“AI could be a game changer for the way we can market and sell books,” says Sara Lloyd, global head of AI at Pan Macmillan, one of the “Big Five” publishers in the U.K. and home of international author brands including Ken Follett, Julia Donaldson, and Elton John. “Machine learning and automation, now combined with the power of natural language processing, opens up a world of possibilities which will enable the relatively smaller marketing resources of most publishers to punch way above their weight.”

It’s a punch that’s already landing pretty hard. Business is booming for U.K. publishers, despite perennial announcements that the toxic cocktail of social media and streaming is about to render their ancient product obsolete. Nielsen BookScan reports that last year the U.K. book market pulled in a record $2.27 billion (£1.83 billion), the highest revenues since accurate records began, a coup fueled by blockbuster releases from Britney Spears and Prince Harry, whose memoir Spare is the fastest-selling nonfiction book in history. In contrast, 2023 was a tough year for both tweetstorms and box sets, with media executives claiming we had reached “peak TV” while Elon Musk continued his consistently impressive job of tanking X.  

Monetizing the backlist

In this landscape, the advent of mature, generative-AI-enabled marketing tools could be just the boost publishers need to reclaim minds—and money—from their shinier rivals, and they have a potent stealth weapon at hand: their backlists. Pan Macmillan, which was founded in 1843, has two centuries’ worth of old prize winners, ex-bestsellers, and underrated gems just waiting to be rediscovered by BookTokers. The other four big hitters—Penguin Random House, Hachette, Simon & Schuster, and HarperCollins—have similarly rich archives to leverage. 

“Over 100 million [book] titles have been published in English alone,” says Nadim Sadek, the founder of Shimmr AI, a London-based “AI-native” advertising platform which, since launching six months ago, has netted sign-ups from more than 120 publishers, including both members of the Big Five and smaller independent presses at a rate of around one per working day. “No more than 5% of these have ever been advertised. Accordingly, they are largely undermonetized.” 

Courtesy of Molly Flatt

Matchmaking tastes and titles

Shimmr uses a three-step process to turn a simple PDF or ePub file into that read you can’t resist. First, “the analyzer” employs a variety of large language models (LLMs) to define the work’s structural patterns and psychological values, or “BookDNA.” Then “the generator” employs an array of diffusion models and other image generators to produce images and copy closely allied to that BookDNA. Finally, “the deployer” shares the content across media channels where readers with matching psychological dispositions are active. 

“Reward models ensure that the campaign is continuously self-optimizing, searching for the most effective combinations of BookDNA, images, and copy, in various channels and against different audiences,” Sadek explains. “Readers, by being matched with more psychologically suited books, also enjoy more fulfilling reads, positively reinforcing their reading (and buying) habits.”

The results are already impressive. “Within only six months of our launch, Shimmr has doubled the ROAS (return on ad spend) of Google’s average benchmark,” reports Shimmr’s chief revenue officer, Brooke Dobson, “and we’re on track to improve this by at least 3 times with Shimmr’s self-optimizing system. In addition, titles on Shimmr see an average of 35% uplift in sales.” 

No wonder, then, that the company has attracted some heavyweight publishing investors and advisors, including former Macmillan CEO Richard Charkin and former MD of Penguin Random House Markus Dohle. 

Balancing innovation with ethics

But incumbent leaders in the industry are still wary about owning up to experiments with AI, and for good reason. Dozens of writers, including John Grisham and George R.R. Martin, are currently suing OpenAI and Meta for training LLMs on copyrighted works without permission or compensation, while copycat AI-generated content threatens to erode already-paltry author and illustrator incomes (an average of around $8,700 [£7,000] a year, at last count from author licensing body ALCS). 

Lloyd, so far the only U.K. trade publishing executive with “AI” in her title, is well aware of the tension between innovation and ethics. “Publishers have to navigate two sets of parallel needs: on the one hand, to innovate and leverage the benefits of AI to make us more effective in selling our authors’ works, while on the other hand, lobbying and pushing hard for the use of their copyrighted material to be respected, recognized, and compensated.”

Sadek, too, is keen to emphasise the ways in which Shimmr safeguards the creators it serves. “Books are hermetically sealed away from the training sets of [our] models,” he insists. “Each execution is built up from an initial single pixel, informed by the BookDNA. No pre-existing imagery is used as a starting point.”

However, Lloyd is also bullish about the need for book trade leaders to proactively embrace and shape AI. “Acknowledging AI’s reach and influence—and the power it is likely to wield in the future—Macmillan (and our parent company the Holtzbrinck Publishing Group) recognized that a centralized role and cross-functional view was vital,” she explains. “The potential of AI to disrupt means the publishing industry cannot risk being left behind.”

So next time you’re browsing for a fresh series to binge on and find yourself buying a decades-old novel that’s eerily well matched to your tastes instead, take a moment to celebrate. You’re helping the publishing industry beat AI at its own game.