Your users stopped visiting your product. Here's where they went — Prathik Roy (Springer Nature)

Prathik Roy is Product Director for Data and AI Solutions at Springer Nature, one of the world's largest academic publishing companies. A quantum chemist and material scientist by training, he spent years in R&D before gravitating towards product management — and has spent the past 12 years helping publishers understand the value locked inside their content. In this episode, Prathik makes the case that publishers are sitting on some of the most strategically valuable data in the world, and that most of them are only beginning to understand what that means in the age of AI.


In this episode, we cover:
(00:00) Introduction: from quantum chemistry to product management
(05:00) The Schrödinger problem: why content value is increasingly unknowable
(08:00) How traditional publishing metrics worked — and why they broke
(11:30) The ChatGPT moment and its impact on scientific publishing
(15:00) Paywalls, subscription models, and the shift to data licensing
(21:30) How scientific content earns its quality — and why AI cannot just follow the citations
(26:00) Why AI developers want bullet points — and what that means for content structure
(29:00) New monetisation models: tokens, outcomes, and data as a service
(33:00) Rights management: rights in, rights out, and why the prohibited section matters
(36:30) Measuring content value when your users live inside AI systems
(38:00) What to do with your content archive: extraction, licensing, and prediction markets

Key takeaways
— Value has shifted from visibility to synthesis. For 20 years, publishing was built on an assumption that value is visible — clicks, downloads, page views. AI has broken that model. A scientist can now get a synthesis of 300 papers without opening a single PDF. Publishers who still measure only what happens on their own platforms are measuring the wrong thing.
— Peer review is a moat — if you know how to use it. The quality signal embedded in scientific publishing (acceptance rates, citations, expert validation) is precisely what AI developers need and cannot easily replicate at scale. That is the leverage point. Publishers who understand this can command significantly higher licensing value than those treating their content as a commodity.
— Structuring for machines is now a first-order problem. AI developers are asking for content formatted with bullet points and question-and-answer structures — formats that run counter to how scientific articles have been written for centuries. Publishers who adapt their content architecture for machine consumption will capture more value from AI pipelines than those who do not.
— Rights management is no longer a legal afterthought. If you are building a data product, understanding rights in and rights out is as important as building the product itself. Prathik's rule: if the prohibited section of your agreement is not longer than the rights section, you are probably doing something wrong.
— Usage-based and outcome-based pricing are replacing subscriptions. The next revenue frontier in data-intensive publishing is not flat subscription fees — it is pricing tied to how content is consumed inside AI systems and the outcomes that consumption generates. Token-based models and revenue-share arrangements are already emerging in scientific publishing.
The interface is disappearing — and metrics must follow. Product managers need to build pipelines that track how content flows into AI systems, how often it is retrieved, and whether the outputs generated are traceable back to the original source. Session time and page views will not tell you what you need to know.
Hidden value is sitting in content archives right now. Podcast transcripts, audio, video — all of it contains extractable IP that can be packaged, enriched, and licensed to financial firms, research companies, and AI developers. Extraction and enrichment is the first step; licensing and outcome-based models follow from there.