The Law VS AI Slop: How Spellbook Helps Lawyers Win With Co-Founder & CRO Daniel DiMaria - S10E16

0:00There is so much innovation that you can do in legal AI. Staying true to your DNA and just focusing on making sure you're an absolute leader at that one thing that's really important to your customers is definitely a necessary component of our success. I think AI soft tolerance is at an all-time low right now, especially among lawyers. So grounding the reasons in citations, you might want to know why a certain entry was entered and where it came from, for example. So you will get those citations. And on the legal research side, it will tell you specifically where this came from. It's not going to be just, this is a law, right?

0:33It's going to be, this came from this statute. And here's a link where you can find it and verify it. On today's Legally Speaking podcast, I'm delighted to be joined by Daniel DiMaria. Daniel is a co-founder and CRO at Spellbook. As a lawyer, he helped build the legal AI tool now used by over 4 ,000 corporate lawyers and law firms. Daniel, having worked in the legal industry firsthand is passionate about driving the goal of making lawyers' lives easier and more enjoyable through technology. So a very big warm welcome to the show, Daniel. Thank you for having me, Rob. I'm excited to be here. It's an absolute pleasure to have you on the show.

1:09Before we dive into your amazing career and all the great stuff you and the team over at Spellbook are doing, we have two very serious icebreaker questions. Firstly, what's your favorite beverage? And secondly, what's your preferred choice of footwear on a typical workday? Yeah, great questions. Preferred beverage. I'm going to do two. If I was going alcoholic, it's going to be a fine Pilsner beer. That'd be my choice. When I lived out in California, there was this local-ish brewery called Trumer, and that was my favorite one. Trumer Pils, not sponsored, just excellent Pilsner. uh i don't know i'm non-alcoholic like ice water you know filter we get a lot of chlorine here in florida so it's got to be filtered and uh footwear when working from home just socks i like steve job maxed my sock drawer it's only like black ankle socks for the last like 10 years and if it's gonna be at the office probably like basic reebok sneakers or something good choices and we agree with those and with that we can move swiftly on to talk all about you so to begin with do you mind telling our listeners just a bit about your background and career journey?

2:17I got into, you know, legal tech and spell book pretty, pretty early in my career, I guess you could say. I went to law school in Canada, where I grew up, I practiced for just over a year before getting into tech. And that's when we started the company that was rally, which eventually became public. Now, we've been at it for a long time. We started rally back in 2018. It's been a long journey, especially in the pre-LLM side of the company. And yeah, so really I've been at the company for most of my career, you could say. Initially in sales, I've done every sales job you'd see in a SaaS company.

2:50I was the first BDR, the first AE, the first sales manager. And yeah, I mean, that's a short view of my path here. Did you always want to be in the law? What inspired you to move into that space? Yeah, you know what? I think I wanted to be a lawyer for all the wrong reasons. I like, like Suits was big when I was in high school. the show. And I'm like, yeah, that's really cool. I studied philosophy in undergrad, which I really enjoyed. And I did consider doing that professionally as an academic for some period of time, particularly studying continental philosophy and existentialism and stuff. It was awesome.

3:22I still pick up some books around in that space now and then. But yeah, I mean, I thought being a lawyer would be cool. I thought I'd walk around in a suit and do important things. And it was very, very naive. But I mean, as a high schooler, I guess you kind of are. And I mean, you know, very quickly in law school, I summered at a small firm after 1L and I saw what lawyers actually do. And I started to think that I, in fact, did not want to do that full time for the rest of my life. That kind of planted the seed early on in law school. And then, of course, when I graduated and article and experienced what lawyers actually do firsthand, I kind of knew that I wanted to do something else.

4:04So, I mean, you know, getting into Rally, which eventually became Spellbook, was probably equal parts knowing that I didn't want to be a lawyer full time and seeing what lawyers actually do and being 100 % sure that there was a better way with technology. So that was probably the two main pieces. Yeah. And you mentioned a couple of times, obviously, Rally becoming Spellbook. For folks who may be less familiar of that journey, why did you decide to change the name? Just give it a bit of context and a bit of journey around that. So in 2018, we launched Rally, Scott, Matt, and myself, our co-founders.

