What we talked about
In this episode of the PreVetted Podcast, we sit down with Dr. Tanguy Chau, Founder and CEO of Paxton, the AI-powered legal platform making waves in the industry by helping lawyers work 10x more efficiently.
With a PhD from MIT and a background in engineering, venture capital, and early-stage investments in groundbreaking companies like Ironclad, Tanguy brings a rare blend of technical precision and business insight to the world of legal tech.
Show notes
Tanguy Chau makes a point that turns out to be more alarming than it first sounds: when lawyers were sanctioned for citing cases that do not exist, most of them had actually asked ChatGPT to verify the case, and ChatGPT had confidently confirmed it was real. He calls it a sophisticated liar, not a lazy tool, but a system that doubles down convincingly. That is why a general-purpose AI is not just inadequate for legal work; it is actively dangerous.
What we covered
- Paxton started with legal research because the dominant incumbents, Westlaw and LexisNexis, required Boolean query syntax and multi-year contracts that put serious research tools out of reach for small and mid-sized firms. Translating natural language questions into a cumbersome query language was a barrier that AI could eliminate entirely.
- Hallucination in the legal context has three distinct forms: inventing a case that does not exist, correctly citing a case but misrepresenting what it actually says, and citing a case that has been overturned by a later court ruling. Paxton built a shepardization system to address the third problem, running millions of cases through large language models to determine which decisions are still valid law, a task that previously required hundreds of millions of dollars in manual editorial labor.
- Paxton operates on its own curated database of statutes, regulations, and court rulings rather than generating from a foundational model. When a case is cited, users can click through to see the actual source document and which court published it. The AI has its own knowledge of the law, it is not simply wrapping ChatGPT.
- The product-led growth approach, free trial, no sales call required, cancel anytime, has resulted in deployment across multiple thousands of law firms globally, including Am Law 100 firms. Tanguy argues this is only possible because small firms battle-tested and sharpened the product before the larger ones arrived.
- From his time as an investor in companies like Ironclad and Case Text (acquired by Thomson Reuters), Tanguy’s clearest lesson for founders is that the counterintuitive move is to narrow relentlessly. Investors want a huge market, but product-market fit comes from solving one specific problem for one specific type of user before expanding. The diagnostic question he uses: what do you do uniquely that people are desperate for?
- His prediction for the next five years: it will become an ethical breach for a lawyer not to use AI. The efficiency and accuracy gains are so large that failing to adopt the best available tools would be a failure of duty to clients.
About Tanguy
Dr. Tanguy Chau is the founder and CEO of Paxton, an AI-powered legal research and drafting platform. He holds a PhD from MIT and brings a decade of legal tech investing experience, including early-stage investments in Ironclad and Case Text.
- LinkedIn: https://www.linkedin.com/in/tanguychau
- Website: https://www.paxton.ai
Episode 48 of the PreVetted Podcast.
Full transcript
Federico Ramallo (00:01) Welcome back to the Prevetted Podcast, where we spotlight extraordinary people and remarkable talent reshaping our world. Our guest today is Dr. Tangui Zhao. He’s a founder and CEO of Paxton, AI-powered legal research platform designed to make lawyers 10 times more efficient. He has a PhD from MIT and experience in both engineering and venture capital. Tangui.
has helped fund and build cutting-edge tech companies, including early investments in Ironclad and Cover.ai. Welcome, Tangui. How are you doing today?
Tanguy Chau (00:40) Federico, thank you very much. I’m very excited to be able to speak with you and your audience.
Federico Ramallo (00:45) Yeah, thank you for joining us. So I wanted to ask you about Paxton. What inspired you to start Paxton AI?
Tanguy Chau (00:49) you
Well, I’ve been involved in the legal tech industry for the last decade, mostly as an investor, both an early stage investor in Aaron Cloud, a contract lifecycle management company, as well as Case Text, a legal research company that was acquired by Thomson Reuters. And so with the advances in AI, and especially with
I was really thinking where do we see the largest level of disruptions happening and there’s a natural fit with the legal industry because you know the ability to read, consume and understand large amounts of text-based documents is a direct and perfect use case for the application of large language models ⁓ and so with that desire in mind
I wanted to make the work of lawyers and attorneys easier, better, faster by applying large language models to their work. And this is how we decided to launch Paxton in 2023.
