In this episode of The Geek in Review, we welcome Feargus MacDaeid and Nnamdi Emelifeonwu, co-founders of Definely, to discuss how their shared experiences as practicing lawyers shaped a groundbreaking accessibility solution for contract review. Feargus, who is visually impaired, and Nnamdi, his former colleague at Freshfields, describe how their friendship and professional collaboration led to a tool designed not only for those with disabilities but for all attorneys grappling with voluminous transactional documents. Listeners learn that Definely began as a way to help Feargus navigate complex contracts more efficiently, and through iterative prototyping, evolved into a productivity suite that addresses universal pain points in the pre-execution stages of contract life cycles.
Feargus explains that his journey to co-founding Definely began with personal necessity: having gone blind from a degenerative condition by his early twenties, he pivoted from a computer science career at Microsoft to law school, relying on assistive technology and immense personal support. Once at Allen & Overy, the limitations of existing tools became starkly apparent—searching for defined terms meant losing one’s place in a 300-page agreement and juggling layers of nested definitions by reading aloud via text-to-speech. The cognitive load was immense. By collaborating with Nnamdi, who recognized that if a solution could serve Feargus, it would benefit everyone, they embraced the principle of “designing for the edge”—creating a platform that brought definitions, clauses, and cross-references into context without interrupting a lawyer’s focus.
Nnamdi takes listeners on a tour of Definely’s three core components: Vault, Draft, and Proof. Vault functions as a dynamic repository for templates, clauses, and precedent documents, enabling users to pull in the most relevant resources from connected document management systems. Draft keeps the user anchored in the current clause while instantly displaying any linked provisions or schedules in a sidebar, eliminating the need to scroll, split screens, or flip between pages. Proof automates common pre-signing checks—verifying cross-references, punctuation, and legal grammar—to ensure a polished final draft. Together, these tools exemplify how Definely streamlines contract creation by surfacing precisely the needed information in a lawyer’s line of sight, thereby maintaining context and reducing manual navigation.
The conversation shifts to quantifying Definely’s impact on law firms. Nnamdi cites a study indicating that attorneys save up to 45 minutes per day—roughly a 90 percent reduction in time spent on tedious tasks—by using Definely’s context-aware navigation. Beyond hard metrics, the founders emphasize “soft benefits” such as reduced cognitive fatigue, higher morale, and improved client value. To capture these less tangible gains, Definely’s customer success team works closely with firms to customize usage dashboards and collect feedback. Feargus and Nnamdi also reflect on the broader legal tech landscape, noting that firms are experimenting with in-house development, acquisitions, and partnerships. They believe collaboration between vendors and firms will ultimately prevail, as specialized expertise in areas like machine learning ops and user experience is hard to cultivate internally and essential for maintaining cutting-edge tools.
Finally, the episode zeroes in on technical and operational safeguards to ensure accuracy and maintain the “human in the loop.” Feargus describes how Definely uses a retrieval-augmented generation (RAG) approach, chunking and embedding each contract so that any language model query is strictly grounded in the document’s own text. By setting the model’s temperature to zero and building guardrails at every step, they contain hallucinations and ensure that the attorney remains the arbiter of correctness. Looking ahead, both founders predict a rise in agentic workflows—small, task-specific language models that plug into a suite of specialized tools—and a greater emphasis on UX design as software shifts from simple point-and-click interactions to more dynamic agent-driven processes. As the hosts close the interview, Definely’s mission emerges clearly: empower lawyers to work smarter by bringing critical contract information into focus, while preserving the essential human judgment at the core of legal practice.
Listen on mobile platforms: Apple Podcasts | Spotify | YouTube
[Special Thanks to Legal Technology Hub for their sponsoring this episode.]
Blue Sky: @geeklawblog.com @marlgeb
Email: geekinreviewpodcast@gmail.com
Music: Jerry David DeCicca
Transcript
Marlene Gebauer (00:00)
Hi, I’m Marlene Gabauer from the Geek in Review. I have here Nikki Shaver from Legal Technology Hub who has some great insights to share with us. So Nikki, please tell us what’s going on.
Nikki Shaver (00:12)
Hi Marlene, so nice to be here.