4:37And what we were trying to solve was the contracting problem. We saw how lawyers actually worked in contracts. There's a lot of Microsoft Word splicing, copying and pasting, searching old documents. And it just felt so odd given where technology was at the time. I mean, we're millennials, we grew up playing Starcraft and interacting with applications and stuff. And it was just odd to be in Microsoft Word late at night, just like splicing documents, when there surely should be a better way if there were apps that could do some of that. Now, these are pre-LLM apps. So we attacked the problem by building a template-based automation system, ultimately.

5:15Our initial customers were transactional private practice lawyers, pretty much in companies of all sizes, large law firms, mid-size, a lot of very small ones. And we did solve the problem to some extent. There were a lot of issues with that approach. Some other tools at the time were available, Contract Express or Hot Docs, some of the listeners might know. We were kind of in that space. We're a little more lightweight and web-based. The UI was more, I guess, enjoyable to use in some ways. But the issue was, too, the time to value problem. You needed to configure your templates to be able to enter the questions, generate the output documents, and the flexibility problem.

5:52I mean, every legal matter, even if you have the most on-rails use case, there's always additional context, there's always nuance, and you need to edit the template after the fact a decent amount to get it to a point of usefulness. So we encountered a lot of the problems that you'd experience as a SaaS company with those two issues in the product. We had churn, adoption was a little bit slow. Now, we were growing. There was product market fit. We had some very happy customers at the time, but we really came to understand the contracting problem. And we knew that our current solution in Rally was not the best fit per se.

6:28And, you know, fun fact for listeners who didn't know this, GPT-3 was released in like 2020. Okay. It was available as an API well before ChatGPT. I remember in like 2020, 2021, like developers were talking about this. And it wasn't abundantly obvious that, you know, this early, good, large, large language model would fundamentally change the way that humans interact with text for the rest of history. It wasn't obvious at that point in time. Now, we were experimenting with GPT-3 like early, early. I remember we were building on it in 21. We launched internally Spellbook in, I want to say, like first half of 2022 and put it in front of our rally customers.

7:11at which point I probably had a few hundred at the time. I think the company was maybe 20 people or so, so we were fairly, really long. And we just saw the dopamine hit, the dilated pupils. They're like, wow, the experience of just seeing text be generated in a Word document, knowing the context was just totally foreign to anybody who had not encountered large language models before. It was absolutely wild. And to tell you how simple Spellbook was, it was simply a cursor that you drop into a Word document and click go, and it would generate a few sentences. And that was just like so revolutionary to the pre-LLM mind to see.

7:47So, you know, that experience, we kind of knew at that point, we're like, okay, you know what, this might be the next level for our company. And the story goes, I'll say briefly, we launched Spellbook in earnest a few months before ChatGPT got launched. The timing could not have been better. ChatGPT got launched, lit the world on fire. Everybody was totally aware of it. and we were already in position we already had you know seo and ads around ai for legal and then eventually you know chat cpt for legal and uh we had a lot of a lot of interest and that basically caused us to rebrand in early 23 and uh yeah we've been spellbook ever since yeah and you've been flying ever since and even from the start you know you've had great traction like you say you've had great product market fit and i know you put a lot of time and energy into that we're in 2026 then for folks who are less familiar with spellbook maybe hearing about it the first time how would you describe what your problem you're solving within a couple of sentences so it really kind of hammers home for folks?

8:42So Spellbook is a, you know, end-to-end, as much as end-to-end can be, a legal assistant, legal AI assistant, built for private practice, transactional lawyers, and in-house lawyers for pretty much any contracting workflow. The system lives, it wedges right into the workflow, it lives in Microsoft Word, and there's a dedicated web app for the multi-doc and knowledge-based features and workflows. It can pretty much help with anything related to contracts. If you're reviewing documents, either based on a rule-based playbook that you configure in the app or just instructing the AI to review a document, it can handle that.

9:16It can redline documents for you. It can respond to redlines. Any drafting workflow, whether it's drafting a document from scratch, editing a document from a precedent set, for example, can automatically do that just based on new instructions or a source file like email or term sheet or whatever. Draft clauses for you automatically, like any good generative AI system can. It has an assistant built in that can, you know, honestly, you can handle, you can, you can throw any task at it in Microsoft Word and it will handle it. Fully aware of the document, the party you represent, your knowledge base, your past contracts, the clauses you like to use.

