YC - Andrew Flashcards
What progress have you made since your application (then you had 2M cases, tens of millions of documents, and nearing MVP)?
We have completed our MVP, and identified a number of potential investments, and have reached out. Now iterating
You said prototype/beta is coming up from November 1st - where are you at? On track?
We’ve already made a couple preliminary models, so this is actually already done. Now we’re just iterating and feature engineering to improve. Now we are pushing towards an investible model Dec 1.
You said your advantage is driven by your combination legal and quant finance and your ability to invest in cases below $500K, lower than nearest competitor. Why do these give you an advantage?
These advantages allow us to successfully build the technology (and the portfolio). These enable us to access these lower cases and enjoy a better cost structure generally
You talk a lot about cost-effectiveness in your app - where is this now?
With higher precision technological models, we don’t need as much human diligence. Also more efficient capital deployments
How do you justify lower rates to LPs?
We have higher precision and more efficient capital deployments, leading to higher returns. LP stay in yo lane and enjoy your returns
You talk about understanding NLP as being how you think about this differently. How would this work? How do you know it’s possible?
Abstractly, court documents contain a ton of information about a case. More concretely, I’ve read successful research papers on similar tasks, and we plan on being more sophisticated than what I’ve read.
Is there no one else doing something similar?
Legalist and LexShares are the two other players investing in smaller cases. Both are using relatively simple models as a filtering system, and their average case sizes is well above what we’re targeting
Legalist average investment size is $500K, Lexshares is average ~$1.5M, both have been trending up.
How is your product different?
Quant-driven approach. Lower cost. Smaller case sizes. Higher precision.
What competition do you fear most?
Legalist if they decide to hire sufficient talent and revamp business model. Same for larger players.
Who are your competitors?
Legalist most directly. Other large firms like Burford, Lake Whillans
Who might become competitors?
Big players may all become more direct competitors with tech investments
What is your growth like?
Had 244 customer conversations. We recently identified a segment of the market that seems very promising. Technical things that have improved.
How large is the market that you are going after?
(I think this is the same as below…but) $640M/year in recoveries in our narrow NY cases. 10B if you extrapolate US (by population)
How many people are in your target market?
33K cases per year for contracts in our population in NY. 500K est. in US., and this is the subsegment of contract cases which we are finding traction in
How will you validate?
Will try to raise with backtest. If not, we’ll invest smaller amounts with our own money
What have you learned so far from working on your product?
Dataset is highly particular, requiring significant domain knowledge to parse
What’s new about what you make?
Quant approach, including NLP. Lower case sizes. Heavier reliance on tech.
How does your product work in more detail?
3 data sources, advanced ML techniques, quant approach to the whole portfolio
What are you going to do next?
Build robust model, get funding, invest in cases
What part of your project are you going to build first?
Model and investing mechanism
What is the next step with the product evolution?
Iteration on model. More feature engineering. Next are NLP iterations
Where is the rocket science here?
Quant investing applied to law; litigation domain knowledge pairing
Why is your product better than Legalist?
We believe Legalist’s models are simpler than ours, as they appear to be focused on building arrays of data largely based on metadata. Our additional tech (focused on language models/NLP) allows us to be more precise in our analysis of case outcomes, which allows us to invest in < 50K investments.
Why do you think Legalist’s models are simple?
From their technology team, description of thier process in a June 2020 SEC filing, and all the other research we’ve done.
How is your solution better?
Better models means lower costs, higher precision. Better portfolio management means higher returns on capital and less risk.
Do you have proprietary data?
No, but as we fund cases we will build a proprietary database
How are you analyzing settlement values?
Estimating settlement values based on similar cases, using full data picture at the time of settlement
Demo?
We don’t have a demo because there is no user interface. We’re happy to show you anything you want to see about our model / backend.
Algothmic Bias
We actually see this as an opportunity. First level, debias the corpus, but in the best version we can actually confront the counterfactual: what if disadvantaged plaintiffs had been given funding?
What has been your progress over 6 months?
AH: Over the past four months on the technology side we’ve built scripts for blah blah blah. And during that time, we’ve diligenced hundreds of specific cases that have been identified by our algortihms, and reached out to the lawyers and plaintiffs in those cases.
Why would you stick together?
We are each emotionally invested in solving this problem for personal reasons, we’ve already handled adversity, and through working together we have developed a deep level of respect for each other. Hell, we all moved to NYC during a pandemic to be together! (If the q is directed at Andrew or wrt ClassWise, mention the dedication levels of the founders.)
Tell us about a tough problem you solved?
Technical stuff.
Tell us something surprising you have done?
Living together. We used GPT3 generated phrases to generate our website.
What’s an impressive thing you have done?
Current tech pipeline to get to multiple input model
Why did your team get together?
We are also data guys, and this is one of the coolest problems out there. But more are really passionate about fairness as we have been personally impacted by the high cost of legal proceedings.
Why did you quit ClassWise? Why will you stick this one out?
ClassWise failed fast. Weren’t even able to help schools for free. Team didn’t include domain experts and also wasn’t as dedicated as Patronus.
Funders preferred explainability over predictability. What are the implications of building something like this for the legal system where results are unexplainable?
Explainability functionality. Counterfactual. See a lot of this as an opportunity. Partial dependence plot. LIME. I don’t think we’ll get this question
How would you build a moat?
Building a pipeline of law firms. Over time we’ll also have proprietary data. (And also it’s hard)
Bottom-up cases
Each case is avg 20K/year after paying our investors. So 50 cases = 1M. 5000 cases is 100M. And there are 200K cases every year.
First hire
Engineer for expanding into new states and helping with model
Tell me about your quant trading background
Built a new quant trading desk in a new asset. Involved data science, algo development, product management, portfolio optimization.
If legalist decides to compete with you, why will you win?
The founding team. Sure, they have the resources to make hires, but it matters that the founding team has serious deficits in Data Science, Litigation, Ops, and Finance. That’s a lot to make up for.