So we all know that over the past two years, Big Law has really gotten underway with generative AI initiatives. Mid Law, though, is a somewhat underserved market. For one thing, it’s fragmented. Mid Law firms come in many different shapes and sizes, and different technologies are therefore appropriate for them. For example, you have your boutique IP firms that do high-end work and often compete at that upper echelon of the market. You have high-volume tort litigation firms that focus on areas
like personal injury and insurance, you have full service firms that are considered big firms in their regional localities. These firms are typically not as far along in their generative AI initiatives and strategy, so Legal Tech Hub is bringing resources to them, both on our platform and in person. Starting on June 18th in Chicago, Legal Tech Hub is running a mid-law tech and innovation CLE seminar series for mid-size law firms specifically. These are one day a
events that will provide three to five hours of CLE points, as well as practical advice on developing AI strategies and approaching the transformation of your business strategically using generative AI. Firm attendees will walk away having seen solutions and met vendors that can serve them, developed contacts at similar firm that they can touch base with on their mutual journey, and with real best practice guidance that will help them move forward in a way that ensures they remain competitive in the market.
And our conferences are always priced fairly. These ones, there’s an early bird registration fee of just $149 for the day. But for all those listening right now, if you reach out to us to let us know you heard this, you will receive a 50 % discount. That’s five zero. Email info at legaltechnologyhub.com and mention the Midlaw Generative AI Seminars and Geek in Review to get your 50 % off code. The events are on
June 18th in Chicago, June 24th in New York, September 3rd in DC, October 15th in Boston, October 29th in Miami, and November 18th in Atlanta. can register at lthconferences.legaltechnologyhub.com.
Marlene Gebauer (02:33)
CLE, AI insights and a 50 % discount. mean, that’s a triple win for me. Exactly.
Nikki Shaver (02:39)
What else could you want, right, Marlene?
Marlene Gebauer (02:50)
Welcome to The Geek in Review, the podcast focused on innovative and creative ideas in the legal industry. I’m Marlene Gabauer.
Greg Lambert (02:57)
And I’m Greg Lambert and today we are delighted to have Feargus MacDaeid and who is the co-founder and chief strategy officer and Nnamdi Emelifeonwu co-founder and CEO of Definely Feargus and Nnamdi , welcome to the Geek in Review.
Feargus MacDaeid (03:16)
Okay.
Nnamdi Emelifeonwu (03:17)
Great to be invited.
Thank you so much for the time.
Feargus MacDaeid (03:21)
Yeah, exactly. Thank you.
Marlene Gebauer (03:21)
Yeah,
we’re really happy to have you. ⁓ Feargus, Definely started as an accessibility solution to assist you in reviewing contracts. Is that correct?
Feargus MacDaeid (03:33)
Yes, yes it is. it started. The idea was born from that. know, Nnamdi and I, obviously we’d met each other, you know, at Freshfields when we were working there together. And, you know, I often really struggled with going through the kind of documentation, especially when you’re working like in transactional law, you know. And I suppose it was just one of those friendships where, you know, Nnamdi always took a real interest in.
how it was I did my work. I previously had a background where I had worked at Microsoft for several years and whatnot and had to transition careers into laws for different reasons. But ⁓ I think really ultimately it was those conversations where we’re both kind of looking at the tasks that we are doing and the jobs that we’re doing and how we’re doing them. And Nnamdi kind of putting the challenge to me, know, kind of like.
what would a good solution look like for someone in my position, you know? And I think, you know, that was fundamentally like driven by a sense of equity and fairness, you know, like just, it doesn’t seem right that somebody is intellectually really capable, will be struggling so much in a law firm to do the same job as everybody else, you know? But I think like after then, Nnamdi kind of put that prototype together. I think when he saw that, I think that’s when he really saw also the fact that, you know,
this is something actually applicable to everybody, you know, because when we sat down and thought about it, a lot of the pain points that we were experiencing were the same pain points. It’s just that mine were more extreme because of the circumstances and much more accentuated, you know, so.
Greg Lambert (05:16)
And just
so everyone that may not know your story, Feargus, I know that you are visually impaired. Can you explain how that affected your life in the practice of law?
Feargus MacDaeid (05:29)
Sure.