9:51It is legal research available, you know, right in the app itself. And then there's a benchmarking component where you can run a comparative market analysis of the document. You can benchmark it against other document types. And then there's multi, multi-step workflows. So it's a very full featured contracting app. Yeah, built primarily for, again, in-house and product practice transaction holders. It's phenomenal. As I said, I've had the benefit of seeing the product and the work that you do. And, you know, we obviously had Scott on the show before. And yeah, I think it's phenomenal, the quality and how client-centric you are and the feedback that you get from people from the market in terms of being really, really helpful.

10:26You know, not one of these sort of tools that is taking people away. It's pretty friction-free and good to use. Okay. You talked before about the rally to spellbook journey and you said at rally you were sort of hundreds of clients now you know spellbook is thousands of legal teams using the product exactly and that's just a great case study and great success story could you talk us a bit about that spellbook growth specifically and any tips for maybe aspiring legal tech entrepreneurs or people looking to build out their own platforms and how you've been able to achieve that growth yeah that's a great question It can go a lot of ways.

10:59I'll speak to what I think the necessary component of Spellbook was to its success. And that would be finding a problem that you have 100 % conviction of and really focusing on that and letting that be the core thing you're trying to solve. Like Spellbook does a lot right now. But the central DNA has always been making contract workflows significantly easier and more enjoyable for lawyers. You know, we were pretty early with LLMs, which was good. But I think for us, like we saw that LLMs were just so flexible and good at dealing with the friction points that we encountered with Rally, you know, which was the flexibility problem and the time to value problem.

11:43I mean, LLMs deal in the currency of text as lawyers do. And we always thought of the contract problem and really just focus on that and iterating on that. You know, there's going to be, you know, other competitors or other tools out there that might, you know, come up with a new feature set that transactional lawyers and house lawyers, you know, want to use for a given workflow. But we always want to be leading on the contracting side, the editing and reviewing of contracts. So I think, you know, there's a lot that goes into the growth of a SaaS company like ours. But staying true to your DNA and just focusing on that and making sure you're an absolute leader at that one thing that's really important to your customers has definitely been a necessary component of our success.

12:25And I love that you said that because I talk a lot about that when people ask me about podcasting growth. And I always talk a lot about TOI, which is topic of influence. what is the topic of influence your show wants to be lasered in and focused on when people are going to chat gbt now or google historically to put in what are those one or two words that they want that pain point relief around that you're delivering on and for us obviously we've been all around legal careers and large part of that is technology in recent years and where that's shaping and changing careers etc etc but we focus on that and you know mentor said specific is terrific right the more you can be specific in times your audience and double down and deliver the better.

12:59So I love it. And clearly it's working for you guys and it's been a huge, huge successful. So congratulations. Just talk a little bit, you touched on some of the features because I know you do a lot more. You've got review, draft, ask, benchmarks, associates, you know, it has grown. So again, folks new to this today and getting excited about hearing about Spell, but just talk us through, you know, maybe one of those features, a few of those features, how they work and, you know, maybe some of the most used of those features and the benefits lawyers are seeing. Yeah, totally. I'll talk about a newer one.

13:27It's more interesting and top of mine and then I can maybe touch on some of the others. So we're in the process right now of releasing a new feature called compare to market, which we're really excited about. It's going to be, I think, particularly used by in-house lawyers. Definitely some of our private practice ones might, might have an interest in it, but it's more in-house focused. So, you know, right now AI, right now there's a problem when negotiating contracts around the, we call it the, this is market problem. It's like, okay, we're negotiating with the counterparty. You look at a specific term and you say, hey, we're going to use this term because it's market.

13:58And the counterparty goes, no, actually, it's not market. I think this is market. And it's kind of like an impasse because there is no objective framework of what market is. And this is normal because to understand market, you need to analyze contracts in private repositories and CLM systems and DMSs of private companies all over the world. And it's really hard to do that. So market data is quite opaque. Now, these pre-trained models that we use, they're trained on a ton of data. You can throw a question at it like, hey, is this term market? And it'll give you an answer, but it's going to be maybe an AI slop answer.

14:33It's not grounded in anything. It's just generating a tech, giving you an answer to serve you, right? So the way that we're dealing with a this is market problem with LLMs, I mean, Spellbook does millions and millions of contracts per year, and it's growing quite fast. So we're taking our contracts that are run through Spellbook, our customers' contracts. We're anonymizing the contracts, stripping the deal terms, and then categorizing the deal terms based on the contracts type, the industry, the jurisdiction, other metadata factors like that. And we're pooling that in a RAG database. So basically these deal terms and the metadata embeddings that go into a vector database.

