For any successful business collaboration, chemistry is key. The elements and compounds leading to the thriving partnership of Matthias Poetzl and Michael Natterer, co-founders of AI-powered patent analytics firm Octimine, can be attributed to a number of personal circumstances: getting out of the finance business after the 2008 crisis, following a passion for innovation after a stint in consulting, and pursuing advanced degrees at the University of Munich. Poetzl’s training was in physics, while Natterer was a computer scientist. Both saw the commercial opportunity of Natterer’s Ph.D. research on natural language processing, machine learning, and patents.
In 2015, after earning their degrees, they went into business together, creating a patent search engine called Octimine. Their offering was so successful that only three years after founding the company, Octimine was acquired by the Dennemeyer Group, a global IP full service provider. Poetzl and Natterer both serve as managing directors for Dennemeyer, and while their responsibilities at the firm have expanded, they also continue to enhance the capabilities of Octimine.
IFI recently talked patent shop with Poetzl and Natterer in a wide-ranging discussion about Octimine’s longstanding involvement in applying machine learning to patents, how ChatGPT might change the patent industry, why patenting software is so difficult, and how using IFI’s data helped Octimine’s business scale. The interview has been edited for clarity and length.
IFI CLAIMS: What led you to start Octimine?
Poetzl: I have a physics degree and a Ph.D. in business administration. After my physics degree, I took a little detour into strategic management consulting in Munich. And then I followed my passion, which is innovation and technology management. At the University of Munich where I did my Master’s degree and Ph.D., we worked a lot with patent data. We used it to build indicators to measure the innovativeness of companies and countries. And during that time, I met my founding partner Michael Natterer. He was working on a machine learning technology to compare patent text to find similar patterns to certain inputs. And that was kind of the idea that led to Octimine.
In 2015, at the early stages of the company, we were an add-on tool for standard patent search tools. We were one of the first ones who really did patent similarity search at a certain quality level. Today, a lot more of these algorithms are public source but back in those days we had to develop them by ourselves. So that was our starting point. We eventually received venture capital, had some financing rounds, and then sold the company at the end of 2018 to the Dennemeyer Group, one of the largest IP service providers in the world. Now I'm responsible for everything that has to do with commercial software applications with Octimine and also our IP management software, DIAMS iQ.
Natterer: My background is computer science and business administration. I studied at the University of Munich. I also studied abroad in Sydney, Australia and in Cambridge, Mass. When I returned to Germany, we were in the middle of the great financial crisis and I had wanted to go into the financial industry at that time, so I was working for J.P. Morgan. But then I thought maybe it was the wrong time for the financial industry because we had stock prices decreasing by five percent every day, and so I decided to do my Ph.D. This was when I met Matthias for the first time. My text data for creating a search engine on patents originally was supposed to predict patent values, but it became something not just for patent values but for search engines in general. We saw that it had a commercial application in using AI in order to automate patent search as much as possible.
IFI CLAIMS: What problem does Octimine solve?
Poetzl: We have a very intuitive, easy-to-use, AI-supported patent analytics tool. For large enterprises, we are often a second tool used alongside their standard patent search tool. But our main targets are medium-sized clients, universities and R&D departments. We solve the problem that finding relevant patent documents is usually pretty hard to do if you're not an expert or an IP professional. You have to have certain skills. You have to know how to deal with patent classifications and keywords and what the difference is between publications, applications and so on. We make this very easy for all end users. We help them find interesting patent documents with the help of machine learning. Of course, we do it with a very intuitive interface and a great user experience.
IFI CLAIMS: Take us through an example.
Poetzl: A typical client would be a technology company that maybe uses public tools like the USPTO, the European Patent Office and other public patent search spaces. Or, they work with patent attorneys who charge a lot of money for their services, and they’re starting to professionalize in IP. They have the problem that they want to know how to deal with patents themselves, how to find documents, and how to get that process into the organization and enable their own people to do this.
With the public tools, you can do basic searches, but you cannot store anything in a useful manner. You cannot share content with your colleagues. You cannot set up patent alerts that come in every week that you can then label and classify on the document. These traditional tools are made for IP experts. They have a lot of functionality, a lot of features, but they’re complicated and old-fashioned. Our clients want something in between. They want to have something more professional than the EPO or the USPTO tools, but they don’t want big tools that are too complicated for them. That's why they come to us.
IFI CLAIMS: What sets you apart from other providers offering AI and machine learning?
Poetzl: Our history and our experience. We have worked for more than 14 years in the area of machine learning and patents. That's a very long time compared to our competitors. We are one of the pioneers in that area, so we have seen a lot. We know what kinds of models work. We know what models don't work. We have always been very up-to-date with what is going on in the market in terms of machine learning models and in terms of new technologies.
We are also unique when it comes to the usability of the software. We are focused on user experience and user interface. We make it so easy for the end user, which sounds obvious, but it's pretty hard, especially when you think about patent monitoring in a way that lets team members be involved in annotation and labeling of documents, and then storing all that work safely. That's a lot of work on user experience and interface to make that function.
