My previous post about BookBub concerned the type of ads book bub accepts. In this post I will look at performance, focusing primarily on discounted books in preference to free downloads. Free downloads might have such goals as generating reviews for the book, establishing a fan base for the author, or encouraging readers of the free book to purchase subsequent works in the series. A heavily discounted book might have any of these goals as well, but may also expect to repay advertising costs through increased book sales. BookBub of course cannot guarantee that ads will generate sufficient sales to earn back the costs, but from the analysis below, I think they typically do. I do not herewith endeavor to contradict BookBub’s published sales predictions, but rather to expand upon them.
See the footnote  for a few words about the limitations of my methods. Because my estimates of sales are speculative, I only present these estimates to compare factors such as book price, book rating, number of reviews, day of week, ad’s position on the page, etc. So do not consider them for any other purpose than entertainment. In addition to the method limitations mentioned in the footnote, sometimes authors run ads in other media on the same day as the BookBub ad. For example, Amazon has Countdown Deals and Free Promotions that might coincide with a BookBub ad. While this is rare, it could make a particular BookBub ad appear to be much more effective than it would have been alone.
Fifteen books – one day:
Let’s first look at a typical day. Here we see book rank for the first fifteen books on the Latest Deals list for 8/3/2014.
Note: Rank is plotted on a logarithmic scale since ranks here vary from #1 to #119,553. On a linear scale most of the data on the graph would just be clutter at the bottom.
The time axis runs from 04:00 Pacific time on 8/3 through midnight on 8/4. The Latest Deals page typically updates between 05:00 and 06:00 Pacific. Although the ad will disappear by 06:00 on the following day, I continue to track the books’ ranks for the rest of that day.
The free books often change rank before the paid books. This may be because Amazon updates free and paid ranks on a different schedule. Some books respond soon after the ad appears, but most take many hours, usually not until around noon Pacific time. This delay could be due to Amazon’s schedule for updating ranks. The early response of some books may occur because the book is already being promoted elsewhere and is experiencing sales/downloads by the time the BookBub ad goes live.
Return on Investment:
As mentioned in a previous post, this is calculated for paid books as:
ROI = (Revenue generated by ad – Cost of ad) / (Cost of ad)
A positive number for ROI means that revenue from increased book sales more than paid for the cost of the ad.
Revenue is calculated from the book’s price and the estimated sales volume, and by applying Amazon’s Kindle Direct Publishing percentages to calculate royalties. Revenue is thus heavily dependent on price, with a $2.99 book bringing in royalties per book that are six times that of a $0.99 book – twice the royalty percentage and three times the price.
Most books in the Latest Deals are either free or $0.99, and more than half of sales go to the $0.99 priced books:
The median $0.99 book does however appear to still make a profit.
For reasons that baffle me, Amazon’s numerical star ratings seem to have no discernible affect on a book’s popularity as evidenced by its Amazon rank. A commenter on a previous post suggested that a high rank may be more a reflection of having only a limited and appreciative readership, and that the same book would achieve a lower rank if it only had a wider audience.
There is a peculiar inverse correlation between rank and volume of sales resulting from a BookBub ad.
Because we already know that higher priced books have a much stronger ROI, let’s break this down.
Rating seems not to matter for the $0.99 books, but perversely seem to correlate negatively with sales of $2.99 books. Forgive me for still doubting that low ratings and high prices are a sure path to success.
Number of Amazon reviews:
There is only a weak positive correlation.
The correlations are still very weak. It does appear though that for the $0.99 books, the number of reviews seems to matter the least. Without the few outlier successes at the other price points, I’d be inclined to think that the number of reviews doesn’t matter for these price points either.
Distribution of ROI:
So far I have only been showing median ROI, which has always been positive, i.e. the ad was profitable. Graphs have started at 0% ROI, hiding points below the line. While almost any grouping I looked at had a positive ROI in aggregate, not all books fell on the profit side. There have been wildly successful outliers, and there have been books that my analysis suggests probably lost money on the ad.
Red = Probably did not earn back the full cost of the ad
Grey = Due to uncertainty in my methods, might or might not have broken even.
Black = Probably earned back the cost of the ad plus up to 50% profit.
Green = Earned back the cost of the ad plus between 50% and 100% profit.
Bright Green = Earned between two and three times the cost of the ad.