Federico Ramallo (01:57) Great. I was reading based on your website that legal research is a big challenge for law firms. Can you elaborate a little bit more on why?
Tanguy Chau (02:08) Yes, Paxson started focusing on legal research first, but we’ve evolved since to do a lot more than just legal research. So we handle legal drafting, document analysis, being able to ask questions around your entire matter. But the starting point was legal research for us because…
know, the market is really mostly controlled by two large incumbent, Westlaw and…
Nexus, Nexus, those are very expensive licenses. They’re both very expensive as well as they require you to sign for multi-year contracts. And so it makes it very unfriendly for a small and medium-sized law firm to be able to use legal research. The other reason why legal research was a really important place for us to start is that the state of the technology ⁓ in legal research was particularly
I would say outdated. In order to do legal research and use the incumbent products, you needed to know Boolean queries. So you would have to put together a series of words and connect them through Boolean connectors. And so it’s a really cumbersome experience. Lawyers went to law school in order to practice law and present clients. But in reality, when they do legal research, they just need to learn an entire new language, which is
in order to perform legal research, they need to read all of these materials. Often time, the cases that the incumbent platforms would source would not be, you know, they would spend a lot of time on legal research. And I think that just like, you know, there’s an evolution of…
doing research generally using Google versus using AI models the same thing applied for the legal space where there was an expectation that you you didn’t want to just be served 10 20 or a hundred different links for you to read you wanted the AI model to really be able to do a semantic search and and compose Read the documents and compose the answer
for the user. And so this is why we believe that legal research was a good starting point for us two years ago. But again, as I said, the advantage of our platform is we have a comprehensive legal research component of the platform, on top of which we started building models and capabilities. Because if you want to really help
lawyers, to do better drafting, to better understand the cases, it starts with knowing what the laws are, knowing what the statutes, the court rulings are, and then on top of that you can really build a comprehensive product. And so the starting point was legal research, and then we started building more more advanced features on top of that.
Federico Ramallo (04:59) Right. Right. Yeah, I can see how lawyers would have issues with a strict search syntax. Something similar happens in the software, in the engineering side, where ⁓ SQL became the de facto query language for databases. ⁓ And it has a lot of, you know,
Tanguy Chau (05:07) Yes.
Yeah.
Yes, that’s right.
Federico Ramallo (05:23) It’s very powerful but also very complex to use. you know, as a software engineer now we have to also learn that language and it becomes, you know, problematic, And then, and particularly for junior developers, right? It takes more effort to learn, right? ⁓
Tanguy Chau (05:41) Exactly,
think what you’re saying is right on point. You’re seeing this across multiple industries. Here we saw that in the legal industry, but as you said, it happens in computer science, happens in accounting, it happens across multiple, this transformation happens across multiple industries. Absolutely, right.
Federico Ramallo (06:01) Yes, and the intentions of those languages are good, syntaxes, languages, it builds a barrier for, in particular, lawyers to be able to use them. So I can see how being able to describe what you want in a natural language is much more efficient for lawyers. I can understand why you chose legal research.
use case as the first feature to build, right?
Tanguy Chau (06:29) That’s right.
Federico Ramallo (06:30) So you talk about legal research. What other features will help lawyers to be more efficient?
Tanguy Chau (06:36) Yeah, so legal research is only just one step of the legal workflow. A lot of the steps that a lawyer goes through relates to understanding the case or understanding the matter. So being able to upload your client documents in an environment that is a safe environment that has passed SOC 2 compliance, ISO 27001.
