Your beta reader loved chapter seven. Your critique partner thought it should be cut entirely. Your writing group had three opinions and a heated side argument about semicolons.
This is the fundamental problem with manuscript feedback: it’s subjective. Good readers disagree. Good editors disagree. Agents pass on books that become bestsellers, and acquire books that quietly disappear. There’s no blood test for “ready to publish.”
Timo Bozsolik-Torres spent years thinking about this problem from two very different angles. By day, he was a machine learning engineer at Google, working on Kaggle (the world’s largest data science platform) and steeped in the question of how algorithms learn patterns from massive datasets. During paternity leave, when sleep became theoretical, he wrote a debut thriller called The Scaevola Conspiracy and discovered just how opaque the path from finished manuscript to published book really is.
At some point, the two worlds collided. What if you trained a model not on what critics think makes a good book, but on what actually sells? Not taste. Not theory. Data.
That question became Iris.
A Machine Learning Engineer Walks Into Publishing
Timo’s background reads like someone designed him to build exactly this tool. After launching two startups and a stint at Google working on AI initiatives and Kaggle, he’d built deep expertise in how machine learning models identify patterns in complex data. He’s a Kaggle Master, which in the data science world is roughly the equivalent of making the varsity team at the national level.
But he’s also a novelist. The Scaevola Conspiracy, inspired by the Cambridge Analytica scandal, explores what happens when tech companies abuse market power. He expected the writing to take four months. It took four years.
Along the way, he experienced every frustration debut authors know well: the opaque submission process, the conflicting feedback, the uncertainty about whether a manuscript is genuinely ready or just good enough to start collecting rejections. Agents might spend minutes on a manuscript that took years to write. And the decisions about what gets published lean heavily on individual taste.
As a data scientist, that inefficiency was impossible to ignore. Patterns exist in commercially successful books. The question was whether machine learning could identify those patterns well enough to give authors useful, objective feedback before they ever hit “send” on a query letter.
What Iris Actually Does
Iris comes in two flavors, both focused on analysis rather than creation.
Iris Editor provides editorial guidance across 25 metrics. Upload your manuscript, and it flags sections where your pacing deviates from genre norms, where reading difficulty spikes or drops unexpectedly, where emotional resonance flatlines. Think of it as a diagnostic tool: it doesn’t rewrite your prose, but it highlights the passages that might need your attention, and tells you why based on data from books in your genre.
Iris Assessment evaluates your manuscript’s commercial viability. In a few seconds, it generates an overall score predicting how your book would perform in the market, how closely it fits its declared genre, and how it compares to books that actually sell. It’s the kind of feedback that usually requires hiring a professional manuscript assessor and waiting weeks for a response.
The analysis spans over 30 dimensions: pacing, reading difficulty, emotional resonance, genre fit, sentence structure, and more. Each metric is benchmarked against a dataset of millions of texts connected to real performance data from platforms like Amazon and Wattpad. Iris learned what sells not by studying bestsellers in isolation, but by analyzing the full spectrum (books that thrived and books that didn’t), training on the real distribution of success and failure.
The practical workflow is straightforward. Upload a manuscript, get a report, see where your numbers deviate from the genre average, revise, and run it again. Timo tracked seven major drafts of his own novel through Iris, watching his overall score climb from 61 to 67 as his edits addressed the patterns the tool identified. Other authors have reported improvements of up to 12 points, which Timo says “can absolutely make the difference between a bestseller that gets picked up by a publisher and an ignored submission.”
The AI That Doesn’t Write
In a landscape where nearly every AI tool for authors is racing to generate better prose, Iris is doing something genuinely different. It reads.
That distinction matters more than it might seem at first. Generative AI (tools like ChatGPT, Claude, or Sudowrite) creates text. Predictive AI (what Iris uses) analyzes existing text and identifies patterns. The difference is like the difference between a ghostwriter and an editor: one produces words on your behalf, and the other evaluates the words you’ve already written.
This positioning has earned Iris a kind of credibility that generative tools struggle to claim. NAC Literary, a literary agency that represents authors, explicitly states that manuscripts improved with Iris are welcome in their submission pile, while manuscripts touched by generative AI are not. When an agency draws that line and puts Iris on the acceptable side, it says something about how the publishing industry perceives the tool.
It also addresses a concern many authors carry about AI: the worry that using it means the writing isn’t really theirs. With Iris, every word is yours. The tool helps you see your manuscript the way data sees it, but it never puts a single word on the page.
The training data is worth noting too. Timo built the model using donated manuscripts, works under permissive licenses, and publicly available sample chapters. The output is purely numerical, which prevents the system from reproducing any training material. For authors worried about the copyright concerns that shadow generative AI, that’s a meaningful distinction.
And the industry is paying attention. The first publishing deal for a book enhanced by predictive AI was signed on Christmas Eve 2024, with a second following in June 2025. Both used Iris. Publication details haven’t been announced yet, but the fact that publishers are signing books that openly credit predictive AI analysis in their development process is a small, notable shift.
What You Should Know Before Signing Up
The pricing isn’t publicly listed. Quantifiction offers a free trial and what’s described as a “small annual subscription” for ongoing manuscript analysis with multiple passes. But the exact cost isn’t displayed on the website. You’ll need to sign up to see current pricing, which is a real friction point for authors who want to comparison-shop before committing.
It tells you what, not how. Iris can flag that your pacing in chapters four through seven deviates from genre norms. It can’t tell you whether that’s a problem or a deliberate creative choice. As Timo puts it, Iris “can’t judge whether a potential problem is a real problem.” You still need editorial judgment to interpret the numbers. This is a diagnostic tool, not a treatment plan.
It’s web-only. No desktop app, no mobile app. You need a browser and an internet connection.
The user base is still growing. With over 2,000 subscribers, Quantifiction is a smaller operation than many established writing tools. That’s not necessarily a drawback (small teams can move fast and stay responsive), but it means the community and support ecosystem are less developed than what you’d find with larger platforms.
It’s a specialist, not a Swiss army knife. Iris does one thing: manuscript analysis. It doesn’t help you outline, draft, worldbuild, or manage a series. If you’re looking for an all-in-one writing platform, this isn’t it. You’d use Iris alongside your existing writing tools, not instead of them.
The Bottom Line
Iris occupies a category that barely existed a few years ago. While most AI tools for authors are trying to help you write, Iris is trying to help you understand what you’ve already written: how it reads, how it compares to your genre, and whether the data suggests it has commercial legs.
If you’re preparing a manuscript for submission to agents or publishers and want a second opinion that isn’t shaped by personal taste, Iris offers something genuinely different. It’s especially useful for authors deep in the revision process who want an objective benchmark, or for self-published authors gauging commercial potential before investing in cover design, professional editing, and marketing.
If you’re early in the drafting process, or if you want AI that helps you write prose, Iris isn’t the tool. And if you’re philosophically opposed to reducing creative work to numerical scores, that’s a fair position. Not every author wants their novel scored like a term paper.
But for the author who’s finished a manuscript and is staring at the submit button, wondering if it’s truly ready, Iris offers a kind of answer that didn’t exist before: not an opinion, but a measurement.