15:10And what our customers can do, they're starting to be able to do, is when they encounter a contract, they can run a comparative market analysis that's based on the millions and millions of contracts that have gone through Spellbook. Spellbook will find the appropriate pool of contracts that kind of fit the one they're comparing it to and give it a term by term breakdown of how each term fits within the market framework that was found by Spellbook. it'll tell them which terms are outside of market that benefit or disbenefit their party. And then they could automatically with the app redline a term and provide an explanation as to why it's now within market, for example, and actually provide like grounded data on this.

15:52So it's a very grounded data driven approach to understanding what market is for a given contract and letting that be a pretty strong reason as to why a certain term should be revised within a contract when you're in negotiation. So that's certainly one that we're really excited about. Of course, we're leveraging, you know, RAG or Retrievalog Metageneration for this, and it's working really, really well. Some of the features that we're really seeing a lot of use from right now that customers are loving at Spellbook, you know, one is associate and the multi-doc drafting. So, you know, as you know, like worst kept secret, lawyers don't draft from scratch, even if the bills look like they do.

16:29You're always starting with a template or a precedent or a previous matter that you did. And Spellbook makes the drafting from precedent experience significantly easier. With Associate, you could pull a set of documents that you did for, say, a similar deal or matter a few months ago. Just like, you're like, oh, I think I did something kind of like this three months ago with this client. So you pull those documents and put them associate, which is basically like a sandbox multi-doc environment, you dump them in, and you can then upload any additional context that you think associate needs to be able to edit these documents.

17:05It's almost like you're throwing it into a blank slate junior associate's brain, like here, I want you to load all this context, I'm going to give you a task. So you load the precedent documents from the previous deal, you upload maybe a term sheet and pull it over 10, maybe just like an email correspondence that in, despite all the noise in the email correspondence, you can distill all the facts needed for the new matter you're working on. Just dump it all in there and say, hey, listen, this past precedent set that I did before, there's some good stuff in there. It's not perfect. I want you to use this to generate draft documents for this new matter.

17:33You'll find all the information you need and much more in this email correspondence go. And then it will start cooking and natively edit all the precedent documents in DocX in the app. And you can then download them and accept all the changes or accept them right in the app itself. So, um, you know, that, uh, experience, which lawyers have done and are doing is, is very monotonous administrative. I mean, there's not, there's nothing like legal judgment involved, but it's very much like a text synthesization experience. And, uh, that's one that Spelbook does particularly well. Um, and we think, yeah, we think that the future of your acting is going to be, is going to be largely done like this where you kind of, you, you, you tee it up and then you let the let the ai go and do its thing so you're not spending your time doing those those mechanical tasks i think that leads nicely on to my next point around teaching ai actually because you've shared that spellbook aims to teach ai to sound like lawyers not replace them right why is this important to spellbook i don't think ai is going to replace lawyers firstly like i don't think it's i don't think it's smart enough to do that right um you know i talked before about the the end-to-end with like you know air quotes end-to-end for those who are watching this, like AI is not end to end.

18:49It's middle to middle. Like you need to start the matter with it. You need to like tee the task up and then direct it to go and do the task. And it's going to be really good at, you know, document synthesization, at taking into account, you know, all the data in the document set that you gave it, for example, so it can go ahead and like do a task, whether it's generating a summary, a table, editing documents, whatever, doing research, whatever it might be. but you need to guide it through this experience. And it needs human intervention at points. And we try and train Spellbook to prod the user with like, hey, like I'm about to do this.

19:25I'm kind of in creative mode right now. Can you point me in the right direction? Like associate will often prompt you and say, hey, I actually need you to let me know like which party you're representing in this document because there's a bunch listed and it's kind of different than the others, for example. And it would go ahead and do that. And then it will tell you kind of how long it needs for the next task, for example. So you're kind of like, you know, guiding it around this path toward completion. And once it gets to completion for the AI, you kind of need to do the remaining 5%. That could be like, you know, assembling the documents in a folder so you can deliver to the client.

19:59Like it's not going to be able to do that just yet, for example. So the lawyer is always in the driver's seat. And it's always kind of, it has the full human level knowledge of the matter that it's trying to complete. but the AI is doing the, like the lifting that would be well suited to the AI. So, you know, again, the document drafting, synthesization, pulling out facts, compiling them, that type of thing. Yeah. Yeah, absolutely. I see this sort of, you know, you're, you're running the race. You are the person you need to be running the race, but you've got a rocket on your back, you know, and Scott described it as having like an electric bicycle, which I think is good.