HIPAA compliance that you know where you know there’s no training that is happening on your confidential client data. It’s really important. But that step is really being able to use an AI system on a firm’s own files and materials in order to be able to answer questions, elevate facts, find the needle in the haystack type of analysis. So that’s…
that’s a really important part of getting familiar with the case that the lawyer has. And then from that, being able to to rapidly draft motions and complaints or legal language for contracts and transactional document settings. These are also…
tools and workflows that are needed for attorneys. So to sum it up, the first part is legal research, to be able to simply ask a legal question, and for Paxton to be able to retrieve information from its own legal knowledge. Second aspect is being able to query, understand, extract documents and files.
and insights from a large amount of information, which is data extraction, the document extraction. And then the third one is use the legal research that you’ve done along with the information that you have around your facts in order to draft the language that is necessary. So drafting complaints, drafting motions, drafting new contracts based on the context of the matter.
Federico Ramallo (08:26) Right. Can we talk a little bit about hallucination? Because I can see how using a generic LLM, I ask questions, then I get sometimes the LLM hallucinates. I can see how that could be a big issue in a legal context. What does that mean?
hallucination impact in the context of legal AI.
Tanguy Chau (08:47) Yeah, think that’s an important question. And something that is specific for the legal industry is that hallucination can mean many different things for a legal AI system. First, it could mean
did you invent a case that didn’t exist? in truth, that’s actually really very, very common when you use chat GPT or just a foundational model, because it’s part of the features. It’s not a bug for the foundational model to invent.
words and languages. That’s how they are built. That’s what they’re built to do, to predict things and to invent the next series of words that come up. But for our industry, it’s a real problem because lawyers need to be grounded in facts and in truth. And so that is one type of hallucination, which means referring to a case that doesn’t exist. But that’s not the only way in which
AI models can hallucinate. For example, a different way that a legal model is considered to hallucinate is if you cite a court decision but the understanding of what the court
decision says is misunderstood. So it’s a little bit more sophisticated because some of these court decisions are very long and they’re very nuanced in what it says. And so you need to be able to really accurately say what was actually said in this court decision and make sure that it wasn’t misunderstood by the AI system. That’s a second way that an AI system, a legal AI system can hallucinate is something that we spend a lot of time.
Correcting for a third way for example would also be because in the u.s.. We use common law system If you have a court ruling a later court or a superior court may have overturned the case So you can cite a correct case and you can understand that case correctly, but a court at a later date might have overturned
the case and so you should not be able to cite it because that is no longer applicable law in the US. And so that is a third way in which a general AI system will hallucinate because it might cite a court decision from 1800 that has been overturned that’s no longer applicable law in the US but that is actually still on the books, right? And so all of these are examples
of what Paxson does in order to correct and to make sure that when we cite the case or when we cite the statute, we’re citing something that actually does exist, that we understand correctly, and that is still relevant and applicable law. And so now you start seeing why a legal AI system is necessary for lawyers and that using a general foundational model is not sufficient.
Federico Ramallo (11:43) Interesting. I haven’t thought about the use case of getting a decision of return. That becomes a very complex process to find, right? Finding those overturns, right? And maybe there’s more than one, right? ⁓
Tanguy Chau (11:55) Yeah.
Absolutely.
The way that the incumbents do it is a very cumbersome system. It was very important and this is why a company like Westfile, NexusNexus, became so essential for the work of lawyers. But the way that they would do it is they would have manual editors that would read every single case that had been published since and annotate those manually. And it would take
thousands and thousands of hours with tons of people that are manually going through this one case at a time and annotating things very carefully. Now the truth is we’re now able to do this at scale in a manner that is consistent, that is ⁓ reliable, and that is systematized using large language models. So one of the things that Paxton did is very rapidly just go over the
millions and millions of cases that we have collected and then build a shepardization system to see what cases have been overturned, which one you can cite, which one you cannot cite. And we’re now able to do this using large language models, whereas in the past it would have required hundreds of millions of dollars worth of manual labor to do that. And now instead of having to charge our clients,
very high fees in order to access this, we can offer it to them at a much, much lower cost because AI has been able to replicate that work and to do that more accurately than what we were able to do before.
Federico Ramallo (13:30) Right, right. Yeah, I think that common law is an imperfect system, but it is so far the best system that we were able to, well, lawyers has been able to implement, right?
