I can be long-winded in my blogs, I know, and there is a lot to unpack here. I’ll try to keep it brief. Famous last words =) Any Netflix engineers reading, it will be worth your time even if you skim fast.
In the last month there has been global news coverage about Netflix losing subscribers (TechCrunch, BBC, More…). There are a lot of reasons this is happening, and MakeUseOf (MOU) gives six good reasons for this. What surprises me is that they don’t include a seventh reason that I know first-hand has driven customers away (e.g. my dad), and I haven’t read about in articles. I’m sure there are some blogs or articles that cover it, but I think I will cover it in a different way with some solid numbers.
As always, “statistics” yay, but they must come with caveats every time and this is no exception. I can’t even tell you how long I have had a Netflix account, how many movies I have watched, how often I liked a movie on that platform that I have seen elsewhere (important for this blog), and a number of other little bits along these lines. Since I will be calling Netflix’ algorithms out, I must acknowledge factors that I certainly hope are part of how it operates. I have serious doubts, but I still hope. Also, some of what I call out has nothing to do with the “recommendation algorithms [that] are at the core of the Netflix product“. Simple numbers and logic will prove that clearly, and is the crux of my argument.
According to one blog, citing uNoGS, “Netflix has over 17,000 titles globally as of April 2022“. Supposedly, as of 2020, Netflix had 2.2 million minutes of content for the Netflix US market. That is certainly a lot of titles and a lot of material to binge watch. Now, are those statistics true? I say who cares, it is completely irrelevant because Netflix does its absolute best to shove the same tiny fraction of content on you, over and over. Even as the available categories have expanded and evolved, their algorithm’s ability to repeatedly offer the same titles has only gotten worse.
The other night I noticed that they had a new feature that came up after loading in, giving me five “Daily Picks”. This was a reminder that I had taken notes and pictures previously about the same problem via the categories. What’s the issue here? One of the picks is on my ‘Watch List’, one I have seen off Netflix (but thumbs-upped on the platform), and two I have seen on Netflix. Why is it suggesting three of five titles that I have seen before? Why keep trying to get me to watch the same thing over and over?
What I wanted to do all along was actually count just how many times Netflix recommended content I had seen before. The platform certainly knows what I watched via their service. If I didn’t watch something on Netflix but gave it a thumbs up, it stands to reason I have seen it. I do that to help its algorithm recommend more new material, not keep giving me the same titles.
By tracking a few simple numbers, it gives a good idea of how bad the problem is. Tracking the total number of titles in a category, how many I have seen, how many I gave a thumbs up, how many got a thumbs down, how many I had seen on the Netflix platform, and how many were already on my watchlist, it paints an interesting picture. Back to disclaimers: you may notice I rarely give a thumbs down. Why? Because the Netflix algorithm is unknown. I may not like a movie because it was just bad, poor writing, poor acting, cliché, or a number of other reasons. If I give it the thumbs down, what part of the movie does Netflix consider in making future decisions? A thumbs up or down is too binary, and now we have thumbs down, thumbs up, and double thumbs up, but it still isn’t enough. For a brief period they had a five-star rating system that went away fast, which is sad.
After all that, consider when Netflix offers me foreign content. That is a turn-off to many, but as someone who watches with subtitles on just about everything, I am fine with it. If I give it a thumbs-down though, how does Netflix interpret that? What about the same movie that appears on multiple categories? As I write this, ‘The Gray Man’ (a Netflix original) appears on at least four out of nine categories as I generate my statistics. ‘Legacy of Lies’, which was sold to Netflix UK, appears on more than four categories; I assume because they just acquired it but I would be wrong, as the rights were acquired by Netflix US in October 2021. It’s hard to understand that part of the recommendation algorithm.
With that last bit, I just confirmed something I suspected about Netflix myself; that it operates as different business units. It isn’t one big company buying rights to content outright, it has limitations. I learned this early on when ‘V for Vendetta’ suddenly disappeared off my ‘Continue Watching’ list after October 31, and re-appeared on December 1. They apparently bought the rights, but that did not include airing the movie in November. Weird.
Anyway, back to how bad their recommendation algorithm really is. I went through every category between 7/27 and 7/28 and took note of a handful of numbers, and corresponding percentages:
- Category – The Netflix category name
- Total – How many movies in the category
- Seen – How many have I seen (regardless of platform)
- Thumbs Up – How many received a thumbs up in the category
- Thumbs Down – How many received a thumbs down in the category
- Seen on Netflix – How many were seen on Netflix specifically
- On Watch List – How many already exist on my watch list
- Total Ack’d – By %, how many have I thumbs-upped or watch-listed
For the purpose of my argument, you can focus on the total movies in a category, and how many I have acknowledged, to Netflix, in some manner. I have either given it a thumbs-up indicating I have seen it (it shouldn’t matter which platform) or have indicated my desire to see it by adding it to my watch list.
In one category, I have seen or told Netflix I want to see 77.5% of the movies offered. That is a bizarrely high and absolutely ridiculous amount of content they are recommending I see… again. Next highest was 75% and then 60.6% and 60%. This isn’t a fluke at all; category after category they do this so ultimately 7 out of 34 categories I have seen at least half the content in. Supposedly 2.2 million minutes of content, yet they want me to keep watching the same thing, why?
I don’t care what the reason is, I just know the system is broken. If you are wondering about them recommending material by “% match”, which is part of their algorithm, I say no. There are movies that are a 50% match or lower in the recommendations. Surely all those minutes of content aren’t below 50%, right? If so then their weighting and percentages are laughingly bad.
One way or another, Netflix needs to improve. They need to realize that they are their own worst enemy when customers were leaving them for this reason before they had serious streaming competition.
(Here is the raw data I generated.)
4 responses to “Netflix: Why People Are Leaving You… (The Unspoken Reason?)”
Spot on – nice analysis. We are still with them as there are certain ongoing franchises we like to keep up with, but are far less likely to use them for movies. We have been continuous members since 1996 and miss the old disc days. I can’t tell you how many great indy films we saw because their old 5-star system was good at matching us with great films. Thumbs up and down is just too “fat-thumb” annoying. And now that they produce their own material, they always push that ahead of other good non-Netflix items. I can’t tell you the last time we found a decent indy through their recommendations. I am not certain that they license them as they don’t see the cost/benefit. In the disc days, having a few copies of a less popular film was a negligible cost for them. Let’s see if they learn the right lessons from their subscriber losses – I doubt it – MBA’s don’t have a clue what normal people seek.
They currently have down / up / double up. But the last time I saw more than two options it went away fast. Not sure how long this will stick around.
If I downvote “Final Score” with Dave Batista, does that mean Netflix would not recommend “Die Hard” even though they are essentially the same movie 20 years apart? Curious about their algorithm.
Curious, Martin McKeay has the same gripe with Kindle recommending books he has read. What gives? https://twitter.com/mckeay/status/1560429095204392960