Yeah, absolutely. like, you know, it turns out impact on my life and law, like I suppose, you know, it kind of goes back a little way, you know, I’ve been going blind since the age of 11. I have a degenerative condition called retinitis pigmentosa. Like I’m 52 now, but by the time I was kind of maybe, I don’t know, 22, 23 ish, I needed some form of assistive technology. Like whether that’s large print or using a monitor and
I had originally done a background, know, BA and MA history, but then, you know, wasn’t able to do that as a career in academia. So because I was able to use monitors, I did a grad dip in computer science and worked at Microsoft for about four or five years. But again, the technology just didn’t exist to support that career fully. And by about like 2005, I’d been married then, text to speech looked like it was kind of coming along, you know. ⁓
So my wife and I was like, okay, we’ll do a law degree. And so I did a law degree at King’s College and the technology just wasn’t really there. So actually she ended up working two jobs and reading all my law books to me. She read and recorded all my notes, everything, you know. And then I graduated out of King’s College, like in the top 10 % of my year and ended up at A &O where I kind of, that’s where the rubber really hit the road. was where the reality really hit me, you know, in terms of
the volume of work, the volume of information and the time constraints and just always trying to find workarounds. my life generally is finding workarounds because I think Nnamdi and I kind of brought that into our mission statement in a lot of ways, know, kind of what are we about? And I suppose it’s partly shaped by my life experience and friendship also, you know.
with Namdeon and how we work together, but it’s ultimately, you know, how do you find the most relevant information in the most efficient and meaningful way possible? Like that’s, that’s kind of like a fundamental thesis of what we do. It’s not just about knowing what kind of workflow or pain points somebody has, but it’s actually how do you address that so that it’s the most efficient and meaningful way? Because
You know, I could take like a hundred steps to solve a problem. But if I can find a way to do something with three or four steps, then, you know, that is actually the ideal. And that, think, shaped a lot. when Nnamdi and I spoke, you know, about the idea that we had, ⁓ a huge amount of that was driven by that kind of philosophy of like, you know, well, if we can solve the problem for me, like if Nnamdi and I can make somebody like me efficient.
getting through the documents, you know, imagine what that does for everybody else. So I suppose another real principle there is that kind of, if you design for the edge, oftentimes you get the middle for free, you know, we ended up with a tool where in principle, it may sound like kind of, it started out as an accessibility driven solution, but it isn’t, you know, it’s actually something that everybody can use.
It’s just that I was the edge and somebody like Nnamdi was the middle and there’s a lot more people in the middle than the edge, you know.
Greg Lambert (09:05)
Feargus, what were some kind of basic issues that you were running into just to both do the job and then also ⁓ ways that if you had an idea that could take those hundred steps down to four or five, ⁓ were there any obstacles in actually implementing those, especially in a large firm like A &O?
Feargus MacDaeid (09:33)
Yes. like to talk about like kind of what my workflow would have looked like before, Definely, it actually wasn’t really that different to what Nnamdi ‘s workflow looked like, you know. So a really good example of that would be if you’re a lawyer, saying you’re looking at a contract and you’re reviewing quite a large transactional document, like Nnamdi and I used to work in banking. So some of those documents could be two, three, 350 pages long, you know.
Typically in those kind of contracts, the way they’re put together is you might have like about 20 or 30 pages of definitions at the beginning of the document. And then sometimes throughout the document, there may also be defined terms actually being created in different paragraphs or clauses along the way, what some people would call an inline definition. So you’ve got to imagine, let’s say,
you’re on page 137, it’s clause 27.3, you know? And there’s some comment or you’re reviewing it or you’re doing some form of analysis on it. And there’s, let’s say, four or five defined terms actually being used within that paragraph or clause. Now, for a lawyer to properly understand that, as everybody listening will probably know, they need to actually know what those defined words mean.
so that they can actually understand the context of the agreement or the clause itself that they’re working on. And the way they used to do that typically would they’d do, let’s say, a control F for the word itself, you know, and they’d end up back up the top. They’d they’d split the screen. Like they’d have the document on the left-hand side with the definition section there and their clause 27.3 on the right side. I think Namda, you used to print paper, if I remember correctly, and have it on your desk, you know?
Marlene Gebauer (11:07)
Yeah.
Nnamdi Emelifeonwu (11:24)
Yeah. Yeah.
Feargus MacDaeid (11:26)
Or, you know, they were like really the methods that everybody used and I was no different. Like I had found very clever ways like with doing control F, you know, especially in the old word like, you know, but one of those things would be, for example, like with shortcut keys, you know, if you did actually kind of a control F B, you would only look for words that were in bold format. So
I knew most defined terms, let’s say were bold, you know? So I’d always do, if I was looking for one of those, I’d do like Control FB and then I’d type it in and that’d bring me straight to where the definition was. But the problem with that is I’m probably back on page eight of the document now and I have to go back to page 137. So, you know, people were always losing this kind of time with just actually navigating the document for the information that’s already in the document, you know?