IFI CLAIMS: What patent trends in general are you seeing?
Poetzl: One area that is really coming up right now and that we are working in is machine learning and artificial intelligence.
Natterer: ChatGPT is everywhere in the news. We recently did a survey in the IP industry and 90 percent of the people have heard about it. That said, 90 percent of the people have not signed up yet to use it. So we find that very interesting. There are many use cases this AI technology can actually solve. Where it works well is in idea generation. For example, I asked for products around automating patent search and what ChatGPT came back with was on target. It gave me lists that I could iterate and make more technology-specific. It is also good at summarizing scientific papers, easily creating a high-quality abstract. I have not tried it with patents so far, but I think, to some extent, ChatGPT can do it as long as a human is in the loop.
IFI CLAIMS: Do you have any sense yet of how ChatGPT could change the patent industry 10 years down the road?
Natterer: It will help with the operational tasks. Think about what a paralegal does in the patent filing and drafting process. This is something where these tools can help. Setting up templates. Easy interpretations. Using it as a search engine—not as of today though because of this hallucination problem, which means it can give you an answer with full confidence but it’s wrong. To give you an example, I looked up a patent. I said, “Hey ChatGPT, can you tell me what EP2049363 is?” It recognized that this was a European patent, but then it told me with full confidence that it was from Qualcomm and covered network devices. Actually it was a patent about ventilated vehicle seats from the automotive industry! So this is a problem when you are using it as a search engine because you don’t have any information about how confident it is and you don’t get any sources.
I can’t say ChatGPT will replace 30 percent of the paralegals in two years. This would be completely wrong. It can do certain tasks. It can speed up the work of paralegals, but I think the job of paralegals will also change over time, in the same way that many other jobs will change.
But in areas like patent drafting and constructing the claims, most patent attorneys will say that is the most challenging part so that is something that is a bit more down the road with ChatGPT. But when it comes to creating a description, the background of the invention, the details of the description, and the abstract, I’m pretty sure that it’s going to be a tool that can help tremendously. So I think it will have a strong impact.
IFI CLAIMS: And aside from ChatGPT? Any other strong trends you’re seeing?
Poetzl: We think about the Internet of Things, where software and the technology of old industry are coming together. I see a lot of filing activity going on. There is also a lot of risk. As we talk to clients, we find that they are a bit scared of software companies moving into their field—of Amazon and other big tech companies publishing patents that go into the old economies.
Natterer: If you want to secure your business, you need to look left and right, not only at the players from the same industry. Taking an example from logistics, companies like DHL are not just looking at UPS and FedEx anymore. They also need to look at companies from other spaces like Amazon and startups in other sectors like drones and other automation processes.
Number of years that Poetzl and Natterer have been applying machine learning to patents.
Poetzl: The other thing is that sometimes the cycles of new technology are so fast that it doesn't make sense to have a patent. In fact, we filed for a patent in 2016 for our technology, and it was granted in one authority, but it was only just granted last year. As of last year, the technology we have a patent on is already outdated. We’re not using it anymore. Especially in the software area, there's so much new technology coming up every year that to file a patent doesn't really protect you. It doesn't really give you a big advantage. So then the question is how patents will be used in the future.
IFI CLAIMS: Does that mean companies will protect their inventions less?
Poetzl: In the promising technology areas, I would say there's more usage of IP rights. I don't see companies tending to protect with fewer patents than they did, say, ten years ago. Protecting your intellectual property is still an important part of your innovation process. There are new ways to track with blockchain, but protecting your core technology with patents and trademarks is still the most important thing.
Except for software. Big software companies don't have that many patents. But in the electronic, sensoric, optical, chemical, pharmaceutical, biotechnology, and telecommunications fields—especially telecommunications—patenting is a huge thing and will be in the future.
IFI CLAIMS: Why is patenting software so hard?
Poetzl: You need to have a technical aspect. A physical improvement of something, something you can actually touch, my patent attorney told me. And in our case, we patented a method to reduce something on the server. So we actually referred to something technical on the machine, not our algorithm. You cannot patent an algorithm. You have to patent something that is technical. That's the way around it. And that's not easy.
IFI CLAIMS: So did going after your own patent end up being a waste of time?
Poetzl: It was not a waste of time. You never know if you’ll need it for defensive purposes.
IFI CLAIMS: Why do you use IFI data in Octimine’s offerings?
Poetzl: In our early years, we were a very small company. And we started getting the data from the different patent offices on our own. We were getting the data from WIPO, from the EPO, from the USPTO and then we worked with another provider to get one or two additional authorities.
That was a lot of work for our small data science team. We wanted our people to focus on the technology and not on how to load the data and do the whole ETL process. So that's why we were looking for a provider who could deliver all the data, all at once, and help us with the data warehouse and the data lake at the beginning of the process. And that was IFI CLAIMS because of the ease of use. That's why we could easily scale up from five authorities to what we have now, which is 48.