Tanguy Chau (13:43) Yeah, look, I mean, it says that the law is a living thing and constantly evolves. so, yeah, it’s, I don’t know that there is such a thing as a perfect legal system, but it is the system in the US, and so that’s the system that we’re working with. And it’s been like this in the UK and in many other former Commonwealth.
Federico Ramallo (13:57) the
Tanguy Chau (14:06) countries.
Federico Ramallo (14:07) Right, right. Yeah, yeah, my comment was more on the philosophical side. So yeah. So how does Paxson ensure 100 % citation accuracy?
Tanguy Chau (14:12) Thanks
Yeah, so we talked about just three of the ways in which an AI system, a generalized AI system can hallucinate. So for each of those, the way that we solve for this, first, we ensure that we have a comprehensive.
database of all of the statute rules, regulations, court rulings in the US that is kept up to date.
that is constantly updated. And so when Paxton refers to a case, it’s not inventing from a database. We’re constraining it to search and cite from a database of cases. So you know that the cases that we cite are part of this database. And the way that we expose this to our users is they can very simply…
click on the link and see the actual case and where it was pulled from, which court it was pulled from. And if, you know, and so this is within the Paxton system, but if they want to go, they can click on that link further and go directly to the source and see that in fact, the Supreme Court of the United States has published this. Here is where it’s, where that decision lies. So that’s how we solve for the first type of hallucination. The second type of hallucination is a proper
categorical, it’s a proper classification of what the cases discuss and how what it says and what it doesn’t say. So summarizing and understanding what were the key elements of the case. And then the third one is this process of shepardization. So we have our own data set and our own database where every single cases that we have has been properly shepardized where we look at the history and future decisions that cite this
and make sure that the case is still applicable and good law.
And so these are just, again, three of the ways that we ensure proper citation accuracy. So making sure that we cite the case and we display it, that it’s properly classified as to what it means and what it says, and that it hasn’t been overturned by future cases. And so that’s the work that we’re doing as it relates to legal research. But again, the platform has now evolved well beyond
good research to do a lot more than that.
Federico Ramallo (16:40) Right. Right.
You talk a little bit about this, right? But I want to give you the opportunity to expand a little bit more on what makes Backstone AI different than other AI applications.
Tanguy Chau (16:55) Yeah, I think that because the legal industry is such a perfect use case for large language models, we see a lot of different competitors and different startups across different practices. But really what makes Paxton really different is that the core legal knowledge on which we operate is not something that we derived from a foundational model. It’s something that…
we built internally ourselves. So if you’re saying like, hey, I want an AI legal assistant, wouldn’t it make sense that this AI legal assistant has its own knowledge of the law? Otherwise, you’re just using a…
an AI model that has no inherent knowledge of the law. just relies on whatever Chad GPT says. And so there’s no like check and balances. And I think that’s a real, real point of differentiation is that for us, we started from saying, if you want to build a proper legal AI system, you need to understand the law as a starting point. And that is something that is very unique to Paxton, that it’s unique to Paxton, but it’s also something that’s tremendously difficult and
takes a long time to build. And then from there, we now have the right foundations onto which we build more advanced capabilities like AI drafting and…
and legal analyses and motions drafting and all of that. But it starts from having an accurate understanding of the law.
Federico Ramallo (18:23) You talk about the use cases. Can you tell us a little bit about the type of teams that could use Paxton? Can a solo lawyer use Paxton or it’s only geared towards larger firms?
Tanguy Chau (18:39) Yeah, so that’s another point of differentiation for us. believe that indeed we’re all in the transition of building AI products, right? And the way that lawyers or attorneys do their work five years from now is going to be completely different than the way that they do today. But at the core, I believe that the products that will win other products are the best product and that the way that you build the best product is you make it available to as many
people so they can test it, they can break it, they can tell you what works, what doesn’t work. And so our approach to doing this is that of a product-led growth company as opposed to a sales-led growth company. And so what it means is that in order for you to access Paxton, it’s very simple. You just go on Paxton.ai and you click start a free trial. Immediately you have access to the tool. You can start benefiting from it. If the tool performs well and you like it, fantastic.