Gold = Earned more than three times the cost of the ad.
Keep in mind that these numbers could be off in either direction. Like a lottery, one at least can’t loose more than the cost of the ticket but the payoff could be big.
This is the most surprising feature I discovered, although perhaps should have been obvious. The first page of Latest Deals lists fifteen books, and below are book characteristics and performance by their order on that list.
The lower sales of books further down the page may have nothing to do with the need to scroll past other books. BookBub may place the books it accurately perceives as most salable at the top.
The books at the top seem to be ones with higher royalties per book – that is, they are at a higher price with consequently higher royalty percentages, and thus earn more per book sold.
The net result (ROI by order on the page) is striking.
Again, this may have more to do with BookBub deciding which books to put at the top of the list than with a special advantage to being the first or second book listed.
Day of the week:
Not much news here.
It would take a lot longer than three weeks to see a trend.
There is really only one metric: Number of downloads per advertising dollar. Since my measurements are calibrated using BookBub’s predicted download volume, I don’t have much else to add other than the distribution of results. Below is how free books tended to be placed on the list, first by free versus paid, and then by genre.
As with discounted books, there was a distribution of downloads/$.
In my sample, no ad generated less than 65 downloads per advertising dollar, with an average of 108 and median of 94. In my imagination there is a cohort of visitors to BookBub on any given day who download everything that’s free, and these obsessive downloaders are why no free book goes undownloaded.
The highly successful outliers in the above chart tended to be New Adult & College Romance titles, which fell in the 9-12 spots on the list – summer reading? Perhaps BookBub could charge more for free NA&CR ads during the summer.
I did not see an advantage for free books appearing in the first (rare) or second (frequent) spot.
Again, I feel I must remind the reader that these sales numbers are just estimates based on the speculative methods listed in the footnote as applied to three weeks of Amazon book ranks. Entertainment purposes only, as they say.
 For twenty-one days (not contiguous between 7/17/2014 and 8/10/2014) I noted the top fifteen ads on BookBub’s Latest Deals, and then checked the Amazon rank of those books hourly. (When I say, “I checked” I mean an automated script checked.) My estimates of book sales are based on how this rank changed during the day of the ad. This has some large uncertainties. Amazon book rank, which typically peaks (reaches its lowest value, indicating higher sales) for a few hours after the ad appears, is not a direct metric of sales, but a simple ranking based on Amazon’s proprietary sales metric. Although the underlying metric is secret, if a book sells steadily, its rank can be used to gauge sales volume. Many authors have disclosed their sales and rank experience from which I have (as have others) cobbled together a formula (actually seven formulas that cover different rank spans) for estimating sales. Fortunately rank responds crisply to sales although with an unpredictable delay — within hours. After the sales stop, rank gradually decays (becomes higher, indicating fewer sales) and may take months (years?) to reach the doldrums.
I calculate the starting daily rate of sale from the book rank on the morning the ad runs, assuming sales are in steady state. I then calculate a higher sales number from the lowest rank the book achieves during the day. Obviously since the formula was derived from data on steady sales, and Amazon’s sales metric includes other factors, its application to this temporary improvement in rank will not cleanly give the incremental sales that day. It should however be proportional. I thus apply a fudge factor that I will describe in a moment.
I compare the various calculated sales bumps over the twenty day period, and compare those by genre with the anticipated sales numbers that BookBub publishes. This gives me the aforementioned fudge factor, which turns out to be around 2.5 – that is, the actual sales from all channels (not just Amazon) is roughly 2.5 times the value the formulas give from the improvement in rank.
Some in the author community have reported starting rank, best rank, and actual sales during the day of an ad. These data give fudge factors that are in the same ballpark, which is reassuring. I calculate a similar fudge factor for free downloads, which also is roughly comparable to what is reported by other authors about movement of Amazon rank and download volume.
As a way of validating this method I compared BookBub’s average sales predictions by genre, with my observations. My measured sales should be both proportion by genre and similar in magnitude to BookBub’s sales predictions. There will of course be noise because of my small sample size and the particular fortunes of the books offered.
As the reader can see, my method gave roughly the same sales volumes. I was surprised, given the small sample size, that my measurements came out so close to their predictions.
The take-home point is that I am trusting BookBub at its word on average sales and calibrating accordingly.