Been a while since I did one of these!ﾠ(See Fourth Year Book Sales, Second Year Book Sales, First Year Book Sales, and Aftermath: The Motivation Hacker.) I managed to completely ignore the data for six years and had no idea how many copies of The Motivation Hacker had been sold. This morning, I figure I should probably check.

I spend a few minutes downloading Kindle and Smashwords reports and finding my old spreadsheet that graphs the charts, then swiftly Emacs-macro my way to matching the Kindle and CreateSpace data format. Nice. But then, a roadblock: I face the horrible prospect of spending 10, maybe even 20 minutes cleaning the data to sum the other channels across 7 wrongly-formatted Smashword Excel spreadsheets. 20 whole minutes!ﾠUnacceptable.

Well, maybe this will be a good opportunity to try out ChatGPT's Code Interpreter–I mean, Advanced Data Analysis feature. Enable in Settings, ok, here we go. Upload all the files. Ask it to do the thing: "Make a simple table where the rows are months and there is one column for total qty for channel Smashwords and another column for total qty across all other channels.". Wow, it's amazing! Automatically tries three formats for the spreadsheet until it finds the right one, corrects missing values where there were no sales for a month, makes the table. First prompt is looking magical.

Here's where I nerdsnipe myself. "Let's see if it can reproduce the graph," I think. "That'll be cooler than just pasting the new columns into the existing spreadsheet."

You can see the whole ChatGPT conversation here. 7 prompts and a full dataset reupload later, I have the graph, which would have had wrong numbers for the totals if I wasn't, you know, making the same graph myself to check its output. I ask it to do one last step to summarize sales by years-since-publish instead of calendar years, but I had left the computer and it had forgotten my file uploads.

Let's take a look. Here's the manually created graph:

And the AI-created graph:

So, what have we learned?

• I'm still faster and more accurate doing basic data processing and graphs for tasks that just need normal spreadsheet munging.
• ChatGPTﾠAdvanced Data Analysis is probably better for things that need more complicated calculation and more data cleaning but less specialized graphing.
• You have to be patient with transient errors (responses stalling halfway, needing to reupload the dataset).
• You have to check its work.
• It's fascinating to watch it go.

Oh right, about the book. We are trying to learn about the book.

• There was a big sales spike in June 2018. It'd be archaeology at this point to find out why, probably some prominent recommendation.
• Otherwise, it's been a slow and steady decline in sales rate.
• We're down from initial 4.29 to 3.72 on Goodreads and from 4.9 to 4.1 on Amazon. The Amazon rating is a bit inflated; Goodreads is more accurate, in my opinion.
• 16,560 copies at $2.21 royalties per copy is $36,597.60, or about $188/hr for the 195 hours I spent writing and editing and 0 hours I spent marketing it.

Sometimes I get to thinking, that was fun, and imagine what it would be like if I spent a lot more effort and tried to make a great book, rather than a good and easy book. And I could promote it! Then I remember that I'm running a startup and have three kids, and also, if I'm going to do a side project, coding is generally more fun and profitable than writing. So, the only books I'm making these days are one-offs, like the print-on-demand AI-generated Among Us choose-your-own-adventure novel I made Max for his 8th birthday, or the consolidated-from-nightly-bedtime-stories Adventures of Ruby Winter sci-fi epic. Perhaps I'll post more on those and other AI projects later.