Pay for it, great. If it doesn’t serve your purpose, no problem. You can cancel anytime. But the point is that it’s made these legal AIS, it’s PACS and it’s…
available to anyone that wants to start it. don’t need to enter a very expensive multi-year agreement or sign up your entire firm or have a budget that is an enormously expensive budget. You can access the tool. The result is that now, Paxson is deployed in multiple thousands of law firms across the world because it’s so easy.
to use and the result is for us is that we get feedback very rapidly from many different use cases we understand and can build the product more rapidly.
Federico Ramallo (20:24) Right. Yeah, I can see I can see how this This this tool will benefit a lot for ⁓ the small small lawyer firms This will be a game changer for them,
Tanguy Chau (20:36) It’s a game changer for small law firms, but also for us as we’ve built a tool that is a more more sophisticated tool thanks to the work and the feedback of small law firms and the many, many thousands of attorneys that are using PACSAN every single day. Our product has become very sophisticated and now is already being deployed across…
and MLAW 100 firms, firms that have multiple billions of dollars in revenue because the product has been battle tested by so many people and now it has become reliable, sophisticated, and is able to handle the more complex workflows that larger law firms typically handle.
Federico Ramallo (21:20) How do you see AI changing the legal industry in the next five years or so?
Tanguy Chau (21:26) I mean this is incredible. This is one industry that will be entirely transformed.
It’s to me, it’s the same thing as like asking how do you think the computer is going to affect the personal computer is going to affect the legal industry back in the 1980s or 1990s. I think the answer is that in the next five years, there will be no one that does not use AI in order to do their workflows. It will be impossible for you not to be a practicing lawyer that doesn’t rely on an AI system.
It’s just, you know, the gain in efficiency are so vast and the gain in accuracy and are so enormous that it would be almost an ethical breach of duty to not use the best AI systems out there in order to serve your clients, in my view.
Federico Ramallo (22:21) Yeah, I think we are in a similar situation like in 1999 where there were so many, know, the dot com era where, you know, you could build so many new products that were not possible to build before. I felt that the same thing is happening with AI right now in so many industries, right? Not only on legal, but on so many others. There’s so many green.
Greenfield opportunities that were not possible before, right? And the amount of tools that we’re building around AI that will help us ⁓ build these new startups are getting better and better every month. In the last year, I’ve seen so many new products, new tools that allow us to leverage better LLMs, right? That it’s getting…
the speed of that development is getting much faster than in 1999. as you said, it’s hard to say where this is going to go, but I can see that a big revolution is happening, right?
Tanguy Chau (23:23) Yeah, absolutely.
Federico Ramallo (23:24) So going back to another question, what advice would you give to law firms that want to start using AI?
Tanguy Chau (23:32) Okay, it’s really interesting because especially in our industry, like the legal industry, there were so many lawyers that were sanctioned for using Chad GPT. you know, and this hits the news. like, you know, this lawyer was sanctioned by this judge because they cited a case that doesn’t exist. so, you know, and so the takeaway
The first takeaway for the industry was AI is bad. Don’t use AI. You can’t rely on it. And I think that has set us back a little bit. But the truth is they were just using the wrong tools. They were using a general AI model that wasn’t a legal specific AI model. And so I think that really the advice I would give is you have
So first is use a legal specific AI model or legal specific AI tool. If you are a professional cook or if you have to use the right tools, you have to buy like a proper sushi knife, you’re not gonna…
cut, you’re not going to be a great sushi chef using a hatchet. So use the appropriate tool. For attorneys, they are legal AI tools. Pakistan AI is one of those that are built specifically to address your needs. So that’s the first advice I would give. The second advice is it doesn’t have to be expensive. It does not have to be expensive, or it does not have to be complicated to use.
go on the paxton.ai and you can immediately access it and try it for yourself. So that’s the second advice is that you don’t need to jump through hoops and go through complicated sales process. The third thing is, and it probably should have been the first thing is make sure that it’s because you’re representing clients. So
Whatever tool you use needs to have basic level of safety of SOC 2, ISO, HIPAA compliance if you’re dealing with medical records and all of that. None of that is safe if you’re using a general AI model. You have to understand why does it cost like 20 bucks a month or however much they charge for the models is because they take all of this information.
and then they train their model. That’s a real problem. That’s a real problem because you’re sharing confidential information with them. Paxton doesn’t do that. There’s no training on user uploaded data. And so that is the most important thing is that it has to be safe for your use. that’s why legal AI models are really important.
