Former Williams & Connolly Partner Discusses Pivot to Legal Research Startup Founder

The original article was published on Law.com, here.

Legal tech startups are popping up left and right with many founders of these emerging companies having spent most of their careers working at law firms.

Like many of these lawyers-turned-entrepreneurs, Sam Davidoff, founder and CEO of legal research tool Align and former Williams & Connolly partner, felt the need to expand his career path and dove further into tech.

Davidoff sat down with Law.com to discuss his professional pivot to tech full-time and Align's case research tool, which launched for general availability Monday.

The following has been edited for length and clarity.

You started Align while you were at Williams & Connolly. Can you talk about your experience of working at a firm and also starting this company?

I had this idea for a product [in 2019], and this was what ended up being our first product. ... [We] ended up calling it Align Binders to distinguish it from Align Research. It was a digital binders product, and that was really based on my experience as a litigator. I was like, "How come I still have to carry binders around? Why can't I be paperless?"

Also, I have always been a kind of technology guy. ... What I realized in 2019 [was] there's this part in the back of my mind that is like, someday I'm going to do something in tech, and I had this moment of truth or maybe midlife crisis, I don't know, where I was like, “When is someday?” ...

I talked to some of the leadership at the firm, and [said], “I don't want to hurt you guys, tell me how you want me to do it. If you want me to walk out the door now, I can,” and they were amazing. ... They were like, "Well, Sam, let's figure out some way to make it work. So why don't you try staying on part-time?" That's actually what I ended up doing.

Williams & Connolly ended up being the beta testers of the product once we got it built, the original product, the binder product. ... Once we got the first product built and out on the market and started making sales, that turned out to be a full-time job and so actually I went to them and I was like, “Guys I'm done. I can't in good conscience keep working here when I don't have time to bill out. I have to work on this company." That was in 2023. ... It is an amazing firm, and they were super supportive of this whole journey.

Can you talk about what the Align Research tool does and the motivation for launching it?

It's [an] AI-based legal research tool, and it differentiates itself from other things that are out there in two aspects: ... It is meant to do case law research. It's not more ambitious than that. It doesn't write memos for you, it doesn't write briefs for you, and it doesn't do more research than ... U.S. case law research. ... You ask it a question, and it goes away for a few hours and it comes back with a set of cases and it highlights the relevant portions on it.

Anybody can try it for free, and if you want to move into our paid tier, we give a few free jobs per month that you can run. ... This is also the second thing that's a little different from at least currently the way AI tends to be priced.

We are pricing it on a per-research-job basis. It's $100 per research job. I believe a lot of AI tooling is going to move toward consumption-based pricing. We've started to see it already with Legora, [which] recently announced they're doing it, but I just think it's inevitable. Unless the ultimate AI providers—Anthropic, OpenAI, Google—start pricing things a different way, and they really can't because that's the way GPU time is priced.

When you say your product differentiates because it's focused on a very specific task or it's focused on a very specific line of research, I'm wondering if that also plays as a disadvantage because it is limited in its capabilities?

There are times in a case where what you want to know is, "I need to get an overview of the law. ... I need to understand what are the leading cases? What are the strands of jurisprudence that might come into play here?" In the olden days, we would go to a library and find a treatise. Nowadays, these AI tools that can instantly write you a custom summary of this area of the law as Thomson Reuters or Alexis can do with citations, etc.

In my experience, where you are often in litigation is in these kinds of roles where what you want is an answer to find me the cases. This comes up all the time where you say, "Has a judge ever in the Eastern District of Michigan granted a motion to dismiss on this grounds? Has that ever happened? Has that come up? Or has anyone ever argued in bankruptcy in Delaware that you can do X?" Those things come up all the time, both when you want to go to advise a client and you want to be able to say, "Hey, this argument's been made before and it's succeeded," or vice versa, it hasn't. ... Those kinds of, "I need to see the cases," comes up a lot. Actually getting an overview of the topic or even an overview of the topic with some citations as a litigator anyway is very unsatisfying.

I want to know everybody who tried this argument, and how did it come out? And that's where, at least in my experience, the bigger tools, they trade their broad, 30,000-foot view on a lot of different sources, and also, they trade their ability to give a response instantaneously for that exhaustiveness.

This is really the opposite. This thing is not instantaneous. You ask a question, and our average time is probably one or two hours. It goes off, it researches, it comes back. It is not the experience of when the chatbot starts generating text right away or within a minute. But we are getting back to you all the cases in which a court held X, Y or Z, or in which this issue came up in this context and that, at least for litigators, matters a lot.

Anytime you have AI writing something for you, you have hallucination risk. It's not possible to get rid of that. It's just the nature of these things. You can ask them to validate, but as soon as there is text written by an AI, there's some chance.

So, part of my philosophy too in this tool is we don't get into the hallucination issue because we're not letting the AI write anything to you. All it does is it returns actual cases and it highlights things in the cases. It could miss something, sure. But what it can't do is it can't give you the wrong information. It's structurally not possible. And that's really a differentiator. And that's what I found is missing in these summaries. You always have this lingering question in the summary: how do I validate it? How do I know it's true?

Where is your case law coming from, and would you say Align's AI technology is model agnostic?

We use Free Law [Project's] Courtlistener database, and on launch day, that is really where all the case law is coming from. We are also building out some additional databases. There are some gaps in Courtlistener's database. ... We have actually built a pipeline to go out and find some of this stuff actually using AI to help. We are building our own database, although our goal is actually not to keep that private, our goal is to actually give that back to Courtlistener.

We are model agnostic. ... Our pipeline for doing the research actually uses multiple models. It's an AI agent, and so it's going through, and in its process, it's using essentially sub-agents to do different parts of it. In our development process we optimize both first for model results and then for model costs.

Currently, we use a combination of models from Anthropic and Google. We're not using currently any OpenAI models, but we test them, and that would change if we found an OpenAI model that worked well.

This tool does the work that a lot of paralegals and associates do. Could this tool potentially take their jobs?

I would not say it's attempting to replace the jobs, but I would say it's attempting to replace certain tasks and a very specific task. Also, I would say, very decidedly not trying to replace certain other tasks. If you're a junior associate and you get asked by a partner, "Can you find for me every example where an expert in California was disqualified on X basis?"

We are very much what would replace the hours you used to bill going onto your research provider of choice and typing in search and terms of connector things and running 20 different searches through California state and federal courts and winnowing it down and stuff like that to get the collection of 30 cases that may or may be in that area. We definitely are trying to eliminate that 100%. First of all, that part of the job was never that fun to start with, and second of all, clients just won't pay for that anymore. ...

Going in and reading them and now deciding which are the 10 cases that I should show to the partner or put in the memo or cite in the brief—we're very deliberately not replacing that. To me, that's work that should totally stay with the lawyer, and the clients should pay for that time because that's the valuable time you want.