Most researchers and startups do not think seriously about patents until they have to.
Sometimes that moment comes when they are preparing to file their own patent. Sometimes it comes during investor due diligence. Sometimes a potential customer asks about freedom to operate. And sometimes it arrives in the least pleasant form possible: a competitor's patent, or a letter from their lawyers.
By then, a lot of decisions have already been made. The product has been designed. Engineering resources have been committed. Investors have come in. Customers may already be using the technology.
So why don't companies look earlier?
There are two obvious reasons: knowledge and cost. But underneath both is a deeper problem. Patent information is widely available. Getting useful intelligence out of it is not.
Patents weren't written for researchers
Patents are technical documents at their core. They describe problems, technologies, architectures, materials, processes and different ways of solving engineering problems.
But they are also legal documents, and that changes the language considerably.
An engineer might say:
"A cooling plate carries heat away from the battery cells."
A patent claim might describe the same idea more like:
"A thermal management assembly comprising a heat-transfer member configured to thermally couple with a plurality of electrochemical cells…"
Or a researcher might write:
"The camera detects defects in the weld."
While a patent could describe:
"An imaging subsystem configured to acquire image data associated with a joining region and a processor configured to determine a defect condition based at least in part on the acquired image data…"
Neither description is necessarily wrong. Patent practitioners deliberately use words such as comprising, configured to, plurality, wherein, coupled to, at least one and based at least in part on because patents are written to define legal scope, not simply to explain an invention as efficiently as possible. To a scientist or engineer encountering this language for the first time, it can feel like reading a foreign dialect of their own technology.
The good news is that this part can be learned. Once you have read enough patents, you start looking past the language and recognising the underlying technical problem, solution and claim structure.
The harder problem is finding the right patents to read in the first place.
The European Patent Office's Espacenet database passed 160 million patent documents in April 2025, drawn from patent authorities around the world. About half of those are granted patents. It had passed 150 million about a year earlier, so roughly ten million documents entered the collection in twelve months.
European Patent Office, Espacenet worldwide bibliographic collection, milestone reached early April 2025.
That scale fundamentally changes the search problem.
Most of us have been trained by Google to type a few words into a box and expect the most relevant answer to appear somewhere in the first ten results. Because of the legal language used by patent drafters, the words are deliberately obscured. That makes finding the right set of patents difficult, and it is why a patent search can return hundreds or thousands of superficially relevant results while still missing an important reference.
Our answer is a proprietary method that uses vector embeddings in a unique way to match the paragraph deep in the description that you were searching for.
Retrieval alone was never going to be enough, and there is now evidence for that. An independent benchmark published in BASES in January 2026 tested fourteen AI-assisted patent search tools on one technical query. Best AI results and Boolean search results had precision of under 50%. This study confirmed what most of us in the search world already knew. Current search tools have poor precision and are expensive.
And patents are only half the story
For technology intelligence, looking only at patents creates another blind spot. Scientific publications often reveal what is happening earlier in the research cycle.
Patent applications are generally published around 18 months after their earliest filing or priority date, subject to exceptions. That means the technical work behind a newly published patent may already be considerably older by the time the public sees it. Scientific papers and preprints operate on a different timeline. Moreover, many deep tech start-ups publish first and then patent, which offers a glimpse into the filing strategy. This is why at Strata we realised that patents and publications are much more powerful when read together.
Publications are easier to search than patents, but there are far more of them. In our own quantum vault, 15,714 patents sit alongside roughly 140,000 scientific papers covering the same field. Keeping up with both records takes dedicated people, which small teams do not have.
So most teams do what rational people with constrained resources often do. They search Google. They read some papers. They check a few competitor patents. They attend conferences once a year. And then they get back to building.
The problem is not that startups do not care about intelligence. The cost of obtaining good intelligence has historically been too high relative to their time and resources.
This is why at Strata we built Technology Intelligence Vaults: curated data repositories that answer the questions you want answers to.
- Should we patent this?
- Who else is solving this problem?
- What approaches have already been tried?
- Where are competitors investing?
- Which technical problems keep appearing across the industry?
- What has moved from university research into corporate filings?
- Where is research active but patent activity still relatively sparse?
- Which companies appear to be converging on the same architecture?
They are technology strategy questions. Answering them properly increasingly requires connecting patents with papers, white papers, prosecution history, litigation evidence, company activity and other public technical information.
Finding technical information is easy. Finding what matters is hard. That is the problem Strata IP is built to solve.
To see how we approach technology intelligence, explore the Strata IP research and resources, including our work on patent-search noise, the genesis of Strata IP, and technology deep dives from the Quantum Vault.