20:34You're still driving it, but you don't need to pedal. Yeah, exactly. And I think that's a good, analogy. Flipping back to when you graduated now, I think I've got a quote here, talking with issues from practice. When I graduated from law school, I didn't realize a large portion of the work I'd actually be doing as a lawyer would consist of splicing, pasting, editing multiple word documents. It wasn't what I envisaged. So when did you particularly realize that the traditional legal workflow was not sustainable for practice? I think it was probably the first real task I was given as a lawyer or article student in Canada.

21:09I had like 13 Word documents open and needed to generate one from all these past precedents that were kind of relevant and, you know, all these documents that, you know, I need to pull stuff from to make sense. And I'm kind of like, wait, like all the data is here. And like the actual task is just like, is just jet polishing this final word doc that I need to deliver. And I just thought that that can't possibly be it. This is like not even a lie. This is true. Like I remember like calling my friends who were lawyers and I'm like, Hey, like, is this what you do? Is there a better way? And the answer was largely like, no, like there's a large part of what you do.

21:47Like, you know, getting to the final documents that is a big part of what lawyers do. And I just found that to be odd. And, you know, again, like I, I didn't, I wasn't like a tenured lawyer with 10 years of practice. Okay. And I wasn't a computer scientist that really understood like the nuts and bolts of text for Doc X. But I had a very strong conviction that this was not the way that lawyers will practice based on the pace of technology that I was seeing just as a millennial who grew up with kind of seeing the analog to digital transition in the world. I just knew it wouldn't be that way. So I'd say the strongest thing that I had going for me at that point was just conviction that this would change and that there would be companies that would deliver this change in some way.

22:28And I wanted to go and explore it. So yeah, I would say that was probably the point. Today's Legally Speaking podcast episode is proudly sponsored by Clio. If your legal management software feels more frustrating than helpful, you're not alone. Many solicitors across the UK delay switching because moving client and case data sounds like a headache. Clio makes it simple. As the intelligent legal work platform, Clio combines context-aware AI with trusted legal research to power your practice. And when you switch, their dedicated migrations team supports you every step of the way, ensuring a smooth transfer of your data.

23:06You'll also get award-winning support 24-5 via live chat, phone and email. So help is always there when you need it. Ready to leave frustration behind? Visit clio.com forward slash UK to learn more and discover why so many UK solicitors choose Clio. Now back to the show. Cool as it sounds, the only constant of business is change, right? Things are going to continually evolve. Things are going to happen, particularly exponentially in the world that we live now with tech and AI. And sticking with AI, we must, of course, talk about governance and security because you've emphasized that governance and security are non-negotiables when embedding AI into contract workflows.

23:45So what should legal teams listening to this be doing to adopt AI responsibly? Yeah, really good question. I think responsibility with AI has, I would say, two different categories, I would say. The first one is responsibility from a data privacy standpoint. You need to make sure that you are maintaining lawyer-client privilege, that there isn't any sensitive data going where you don't want it to go. So our customers, whether it's in-house or private practice, they typically have an IT component or questionnaire that they'll want us to fill out, or they'll interact with our trust portal that will describe all of our security protocols.

24:21and they're always satisfied with what we do. I mean, like, firstly, Spellbook has zero data retention at the LLM level. So there is no LLM training happening. None of your data is leaking to an LLM where it's going to train, you know, any Claude models or OpenAI models or anything like that. The next is around, like, the configuration side. So you can configure Spellbook quite a bit, and that data is going to be stored, of course, in Spellbook servers. We're SOC2 compliant. We're HIPAA compliant. We really do cover that data privacy component of it. So, you know, I think like knowing that is really important.

24:52When you go and use an AI system, you have to make sure that it is secure and that these are the level that you're comfortable with for your firm and your practice. And the way that the law societies manage it is they basically say like, okay, like, you know, the law firms and the lawyers are responsible for security. You know, they're not going to go and audit every tool out there. That'd be impossible. So it is your responsibility and, you know, make sure that that kind of checks the box. Now, the next component, the next category of AI responsibility, I would say, is on quality AI outputs that you can actually deliver.