And I think another issue with that then is, you know, the cognitive load that comes with that then, because for someone like me then, it’s like, you know, if I see a defined term being used and I now have to go to that defined term, I read it with like text to speech or whatever, because I used audio, but it would read back to me. And then there might be another defined term used within it, you know, and now I have to go look that up. And now you’re trying to hold all that context in your mind.
and then go back to the provision that you’re reading on page 137. you know, trying to implement a solution, there was no solution, you know, like that’s why we built what we built, I think, ultimately. Like, I don’t know how many times Nnamdi asked me in the lifetime of our careers together at Freshfields, where he was like, surely there must be something that can make this easier. There has to be something, there’s got to be a better way. And I used to always say to him, look, Nnamdi , like,
Marlene Gebauer (13:17)
There’s gotta be a better way.
Feargus MacDaeid (13:23)
If there was, would know it. know, like I used to know how to do registry hacks with Word to change like the underlining. You know, I used to know so many different little things, but there was nothing that could save you that time in navigating and looking for the information. another very simple example of that would be, let’s say you knew something was in clause 27.3, you know.
Very few people actually realise it and they’re shocked to know it when they do. But if you do a control F in Word and you type in 27.3, if it’s auto numbering, Word won’t find it. It won’t find auto numbered text. So that’s no good to you.
Marlene Gebauer (14:06)
something people don’t
always know. mean, most people wouldn’t know that.
Feargus MacDaeid (14:10)
Yeah, exactly what I do. I do because I tried so many times to figure out a way to be able just to find so that if I did end up on page eight, I could just do control F twenty seven point three and I’d be back to where I was, you know, but it just didn’t exist. know, and I think that was I think Nnamdi could see that too. He could see all these different things and he was like, dude, there has to be something better, you know. And that then was like, you know, when that conversation started, you know,
Marlene Gebauer (14:11)
⁓ Right.
Greg Lambert (14:34)
you
Feargus MacDaeid (14:40)
through our kind of friendship and just being work colleagues, I think it started shaping itself into like, if you were looking at this 27.3 provision on page 137, what would ideal look like to you? And those elements that we both knew to be true. One thing was we knew that the word that I’m looking at
on that page, we knew it existed in the document with a meaning to it. So we knew the information was in the document. We knew that like, you know, it was always structured in a very particular way, you know, whether it’s like some form of stylized font followed by the meaning or whatever it was. But there was always a way it was done. And I think we’re to say, like, what if I could just find a way to locate that in the back end, let’s say under the hood?
and then just pull it forward so that I don’t have to go looking for it, you know, that it would actually just appear beside me, you know. And now I’m there. I don’t lose any context. can see the meaning right beside me. I can read it. I can interact with it without ever losing context. And I think I can’t remember. Nnamdi has a really nice way of saying it. I can never get it right because he always says it so well, but it’s about, you know,
the ability to kind of read and review and navigate your document without ever losing context. know, you’re always where you are.
Greg Lambert (16:14)
And Nnamdi I’m going to ask you some questions about the, efficiency gains here, but I think let’s back up a little bit. Can you, can you give us kind of a 30,000 foot overview of what it is that Definely does? I know it’s a lot of, you know, the, pre-execution stage of ⁓ a contract life cycle, but can you give us your, your elevator pitch?
Feargus MacDaeid (16:35)
Thanks.
Nnamdi Emelifeonwu (16:36)
Yeah. Yeah.
Sure, sure,
sure. So I mean, the concept of Definely is, you know, obviously evolved since the initial idea came ⁓ from Ferguson’s problems and the use cases he had. So we’ve now really morphed into and developed into.
⁓ a productivity suite and really the best way to describe what we do is to liken it to making a statue and if you allow me to make the analogy. when you’re making a statue the first thing you typically do is get your raw materials.
be it your granite, be it your iron, it your ice. And so what Definely vault does is emulate that in the sort of legal drafting process. So when you’re trying to create a contract, the first thing you typically do is look for the best templates to start with, the best provisions, the best clauses, and you typically go to your knowledge management.
you know, lawyers or maybe go to the partner and deal with your own associate. So what vault does, it allows you to actually link into your DMS to access those provisions, you know, be it the best clauses from a similar deal that was done two years ago, the best definitions or even the document that’s closest to the current, you know, transaction. So that will be sort of the starting position. Once you’ve then done that, you know, the next phase in that statute creation process is to actually refine it, carve it up, create this sort of shape of the statute.
