Systematic FTO: What I Showed The World In My Webinar With Qthena

On 30 June, I did a webinar with Jonathan Laidler, Chief Growth Officer at ipQuants, the company behind Qthena. The topic was AI-assisted systematic Freedom to Operate (FTO). We had 60 minutes. FTO is a big elephant, and 60 minutes is not enough to eat the whole thing, but here is what I could fit in.

A bit of background, because it matters 

I started as a patent attorney in 1996. Before that, in 1995, I spent six months as an auxiliary examiner at the German Patent Office. Back then, prior art search meant paper. Boxes of it, sorted by IPC class, in steel cabinets. Once a month, a trolley arrived with ten or fifteen kilograms of new patent publications, and we spent the day sorting them into the right boxes. That is where I learned to search systematically, one class at a time, and I have carried that method with me ever since, first onto computers, and now onto AI.

I am not part of Qthena. I have no stake in the company. I give talks and training on AI in the IP profession, I have used more than 40 different AI tools in this space, and when I found Qthena, I recognized something different. It analyses documents from the first line to the last, not by estimation, the way most large language models do with retrieval-based (RAG) pipelines that only pull the “most relevant” snippets. For FTO work, where missing the wrong paragraph can mean missing the wrong patent, that distinction matters enormously.

Landscape FTO versus systematic FTO 

I told a story in the webinar about a startup of mine that was acquired in 2003. The buyer’s lawyers looked at the FTO I had done and told me it was no good because it was only a “landscape FTO.” I pushed back, because it was not a landscape FTO. It was systematic. The extended FTO they commissioned to prove me wrong found nothing I had not already covered, and that was expensive proof that the method works.

The difference matters. A landscape FTO gives you a probabilistic result: here is your product description, here is an output, trust it. A systematic FTO is controlled. You know what you have covered, you know where the gaps are, and you can explain your work to a judge if it ever comes to that.

Four myths I addressed 

Having a patent does not mean you are free to sell your product. Searching only in your operating countries is not safe, because patent families cross borders in ways that only surface in a worldwide search. FTO is not a one-time exercise; it needs to track the entire development of a product. And 100% coverage is never possible; what matters is doing the work properly and starting early.

Cycles, not a single pass 

My method splits FTO into cycles. Cycle 1 starts from nothing but the product description: an AI semantic search, ranked by relevance, reviewed by a human. That output, the flagged patents, IPC classes, and competitor names, becomes the input for cycle 2, which refines the search and does a first infringement triage. Cycle 3 and 4 repeat the process with progressively sharper inputs. Confidence climbs from around 80% after cycle 1 toward 95% by cycle 4, but each additional point of confidence costs roughly ten times more than the last. That is the law of diminishing returns, and clients need to understand it before they commission a search, not after.

In practice, a semantic search with Qthena typically returns 30 to 100 candidate patents. Qthena classifies these into relevant, maybe relevant, and not relevant, and that classification is reliable enough that a full review of 100 patents, which used to take me an entire day, now takes under an hour. From those 100, you typically end up with two to four patents that genuinely warrant a full infringement analysis. That is where the real legal work, and the real cost, should go.

Where the human stays firmly in charge 

None of this replaces the attorney. Qthena surfaces candidates, extracts IPC classes and competitor names, and drafts a first classification. The attorney still does the legal judgment: claim scope, jurisdiction, design-around options, and the signed opinion with a defined confidence level. Under the EU AI Act, a law firm deploying AI for client-facing FTO work is the deployer, and carries full responsibility for the output. A documented expert review is not optional.

What it means in practice 

Across the roughly 300 patent professionals I have trained on AI tools over the past two and a half years, about 20% cut their working time by a factor of four or five using this approach. Most others at least double their speed. Results vary, as they do with any new tool, but for systematic FTO specifically, this has been the single biggest shift I have seen in thirty years of doing this work.

Here is the full recording of the webinar, including the live demo of the search and analysis workflow inside Qthena.

IP Lawyer Tools by Martin Schweiger