This is contrAnalyst, your personalized guide to contract risk management
SELECT contract, risk, big data FROM legal JOIN tech USING hackathon;
Overwhelming. That is the word I would use to describe the Berlin Legal Tech Hackathon if asked to do so. Further descriptions apply as well, but let’s forget that for a while and start at the beginning.
How is big data able to assist the legal users? Where is ample data available which is suitable for training IT systems? And which needs of a broader audience could we satisfy? These were among the questions with which our group started its activities.
That basically brings in the two perspectives we tried to match at the beginning of our group’s actual hacking session:
- What is the customer need? In other words, look at it from a problem and need/want oriented perspective.
- What is easy to achieve and leverage? That is, have a look at it from a perspective of possible solutions and feasibility.
Please note, that we’ve sticked to the “customer first” approach when ordering the perspectives. Solve a need, and a business will most probably follow.
In such discussions, the diversity of our expertise and backgrounds came handy. With all that power thrown in, quite quickly, an image of our expected product materialized:
contrAnalyst, the user’s “personal” contract analyst, displaying both risks of a specific contract as well as an overview of the risk exposure in general (e.g. how many contracts with potentially void clauses are in place).
Now that the big picture has set in, let the hack begin…
The process was well iterative, with alternations of discussions in the group’s entirety, preparations by the topic specialists and discussions in smaller groups. For one thing, this proves most effective from a time perspective. For another, with the regular sharpening iterations, the understanding quickly increased and learning was easy.
Within the first round, Marco created a first model; Xavier started with a dashboard design, Idan worked on the graphic charter and the lawyers explained what the focal points regarding the contracts should be and started with providing contract examples as input. Parallelizing work and trying to go in the same direction for a common goal. Thanks to a proactive networking during the hackathon, we also got the chance to use an OCR (Optical Character Recognition) solution for free which was able to solve the conversion of PDF documents to texts and simplify the text recognition process. First outcomes were soon seen, already starting to show what would be possible. Unexpectedly, we were to going to finish our project.
Soon enough, though, the first obstacles came into view: After feeding the first contracts into the model to train, we slowly got a better image of what is doable immediately and with reasonable effort compared to what is rather hard to realize. Our initial task of testing the validity of clauses was somewhat too big for the day and a half of a hackathon. For major parts of the group, this helped us learn about the necessary effort as well as about suitable input for learning models.
Now then, refocusing time. Just for the start, we decided to work on more standard, readily accessible information within the contract: the (remaining) term. With the little training set we had (about 40 different contracts), even this proved difficult; machine learning can provide amazing solutions but always at the cost of sufficiently high amount of data available; hence, we developed ideas to reduce the complexity even more and manually created varying clauses regarding term, covering most important variants, however most having a somewhat similar structure.
In the end, working the topic from various sides and perspectives, with joint effort, we were proud to show a visual prototype at the presentation, showcasing a dashboard and a possible tool structure.
We were happy with the outcome, the jury of the Berlin Legal Tech Hackathon pitch liked it, and we were happy to leave with the third prize.
Thanks to the jury for this, the sponsors of the event and prices and even more for the organizers to set up this great event with an incredible drive and spirit!
Team (you may contact us via medium or find us on LinkedIn):
- Idan Nesher, our designer, who is able to make almost everything look appealing
- Marco Scaravelli, Italian and our IT and machine learning specialist
- Xavier Lavayssière, French and the most customer focused of all of us
- Nuri Khadem, lawyer, and very into IT
- Baltasar Cevc, a former IT guy turned lawyer
- David, another lawyer, who unfortunately had to leave before the pitch
The team, the approach of which perfectly matches one of Silicon Valley’s favorite quotes:
“Ever tried. Ever failed. No matter. Try again. Fail again. Fail better.”
(Samuel Backet in Worstward Ho)