video to others who have watched similar videos. But unless enough of your subscribers watch the video, it is very unlikely that the algorithm will even bother to recommend it to others.

It is highly recommmended that you read the documentation on How the YouTube algorithm promotes videos so that you can better understand how it influences your video views and how Awdisea is able to help you overcome the major challenge in growing your channel.

By breaking down the conscious and subconscious thinking process that goes into coming up with an idea for a video and creating parameters on a granular level, it is possible to feed this data into a neural network that AI models are built on and then with fairly good accuracy predict how many views a video will get. The AI model does however need additional channel metrics to make this work. These metrics include things like the number of subscribers you have, how many total views your channel got, how often you upload, the number of likes, dislikes and comments your channel has on average, what is the average percentage of subscribers that watch new videos you post, and a host of other data.

Instead of trying to crack the YouTube algorithm, YouTubers need to crack the thinking process of what goes on inside the minds of really good creators. That is really what Awdisea is all about. It helps YouTubers break down their thinking process into actual data. It then takes this data and predicts how well the video might do based on how well similar videos did - not only from your channel but also from other channels that produce videos similar to yours. Provided that you supply Awdisea with enough relevant data and publish your video using the thumbnail and title that you provided it with, Awdisea will later on check how well the video performed and compare it to what it predicted and then adjust its neural network to improve the accuracy. Initially the accuracy will be off but over time it will improve.

Training Awdisea's AI model does require some human interaction in order to help guide the AI avoid making wrong choices. But over time, human interaction decreases as the model becomes more accurate.

As more YouTubers who produce content similar to yours use Awdisea to train the model, predictions become even more accurate. For this reason, it is strongly recommended that you reach out to similar channels and tell them about Awdisea and get them to use it as part of their pre-production.

Some YouTubers are in a financial position where they can afford to produce a short clip of the video without having to produce the entire video. For those in this position, it would be extremely advantageous if this clip could be reviewed and feedback provided not just to the AI but also the YouTuber who produced the clip. This is effectively what Walt Disney was doing. The ideal way to do this is to produce only the first minute of the video and make it privately available to only a portion of your subscribers. These selected subscribers must not in any way be persuaded to watch the clip. That defeats the purpose of the test. The test is to see whether they are attracted to the thumbnail and possibly the title, to be engaged enough to want to watch the clip. If they do choose to watch it, the amount of time they spend watching it is recorded. This is one of the most important metrics YouTube Studio shows on videos. You want to know at what time in the video users start dropping out.

Unfortunately YouTube doesn't offer this kind of test environment. Some YouTubers get around it by showing preview clips in advance to paying Patreon members. While Awdisea currently does not support enlisting a channel's subscribers, it is on the list of TODO things that will eventually be available. The reason you want input from your subscribers is because the YouTube algorithm places important value on how your subscribers react before suggesting it to others. If you can't get enough of your own subscribers to watch the video, the algorithm isn't going to bother recommending it to anyone else.

Ultimately, Awdisea is about an AI whose only purpose is to figure out the thinking process that went into deciding what video to make and tracking whether that thinking process resulted in a very good video performance. It then lets other YouTubers who produce similar content to tweek their own thinking process along the same lines as those YouTubers who made better videos. It isn't about duplicating someone else's videos but rather about adopting parts of someone else's way of thinking to help improve their own videos.