Federico Ramallo (26:04) Yeah, I mean, I can see how a lawyer will be sanctioned by using a judge’s GPT. But I think that if we can unpack that a little bit, I think that part of the issue is that the lawyer sent reference a case that doesn’t exist. So the lawyer didn’t do the homework to make sure that that information is accurate, right? And that’s a big no-no.
Tanguy Chau (26:21) Yes.
Federico Ramallo (26:28) right, regardless of using ChachiPT or not, right? the…
Tanguy Chau (26:32) But you know,
it’s funny, it goes a little deeper than that. Because they did the homework in a sense. They’re asking ChatGPT does this case exist? fact, ChatGPT says, yes, it absolutely does exist. Are you sure? Yes. Can you tell me more about this case? And it makes up the story. And then afterwards, when the judge, the opposing, when the defendant counsel says, we cannot find the case.
Federico Ramallo (26:45) Hahaha
Tanguy Chau (26:54) They go back to ChatGPT and say, hey, where did you find this case? And it tells you, I found it in this thing. It’s not that they didn’t check. They checked with ChatGPT, and ChatGPT was doubling down and telling them, no, it’s absolutely true. Trust me, it’s true. Trust me, it’s true. And then it’s only until this whole thing unravels that you realize that.
It’s more complicated than that. People think that hallucination is like, oh, lawyers just didn’t do their job. It’s not that. It’s not just that. It’s that chat GPT is a very sophisticated liar. It will tell you that it exists. It will make up a lot of fact. It will sound so good about it, right? That it’s like a professional sociopath. It’s just like, that’s what it was.
Federico Ramallo (27:30) Ha ha ha ha ha!
Tanguy Chau (27:44) I don’t think that it was just as simple as the lawyers were lazy and didn’t do their job. They actually just didn’t know what they could trust and what they couldn’t trust. That’s really what the difference is. And so the takeaway is you cannot trust Chad GPT, but you can trust Paxton. That is the difference.
Federico Ramallo (28:05) Yes, yes, I see what you’re talking about. My position on that is because when you get into a car, even if you’re driving and somebody jumps in front of you, you’re still liable for the person, even though you’re not guilty because you didn’t want to kill the person. But you’re responsible because you decided to get into the car and drive the car.
Tanguy Chau (28:25) Yeah.
Federico Ramallo (28:28) I feel that something similar happens here with Chachi P.T. you know, because the tool is so complex that, and as you say, can lie to you, You know, even though the lawyer thought did all the homework, you know, at the end of the day, he’s still responsible because his signature is, you know, on the document that he’s presenting, right? ⁓
Tanguy Chau (28:46) Yes, absolutely. That’s
right.
Federico Ramallo (28:51) Yeah,
yeah, yeah. But I can see that how the lawyer will trust the LLM because, as you said, it’s very good at lying to you, right? And that’s the challenge of using ChachiPT. And the thing is, you don’t know when it’s telling the truth and when it’s not, right?
Tanguy Chau (29:07) Yeah.
Federico Ramallo (29:17) I can see why Paxton AI would be a great fit for lawyers because then they have a higher level of trust on the tool, right?
Tanguy Chau (29:27) Yes.
Federico Ramallo (29:28) Great, jumping to another topic. You’ve been an investor in Ironclad and other tech startups. What patterns do you notice in successful legal tech?
