On the path to automated patent workflows

Not too long ago, automating technical content-related patent work sounded a bit like science fiction. Now, it’s starting to feel much more real – not just as a dream or distant goal, but as a set of concrete steps we can take, features we can build, and use cases we can cover.

As we've mentioned before, our goal is to automate repetitive, substance-level patent work in a way that is transparent, verifiable, and trustworthy. If you're curious about what exactly we do – and don’t – mean by that, I explain it in more detail in this blog post.

In recent months, we've introduced a range of performance enhancements and features that pave the way for greater automation. IPRally’s unique Smart filters alone now answer hundreds of thousands of questions in IPRally each week – that’s a lot of patent reading outsourced to AI! The fact that users keep coming back and asking more proves that they get a lot of value.

Image search has also been an eye-opener for many, as it makes it very tangible that the AI really understands technology. With it, you can both speed up work and unlock entirely new use cases, like infringement analysis. Multi-modality both in search queries and in the search corpus is the future without doubt.

With the recent Projects launch, we’re giving users more control over how related search cases are grouped, reviewed, and managed. Besides making manual work more efficient, Projects will also act as containers for fully automated workflows.

The new Refine function streamlines free-text searches by filtering out non-technical content and transforming key features into a “mock claim” that is optimized for IPRally's graph AI.

We're proud to be trusted by a growing number of patent professionals, helping them work faster, more effectively, and with greater confidence. 

Towards automated, yet transparent, workflows

A typical patent search is a challenging task, requiring engineering and patent domain knowledge, tool proficiency and a lot of persistence – if done in the traditional way. It involves identifying and interpreting relevant prior art from among hundreds of millions of documents, written in multiple languages and often marked by inconsistent terminology open for interpretation. Meanwhile, data volumes continue to grow, whilst technologies evolve rapidly and increasingly overlap, making it ever more difficult to define clear boundaries between technical domains.

Deep research-type Generative AI products (like ChatGPT or Gemini Deep Research) have found their way to professionals’ toolboxes in e.g. engineering, science, e-commerce, and healthcare domains. They excel in complex tasks that require both information retrieval and analysis, often in an iterative way. 

While there is a strong resemblance with patent search and analysis, general-purpose deep research tools fall short when it comes to professional patent searching. Whether their search tools capture the most relevant patents from the 150+ million ones out there is very much hit-and-miss, often based on random web searches. Also, their patent domain knowledge is still on “junior level”- or applied too naively in the analysis stage. 

These parts are, however, where IPRally excels.

IPRally is purpose-built to address these challenges using technology, with a mission to make information-finding effortless and elevating humans in the value chain. As I write this, we’ve been developing and training our AI and building the product around it for over seven years. Also, we’ve been adding so-called AI Assistants like Smart filters gradually, allowing the user to sit in the driver’s seat.

While we keep doing all that, our aim is to take a leap forward and release the first version of fully automated search workflows into the hands of selected users later this year. Think of it as an extremely fast co-worker. For example: You input an invention disclosure to check its novelty, go grab a coffee, come back to a structured, pre-analyzed set of search results. A process that today could take hours – and not so long ago days – would be done in a few minutes.

The outcome will be a fully populated IPRally Project with:

• Key aspects and features of the invention automatically detected 
• Automatically generated searches covering the invention in the best possible way  
• Each search result analysed in feature level 
• The most relevant results identified based on the analysis
• Draft reports in multiple formats and levels of detail

All this ready for the human expert to intuitively and interactively review and verify, and to make decisions – faster and with less effort.

Transparency, human-verifiability and trust are our keywords.

Prior art searching first

For building trust, it is not smart to bite off more than you can chew. 

We need to start with use cases where automation provides value even when imperfect (well, which patent intelligence project is perfect anyway?) and that can be easily measured and double-checked by humans.

Prior art searching is the perfect use case to start with. This is particularly true for pre-filing novelty assessments, patentability analysis, and opposition or invalidity proceedings. A quick search, even if imperfect, has only upsides. You might immediately find the killer prior art you need for key decisions. And if the results aren't satisfactory, you always have the option to fall back to traditional search methods.

While Freedom to Operate and other more risk management focused use cases seldom have 100% confidence requirements, they are usually considerably higher. Successful prior art search automation will also build trust and pave the way for such use cases, too. 

This was also the clear signal from our users in a survey that we did a while ago.

Looking ahead

In the slightly longer term, the goal is trustworthy automated virtual workers tuned for specific tasks. They would run whenever needed and optionally be triggered by key events like filed invention disclosures (novelty search), submitted R&D project proposals (state of the art and/or FTO search) or published or granted competitor patents (invalidity search). 

As a result, stakeholders would get practically immediate, reasoned and structured analyses ready to support informed decision-making. We are building the technical foundations and key components for all that.

While humans need to have the ultimate decision-making responsibility and occasionally roll up the sleeves for grassroot work too, I'm personally more and more confident that the AI is ready to handle a lot of the “grinding” involved. It will also solve a significant portion of tasks with high enough confidence without human assistance. 

Given the sheer volume of tasks and data involved, virtual workers will undoubtedly improve productivity, quality, and consistency, especially in large organizations.

In two years, coming to a workplace without patent automation will feel like coming to work without a computer today. The key elements of automation are starting to come together.

What do you think? I’d love to hear your thoughts. You can reach out on Linkedin, or drop me a message at product@iprally.com.

‍