How Ian Schick innovated in Legal Tech through AI and turned a challenge into an opportunity
Ian Schick is the CEO and Co-Founder of Specifio, a Natural Language Generation company that automates the mechanical and tedious parts of drafting a patent application.
Specifio has recently been listed in the AI Time Journal TOP 25 Artificial Intelligence Companies 2018.
Before founding Specifio, Ian was working as a patent attorney in a global law firm.
The patent application drafting market had been stagnant for over 10 years, with shrinking margins and fierce competition.
Q: What is the story behind Specifio?
I was building a great patent practice in a large well-known law firm and I was having the same struggles as a lot of people are having. I was getting work in the door, but finding good people to get through that work was just a real challenge. And then, of course, there’s the price pressure. Being more senior at the firm, my rates were so high that this was work that I needed to oversee rather than do myself.
I had worked with several clients in the AI space working with technologies such as neural networks and other cutting-edge technologies. Through that experience, I learned a lot about the capabilities of AI and started having ideas about how it might apply to my own work as a patent attorney.
One day, for whatever reason, I put my ideas in an email and sent them to Kevin Knight, a professor at the University of Southern California and a former client of my first law firm (I had worked on the patent portfolio of his last start-up years earlier). That started a conversation between Kevin and me which culminated in the vision of Specifio.
Within six months or so we had a working prototype, and it was compelling enough to convince me to leave my law firm job to co-found Specifio.
Q: Who are the customers and what are the benefits of using Specifio?
Our customers are law firms and in-house legal departments who file patent applications directed to software-related inventions. Our users report time-savings on their patent application projects of about five to eight hours.
These projects are typically 20 to 25 hours so, best case, we are turning a three-day project into a two-day, which would be a big deal in any practice.
Some firms are really struggling with profitability with these practices, so our service helps them be more profitable. Other firms are trying to compete on price and with our service they can provide the same quality of work product but at a slightly lower cost. That’s kind of the beauty of what we’re doing. In ideal cases, the time saving is enough that there’s room for a three-way win: we take our fee, the law firm makes more money, and their client gets a break on the price.
In ideal cases, the time saving is enough that there’s room for a three-way win: we take our fee, the law firm makes more money, and their client gets a break on the price.
Q: How does Specifio work?
There are three parts in a patent application.
- The claims, which is a numbered list of sentences that very precisely define what is being asserted as the invention.
- The specification, which is a long body of text that explains how to make and use the claimed invention.
- The drawings that go along with the specification.
Crafting commercially-valuable claims is the part that I regard as an art form. It’s almost like a blend between writing computer code and writing poetry. It is very nuanced and, at the same time, it is very syntax-driven; so, we leave the claims to our attorney users.
Specifio takes the attorney-written claims as the input that our users send to our system.
They can send the claims to our fully-automated system via email or via our Microsoft Word Add-in (available for free on Microsoft’s AppSource). We provide a first-draft patent application back to the user by email. The generated documents are delivered in a password-encrypted zip file.
We can take two to three pages of attorney-written claims and transform that into a 20- to 30-page first-draft patent application in about two minutes. So the process will start and end with the attorney. They send the claims, and then they have to review and bolster the application after they get our draft. But it’s that middle chunk of the job that we’re really helping our users to fast forward right through. We also generate some basic drawing figures to go along with that.
We can take two to three pages of attorney-written claims and transform that into a 20- to 30-page first-draft patent application in about two minutes.
Q: Have you encountered any resistance or skepticism while bringing Specifio to the market?
The way our approach works is that we think in terms of writing projects.
Writing to us is something that can be composed into sub-tasks. And if you think about this sub-task, you need to be categorized as creativity, or judgment-driven, or more mechanical in nature, right?
It’s our ambition to automate the mechanical aspects and leave the intellectual heavy lifting with the attorney. We’re not trying to provide a substitute work product, we’re not trying to replace lawyers. We are simply providing a way for them to get through their projects in a much more efficient manner. So that kind of removes the threat of “Here comes AI to steal my job”.
It’s our ambition to automate the mechanical aspects and leave the intellectual heavy lifting with the attorney
And with that obstacle out of the way, it really just comes down to “do they like the work?”, “do they like the results?” and so far, generally the reaction that we’ve gotten is, yes.
Q: How do you train your machine learning models?
The patents are a unique area at least in the law in that there is so much publicly available information. There are millions of patents out there that have been published and at least that many patent applications, it’s just tons of training data. If we were working on contracts or something like that, that kind of document is usually kept private so it would be harder to get training data, but in certain areas of law, you can get public information. It’s a great opportunity.
Security and confidentiality
The information we process is highly sensitive: it is kind of the crown jewels of the law firm’s client and we respect that. I am an attorney myself, I practiced for 10 years on. No person on our side is ever exposed to the confidential subject matter, we delete it permanently after used and we don’t train our models on that confidential subject matter. As I said, there are millions of publicly available examples out there for us to train our models.
Q: Tell us about the Specifio Team.
We hire primarily out of the University of Southern California here in Los Angeles.
All of our tech guys are from the graduate program. They are the best and the brightest; they are brilliant computer scientists. On the business side, we are also growing and hiring in sales and customer support.
Q: In which direction are you bringing the product development forward?
We are pushing hard on all fronts. The NLP part is a never-ending project. We are constantly tweaking our system, making it better, implementing new tools as they come out.
We are continuously improving our models and adding intelligence, and more flexibility on the input as well as a richer output.
The NLP part is a never-ending project
Kevin, my co-founder, leads this area.
A Feature we recently launched
We can generate patent applications that look and read like our users own work and that’s been pretty huge with respect to the rate of adoption.
An interesting feature that we launched recently is the ability for our users to customize our output. The customizations are a series of worksheets that a user can click through and in a few minutes customize the appearance, content, and structure of the patent applications we generate for them.
We can generate patent applications that look and read like our users own work
Q: What are your business expansion plans?
There are a lot of opportunities in the patent application market.
So for example, right now, our service automates patent applications that are directed to software-related inventions, which I’d say is about 20% of the market. This year, we are working on expanding our coverage to include other areas of technology and trying to get from 20% of the patent market to around 80%.
We could also expand to different languages and it’s not a huge job to add languages thanks to how our core technology is built. There are five patent markets worldwide that make up 80% of all patents filed: Europe, Japan, Korea, China, and the US.
Opportunities in the patent world
Beyond patent preparation, there’s plenty of room for what my founder friends and I refer to as “patent tech” in the life cycle of a patent, from the inventor disclosing the invention to the patent attorney through preparing the patent application, filing it and prosecuting it at the Patent Office, and then post-issuance activities. That entire process is historically almost 100% manual, and there’s just a ton of inefficiency in the different areas and a ton of opportunity for automation. There are several really exciting companies already working on different parts of the patent pipeline.
We view ourselves as a natural language generation company and we will likely stay in document generation. We may eventually try to automate other documents in the patent world or maybe even outside of patent but still in legal.