you’re trying to do and that’s what Definely draft does and that was actually that first product that we released which Feargus alluded to. ⁓ allows you to essentially
review and access and navigate through a contract without actually having to ever leave the provision or the context of your review. So you can access your clauses, your provisions, your defined terms. You can also interact with that information ⁓ from where you are and not just from a single document, but if it’s, as is often the case, it cross refers to multiple suite of documents. You can access any information from any document.
related to that particular document without having to leave where you are. And then the final stage of that statute creating process.
is that you then, you know, wanna put the statute on display or you wanna sort of sell it onto somebody. have to polish it off, refine it, make sure it really looks really nice and presentable. And that’s what the Definely Proof Dots, it’s essentially a ⁓ proofreading solution that allows you to clean up and sanitize the document ahead of, you know, signing or sending it out ⁓ to the other side for review. So it does sort of your typical proofreading tasks, like making sure that there are no missing cross references, you know,
sort of punctuations have been fixed, know, grammatical terms are correct, you know, from a legal standpoint. So that’s how sort of the, you know, suite of products are brought together.
Greg Lambert (19:30)
Definely clearly delivers what?
you’re calling a significant time savings and I’ve read on the report that ⁓ it’s like 45 minutes a day, ⁓ that there’s a 90 % savings on some tedious tasks ⁓ that are going along. ⁓ I will say that, ⁓ for example, I was giving my Lexus people a hard time because they came out with the report that… ⁓
Feargus MacDaeid (19:45)
Okay.
Greg Lambert (19:59)
through a number of 344 % ROI on
using their product. And I was like, do you think anyone believes that? So what I was really wanting to know is, one, how do you go about kind of looking at the true measures of ROI? Because I think most law firms, at least right now, are really struggling to put
Marlene Gebauer (20:10)
Show me the money.
Greg Lambert (20:28)
a serious number on it that’s not just kind of throwing something at the wall and pretending that it’s right. you ⁓ know, can you… Yeah.
Marlene Gebauer (20:37)
It’s because it’s hard for them to capture. Like they have a hard time
doing that themselves in order to compare.
Greg Lambert (20:43)
Yeah, so how do you how do you kind of structure when you look at return return on ROI?
Nnamdi Emelifeonwu (20:50)
So there are various ways ⁓ we look at it. We’ve actually done a study and set up standards with some of our customers. So for example, we did a study with the lawyers who actually found, they tested this with their lawyers that it was saving them up to, I believe the time savings was 45 minutes a day when utilizing, defying me to utilizing sort of the typical.
processes that they use. Because if you think about what Definely does at its core, it is essentially the quickest way really to access information in the contract that allows you to read, understand, and interact with that information. So if you look at the typical workflow of a lawyer when you’re trying, as Feargus sort of alluded to, I’m on clause 10, and I’m trying to review perhaps clause 40 of the same document.
typically what I would do is either scroll down to clause 40, I might ⁓ try and use Control-F, which Feargus made the point that it actually doesn’t work on cross references, or I might actually print out a hard copy version and I flick through sort of the other version on screen. ⁓ Whereas Definely, can actually just now click on clause 40, it comes out right there in front of you on the sort of right-hand side panel, and you can read the information, but also not only clause 40, but any nested ⁓ information to allow you to actually get
the full context of your review of that particular provision. And you can interact with that information, right? So let’s say I’m at clause 10 and I open up clause 40 and then within clause 40, CROSS refers to schedule two. I can actually, again, access schedule two without ever having to leave clause 10. And if I wanted to make an update to schedule two, I can do it again from where I am without having to scroll all the way down to.
schedule two. Now if you can sort of theorize and imagine how the typical process ⁓ is when you’re doing all of those inefficient sort of workarounds that I mentioned, you can easily see how much quicker it is. And then the other things are, know, softer things like again that ⁓ ability to just retain context and gain your full understanding ⁓ of, you know, the particular ⁓ information that allows you to give the best, you know, review and hopefully the best advice to your clients. So
I think that there are various sort of ways we justify ⁓ ROI and there’s no better way than ⁓ your customers ⁓ speaking on your behalf. We have very, very strong ⁓ usage across our customer base ⁓ with our products. So I think that that’s how we do it.
Greg Lambert (23:31)
On some of these, what are referred to as soft benefits, and that’s the morale of the attorneys that are doing the work or enhanced value. Are you providing any type of framework for the customer so that they can track those more soft benefits of your product?
