Did Amazon Change Algorithms to Protect Their Readers?: The Ten Thousand Gatekeepers of KDP Select

Anyone who follows this blog knows that last May, Amazon drastically changed their popularity lists (available on the left sidebar of the main Kindle store) to change the way free downloads were factored into the ranks. On last week’s Self-Publishing Podcast #42, I was asked whether this change was done in order to present readers with better books.

The short answer: yes.

The longer answer: not necessarily better books, but certainly more profitable ones. That’s a very important distinction to make right off the bat. In all media, there’s an ongoing, centuries-long debate about whether a work’s value is based on its commercial appeal or its artistic qualities. As it turns out, I have nothing to contribute to that debate. So what follows should in no way be taken as a judgment of books that have failed to thrive under the recent Select model. Some of my books did worse as well.

But here’s what we know. Between the birth of the Select program in December 2011 and mid-March 2012, all it took for a book to hit the first few pages of its category after a free run was a few hundred downloads. 2000+ would essentially guarantee you’d be near the top of your category, probably for 2-5 days. Because a free download was weighted the same as a paid sale. And very few books are currently selling hundreds of copies per day on Amazon. Right now, about 1000 sell 100/day. Maybe 500 sell 200/day. And only something like 100 sell 500+/day. The numbers were a little lower a year ago, but not by all that much.

Meanwhile, every day, freebie aggregate blogs were pointing their readerships toward several dozen free titles. The biggest blogs had tens of thousands of subscribers, more or less guaranteeing every book featured would pick up at least 1000 downloads. There was some level of curation involved–covers had to be at least halfway decent, and there was typically a rating threshold of some kind–but the blogs had no real way to test the commercial potential of the books they mentioned. And when a book is free, the resistance to downloading it is much, much lower than when that book has a price tag attached to it.

The result is that a lot of books with lower commercial appeal wound up displacing books with higher commercial appeal. On Amazon’s popularity lists, 1000 free downloads beat 100 paid sales, and new Select books were picking up thousands of free downloads every single day. The gatekeepers weren’t strong enough to keep out the low-appeal books, meaning readers were less likely to buy the books in front of them or to be satisfied with the titles they did purchase.

What was the solution? Well, Amazon wasn’t about to start curating these books themselves. Amazon is all about letting massive numbers of consumers reach their own decisions, proving in the most meaningful possible fashion which books have the highest commercial appeal. So some churn of their lists was probably a good thing, as it broke up the stagnation of long-term bestsellers (by the way, the iBookstore is currently struggling with this problem) and presented more voracious readers with fresh material. But this was too much, and it was too unregulated.

The answer was to raise the standards for which books would get prime placement. And in typical Amazon fashion, they would tie that standard to consumer behavior.

In March, they started testing new popularity lists; in May, there was a new algorithm. The winner no longer weighted free downloads equally with paid sales, but at something near a 10:1 scale. And instead of weighting the last 1-7 days of sales + downloads, it looked at the last 30.

So instead of needing 2000+ downloads to land high on the charts–a number most decent-looking books promoted by the top sites could cross; about 100 free books managed that number of downloads per day–their new formula required somewhere between 8000-20,000 downloads to really hit it big. The more niche or iffy books couldn’t hit those numbers anymore; fewer than twenty per day could climb those heights. With exceptions, the only books that could rake in that many downloads were the ones that would have guaranteed commercial appeal when plunked in front of readers. The gatekeepers–readers, making their download decisions one click at a time–were made stronger.

They crowdsourced commercial appeal. In the environment of the time, one or two or three thousand readers downloading a free book wasn’t a terribly accurate predictor of that book’s potential. But if you upped those numbers ten times over–to ten or twenty or thirty thousand reader downloads–you had a much more accurate barometer for which books would sell when they were awarded with extra visibility.

It was a net gain for readers, who had an easier time finding appealing books, but a net loss for writers, fewer of whom could pull in the number of downloads required to hit the jackpot.

Again, I’m presenting this without judgment. A book’s surface appeal, which prompts free downloads, doesn’t necessarily represent its deeper appeal, which prompts word of mouth and long-term sales (to say nothing of literary or artistic appeal). And Amazon’s current algos aren’t perfect. Certain factors–the readership demographics of the major blogs, crossover appeal of the larger genres, Amazon’s categories, etc.–means that certain subgenres (romance, thrillers, etc.) have an easier time of it than more niche subgenres (epic fantasy, Westerns?, etc.).

This is just my narrative of what happened. Amazon’s standards/algos weren’t high enough to deal with the emerging free book market; the rewards for making your book free were disproportionately high compared to their average commercial value.

So they raised their standards. And a lot of authors were left scrambling for a new solution.

 ~

The silver lining to these changes is that we as authors can take advantage of the raised standards to gauge the appeal of our own books. But since this post is already closing in on 1000 words long, I’m going to tackle that in a followup.