Tanguy Chau (29:39) I think that there is something counterintuitive in building a startup, which is investors are ready to ask you, how big is your market? you know, like, what are all these different workflows? And they want to know that it’s like a really big outcome and that you can do everything for everyone. Because that’s the only way that an investor makes money, right? Like if you just become a like…
10, 20, 50 billion dollar, 100 billion dollar company. The counterintuitive part is that that’s what the investors are asking you. But for you as a founder, actually I think it’s quite the opposite. It’s narrowing down. It’s making things much, much more narrower. And just to do one thing correctly. So I find that a lot of startups just end up failing because they can’t…
reach product market fit. And they can’t build a simple, one simple thing for one type of user. And so they never really quite satisfy someone. And so for us, you know…
There are many different types of lawyers. There are many different practices. There are lawyers that work in-house. are lawyers that work at small startup. There are lawyers that work at big companies. do immigration law. They do security law. It’s super, super difficult to build an AI system that serves everybody at the same time. So my advice here is…
Federico Ramallo (31:04) Right.
Tanguy Chau (31:05) that in order to find product market fit, you need to be very narrow. You just need to pick like one lane, and it’s not about the size of the market, it’s just about getting the first workflows and the first use cases correctly. And so oftentimes if you’re not getting traction, it probably means because you’re trying to solve too many problems for too many people, so just pick one of them. It does not matter if it’s a big market or not. You can just do
the next, you’ll be able to work on the next iteration later on. But you just have to get workflows to understand the workflows and just pick one ideal customer profile and just satisfy that person. That would be the pattern of where I’ve seen successful…
successful companies. So it’s narrowing the problem to one single problem first before taking on too much.
Federico Ramallo (31:59) Right, right. can see, I’ve seen that happens as well. I mean, there’s so much, you know, news coverage or eyes on the successful startups, right? Everybody wants to be the next, you know, Uber, the next Facebook, the next Airbnb, whatever, you know, I’m just mentioning random names of big startups, right? Everyone wants to be the next unicorn, right? But because those are the ones that you see on the news and everywhere, right?
But I think that that’s like the football players or the stars, like the movie stars. You see one movie star, you see one football player, but for one football player, there are thousands and thousands of people that almost made it, but never did for different reasons. So I think here happens the same thing.
if the investor wants to optimize their investment and sometimes they’re doing this, ⁓ they’re bidding to multiple startups and they are, they, if they get one unicorn, they don’t care or they care less about, you know, all the other startups that didn’t work. But as a founder, you know, if you can find one single, you know,
one single use case and you can deliver on that, then you can start building with that. You can prove that the product works and then you can scale from there, right?
Tanguy Chau (33:25) Yeah, absolutely. I’m not surprised that as an Argentinian, you went straight to the soccer analogy, to the football analogy. ⁓ But I’m sure it’s normal, and it definitely resonates.
Federico Ramallo (33:34) I
Yes, yes, I actually am the only Argentinian that doesn’t follow football. Can you believe that?
Tanguy Chau (33:45) I’m very surprised. I did not know that existed.
Federico Ramallo (33:51) Yes.
Yeah, I’m a unicorn myself. There you go, yes. Yeah, there was too much violence around football that put me off to go into the matches, right? And that’s kind of why I took my distance, right? And yeah. Yeah, I actually have a very interesting story about that because I was on the bar, right? And then…
Tanguy Chau (33:58) There it is.
Federico Ramallo (34:16) We went, what is the name of the studio of the basketball place? 49th, right?
Tanguy Chau (34:24) 49ers is the American football. I think it’s Golden State Warriors for basketball.
Federico Ramallo (34:29) I don’t remember the name of the team. So I called them the blues and then the reds, right? And they both showed up from different stairways. And we were on the bar and then they were getting into the train, right? And from my experience, when the two teams show up, there’s going to be a fight, right? So for me it was like that.
All of them were agitated, know, everybody was singing and shouting and whatnot. And I was like, okay, we’re going to have a fight. What is going to happen? Right. And then they start jumping one on one side and the other side, you know, shouting things to each other, know, teasing each other. And we, the train started rolling and that was it. You know, they start, they continue that, you know, playful teasing, right? Zero violence. And I was…
surprised because that’s not what I’m used to. So I’m happy to see that that was playful and naive and childish, whatever it is. But it was safe and fun. So one other question about the invest.
Tanguy Chau (35:30) Yeah.
Makes sense.
