Your Tool for Writing Best-Selling Novels

Want to write a best-seller? Together, you and I will figure out the secret.

As I’ve posted before, data scientists have developed software that can predict whether a novel will be a best-seller. These text-analyzing tools are about 84% accurate, but only work when they have text to analyze. That is, you have to write the book first, and then run it through the computer. Not super helpful.

Not available. Anywhere.

We can do better than that, at least as a mental exercise. W. Somerset Maugham is reputed to have said, “There are three rules for writing a novel. Unfortunately, no one knows what they are.” Let’s figure out the rules Maugham said no one knows, but not just for novels—for best-selling novels.

I don’t know about you, but I’m not a data scientist and don’t know a thing about creating text-analysis software. I don’t have the time or spare cash to buy all the best-sellers for the last few years and input all that text into a computer.

Still, we won’t let minor details stop us. In fact, since we’re creating imaginary software, we’re free from bothersome facts that constrain real data scientists. Our best-seller writing rules needn’t involve things that computers are good at counting. We can come up with any rules we want.

Let’s start with a name for our imaginary software tool. Perhaps Best$ell 1.0. Not very good, but we’ll use if for now and get our Marketing Department to work on a better name.

Let’s imagine a list of attributes that Best$ell 1.0 will use. We’ll use this list for starters:

  1. Luck
  2. Amount of promotion of novel
  3. Appeal of book cover
  4. Existing fame of author
  5. Appeal of main characters
  6. Difference from other novels
  7. Addresses a current or emerging topic
  8. Addresses a controversial or taboo topic
  9. Amount of sex or violence
  10. Quality of prose

You could come up with different attributes, but that list should be okay for Version 1.0. Let’s say Best$ell 1.0 can easily measure all of those attributes. Let’s also say a greater amount or degree of any of those attributes gives a manuscript a better chance of becoming a best-seller.

Looking back over our list, I see one problem. Some attributes are beyond the author’s control. The first one depends on chance. The publisher controls the second one, for the most part. Attributes 3 through 5 depend on reader reaction. Attributes 6 through 8 depend on society in general.

I put the list in rough order from least author control to most author control. The author has some influence on all the attributes except number 1, but has greatest control over the latter items in the list.

Moreover, not all the attributes would be equally important. Best$ell 1.0 would know the weights to assign to each attribute, of course, and it may well be the last items in the list outweigh the first ones. That would give the author greater influence.

Of course, that last attribute might be fully under the author’s control, but it’s not a very actionable attribute. How, exactly, do you write high-quality prose?

Well, it looks like Best$ell 1.0 has a few bugs and isn’t ready for release. But it’s a start. The next version will be much better, given the talent and expertise of our top-notch team: you and—

Poseidon’s Scribe

Best-Seller Foreteller?

What if a soothsayer could tell you if your manuscript would become a best-seller? If you were a publisher, you’d hire that soothsayer, right?

Throughout the history of the publishing industry, editors and publishers had to make buy-or-reject decisions based on experience and gut feel.

Welcome to the Age of Big Data.

Crystal ball image from Wikipedia

According to an article in The Telegraph , researchers at Stony Brook University used computers to analyze writing styles and could predict whether a book would be successful with up to 84% accuracy.

Following up on that, Jodie Archer and Matthew L Jockers wrote The Bestseller Code, a book about their algorithm (the “bestseller-o-meter”) that analyzes character, plot, setting, style, and theme to make its predictions. According to an article in BBC Culture, this strangely named algorithm is also highly accurate.

More recently, I read an article in BuiltinAustin about a company in Austin, Texas called AUTHORS.me that has developed their own algorithm, StoryFit, which they market to publishers.

These algorithms chew on massive amounts of data—thousands of novels—and perform statistical analyses. After being given test data about past novels for which the success or failure results are known, the algorithm “learns,” or at least develops rules, to distinguish best-sellers from flops. You then apply the algorithm to an unpublished manuscript and make a reasonable prediction. A crystal ball for novels.

Could this lead to a world where publishers reject your manuscript because their algorithm said it wouldn’t sell? Or a world where authors could edit their manuscript to add in the aspects such algorithms judge to be indicative of success? Could the writing and publishing of novels be reduced to a numbers game?

Not quite yet, apparently. The Stony Brook University algorithm struggled to predict the success of books in one genre—historical fiction. Also their algorithm “predicted” Hemingway’s The Old Man and the Sea would flop. Archer and Jockers’ bestseller-o-meter rated The Help by Kathryn Stockett as meh. Further, the novel achieving their algorithm’s highest score (The Circle by Dave Eggers) was a commercial failure.

Certainly, these artificially intelligent systems will improve and get more accurate in the coming years. They’ll identify trends in how the reading public’s tastes are changing. Maybe the algorithms will never be 100% right, and some books they reject will succeed and vice versa. Every now and then, an author tries something new and it sells well despite being unlike the norm. They do call them novels, after all.

As publishers make increasing use of tools that predict a novel’s success, and as authors begin to use similar tools to tune their manuscripts for market success, could it be that overall novel writing will improve? Will that lead to an increase in readership, a renewed clamor for books by the buying public?

I hope so. In the meantime, my new big-data algorithm has just finished analyzing all my previous blog posts, and states there is a 99% probability I’ll conclude this one by signing it—

Poseidon’s Scribe