Zedoc, a new AI-driven metadata platform, is offering publishers the ability to streamline demanding tasks, including cleaning up book metadata, accessibility enrichment, and structuring product data for discovery by AI systems. The platform's current offerings include subject classification (Thema, BISAC, and Dewey), speech-to-text transcription, ONIX-compliant record generation, and audiobook chapterization.

The company is a spinoff of Beat Technology, a Norwegian e-book and audiobook company that provides white label apps to clients in a dozen countries, including for IPG in the U.S., Gardners in the U.K.

PW spoke with Zedoc founder and CEO Njål Hansen about how the platform works and how it grew out of problems Beat encountered running its core business.

Can you tell me about the origins of Zedoc and how it connects to Beat?

Over the last years, [my colleague] and myself, we've been working on Beat, which is providing white label e-book and audio apps. Currently, we're serving one of the largest Dutch services, and we have services across approximately a dozen different countries. So that's where we came from. Then one of my colleagues in Beat, half a year ago, showed me his side project, where he was using AI to solve these problems that we had in Beat, and then beyond that, creating other assets as well. It got me excited.

What made that side project significant to you?

That was, for me, an eye-opener in sort of how much of the problems that we had seen was actually solvable in a much more efficient way by use of AI.

What were the challenges in turning that into a production tool, rather than just an experiment?

You can sit down and prompt the AI engine and you can get some kinds of results. But once you want to produce something professional, there's all kinds of guardrails that you need to put into place. So we did that in a very early stage. We sort of engineered a solution that at least got to a place where we had predictable results. That's one of the downsides of using AI, is that it's not really deterministic. So you get these kinds of fussy results sometimes, but doing it the right way—making not only one hit with the AI engine, but maybe a kind of a loop where you prompt it until you get the result and then you move on.

How did you settle on the specific use cases Zedoc is built around?

We want to make books that are invisible to the market, visible. The problem, to be very clear, is the lack of metadata, poor metadata, poorly formatted metadata. We created is an agentic platform, one where you can have different agents connecting together and performing tasks. We call that a pipeline. With a pipeline, we can solve the ONIX [metadata] problem—the metadata problem. Then we stepped back and saw that similar pipelines can be used for a range of different things.

We started with ONIX enrichment, and then added accessibility enrichment and created a way scoring books for accessibility. And now we are offering tools for enriching and structuring data for LLMs and AI discovery.

Why is AI discovery becoming a priority?

A lot of the publishers are now seeing search less and less driving traffic to their sites. A lot of the conversations are happening in the LLMs. It's affecting traffic, but it's also affecting search. The estimate is that 90% of content delivery will run through AI assistants by the end of the year. So you need the metadata to do that, but you need to serve it in a certain way for the LLMs to pick it up as well.