Federico Ramallo (35:41) you your role as an investor. How do you advise to do customer research in order to find product development fit? Product fit, yeah.
Tanguy Chau (35:54) ⁓ product market fits, right? So I think that our approach is to answer a single question. It’s just trying to say, what do you do uniquely that people are desperate for? And if you can answer that question accurately, I think…
Federico Ramallo (35:56) I’m proud of MarkerFit, thank you.
Tanguy Chau (36:10) there is something, it’s a very difficult question to answer. It’s a very simple question to ask, but a very difficult question to answer. What do you do uniquely that people are desperate for? And so being able to answer that, that’s just the entire process. The way I think about this is,
In the process of growing company, there are questions that are product hypothesis and there are questions that are growth hypothesis. You want to separate the two of them. You don’t want to work on things like growing the company and growing revenues and marketing and sales until you figure out like the product hypothesis, which is like what do you do and for who and why does it matter? Because otherwise you just end up growing something
that is not solving any problem for any specific individuals. And the way that we approach
product hypothesis, which is to answer this question of what do you do uniquely that people are desperate for? Originally for us is through design partners. So we work very closely with certain types of lawyers to understand what their workflows are, to understand what the platform is doing, where it’s failing, and then just really getting a lot of feedback. getting that.
that feedback loop as quickly and as tightly as possible, I think is really important. Because the truth is that I think what makes a company successful in the end is the rate at which they learn rather than getting everything right. Most of the hypothesis that I form are incorrect.
makes it, but I run experiments rapidly with the team. Those experiments don’t cost a lot of money and I have the answer very quickly. And so that is how we try to validate our product hypothesis as quickly as possible. So yeah, those are a few insights that.
It’s the rate at which you learn that matters. You are going to make more mistakes than get things right. And then to separate the growth hypothesis versus the product hypothesis. And that in the early stage of a company, it’s really just about answering a single question, which is, what do you do uniquely that people are desperate for?
Federico Ramallo (38:15) Great insights.
Great, so what’s next for Paxton?
Tanguy Chau (38:21) Yeah, I think that we’ve built a product that’s incredibly useful product for attorneys that do drafting and file analysis and legal research. We focused more and more on certain type of practices.
you know, like securities market and personal injuries and things like that. And we’re, we are expanding rapidly towards more and more practices. So you can now not only just use Paxton, but you can use Paxton across more a larger group of lawyers. And then you can also directly integrate it in the work where you do the work. So integrations with document management system integration with
other tools so you end up seeing and being able to use the tool where you’re used to do that and it makes it easier to use the tool on a day to day basis as opposed to having to change the way that you do the work. those are things that we’ve been spending a lot of time on.
Federico Ramallo (39:33) Very good. Very good. So we’re running out of time. I wanted to appreciate Dr. Tangui to join us today. Any last remarks before we wrap it up?
Tanguy Chau (39:46) No, think, I mean, I really enjoyed the conversation and I hope that some of this resonates with you as well as…
the listeners. Again, the part that I think is nice with Paxton is that you can use it. So go on, try it out, see what we’ve built. I’m sure that a lot of your listener, whether they’re in startups or in different companies, would benefit from
having a legal AI system help them make sense of contracts, help them understand how to operate in their geographies, make sure that they comply with local regulations. And then the truth is that it now no longer has to be a very expensive endeavor in order to launch or to answer legal questions. are tools like Paxton that that makes…
doing legal work a lot cheaper, a lot faster, a lot more accurate, and these tools are really changing the way that the legal industry is operating and that clients of legal firms are also benefiting from that. And so go check it out on paxton.ai and if there are any…
questions, feel free to always reach out to me or to my team ⁓ at hello at paxton.ai. And again, it’s a pleasure to be building in this space.
Federico Ramallo (41:09) Great, great. It’s great to have you here. I truly appreciate it. And we’ll put all the links in the description of the episode so people can go and check it out by themselves. Yeah. Thank you.
Tanguy Chau (41:22) And I wanted
to thank you for making time to speak with us and ask me all of these insightful questions and having a really good time together with this conversation. Thank you, Federico.
Federico Ramallo (41:37) Thank you.