Nnamdi Emelifeonwu (23:53)
Yeah, so we have a ⁓ very, we work very closely with our customers. We have a very capable customer success team ⁓ who regularly have touch points with our customers just to make sure that they are seeing the usage and the ROI on.
on sort of the suite of products. we have various touch points. We regularly train them. We give them sort of unlimited training. And actually, our product suite is actually very easy to use. It doesn’t actually require ⁓ much training. But whenever there is ⁓ the need for it, we are very much ready to get on ⁓ the call or go on site to help them with that. Our support is best in class. We typically return. ⁓
back to our customers within a matter of hours, ⁓ you know, whenever they reach out to us. And, you know, we work with over 100 law firms and corporations around the globe. So you can only imagine sort of the effort needed to do that. But it’s something that we really prioritize as a business. And as a result, that’s the reason we have such a high retention rate and high scoring when it comes to, you know, NPS scores and so forth.
Greg Lambert (25:07)
Yeah, because
I imagine ROI, it’s easy to kind of look at numbers and then divide some kind of denominator into it. for some of these other things, and especially the things that will actually help you retain the customer because it has made the life ⁓ of the lawyer easier or better, ⁓ those are really kind of hard to quantify.
Feargus MacDaeid (25:32)
Yeah.
Nnamdi Emelifeonwu (25:33)
And actually there’s no
better testimonial or.
You know feedback on ROI then your customers speaking on your behalf. So just to give you an example ⁓ about just ⁓ Before you actually know about about six months ago one of our customers You know one of biggest law firms in the world They the partner in their ⁓ agent office was doing a deal was sharing his screen utilizing the finding to access and read the information really quickly His client which is a major investment bank and actually probably arguably one of the most prestigious ⁓ Investment banks in the world was on that call and asked him, you know, what’s that software you’re
using, he made a referral to us. ⁓ We got in contact with that bank. It took us probably about two months to, you know, from sort of initial contact, training them, doing a pilot, and then they’ve now rolled out Definely across the entire legal team. So there is no better, you know, example ⁓ or exemplify of, you know, the ROI you’re providing for your customers than things like that, which we appreciate greatly.
Marlene Gebauer (26:33)
So I’m
Nnamdi Emelifeonwu (26:33)
right.
Feargus MacDaeid (26:34)
you.
Marlene Gebauer (26:34)
⁓
I think this is really great. We’ve been really getting down into the weeds about like how Definely works and how it’s, ⁓ you know, satisfying it’s it’s customers. ⁓ I want to go a little, ⁓ a little higher level on this, this next question. you know, we’re seeing in the news, like, you know, major firms like, you know, you know, Sherman and clearly got leave and others, you know, they’re partnering, they’re incubating, they’re developing or, actually acquiring tech tools.
⁓ you know, even some large, you know, tech companies like Lexus and Thompson Reuters are investing in tech that, isn’t their own. ⁓ I have kind of, have two questions. So first, you know, how do you partner with firms navigating this shift and do you see collaboration or competition emerging between legal tech providers and the film, the film, the firm, firm built solutions. my gosh.
Greg Lambert (27:31)
Easy for you to say.
Marlene Gebauer (27:32)
It is not.
Nnamdi Emelifeonwu (27:32)
Yeah,
Marlene Gebauer (27:33)
Apparently.
Nnamdi Emelifeonwu (27:34)
so yeah, those are two really ⁓ good questions. I think from our perspective, I think the best ⁓ way to, ⁓ and I think the first question alluded to how do you work with the organizations or the firms, is that correct? Yeah, so it’s.
Marlene Gebauer (27:51)
Mm-hmm, like that are doing these things.
Nnamdi Emelifeonwu (27:53)
Yeah, so it’s really about, you know, solving real workflows, real, you know, real problems that at the end of the day, the end users, the lawyers have. So, you know, I think Definely started off as a product that was solving a real use case that Feargus had, and that sort of had much wider application for, you know, really anyone working on contracts. And what we do once we then sort of bring a
customer on board, we work sort of hand in glove with them to really develop ⁓ the product and make sure that the product is fits for purpose for their ⁓ users. We regularly customize ⁓ the product that they have so that actually no two customer actually has the same product depending on how the utility that they’re seeing from their users. So to give you an example, there’s a product we launched ⁓ a few months back that actually came from feedback from a super
user, one of our customers saying, hey, this is how we use this product and these are the workflows that we’re trying to solve utilizing your product. it didn’t take too much effort for us to actually implement a solution to address that. So we were able to do that and obviously ⁓ satisfying the customer. Whether I see or whether we see collaboration or competition, I think time will tell. ⁓ I think that one of the things we’ve learned and working with over 100
customers, it’s always really good to collaborate, ⁓ first and foremost, to really understand the pain points of the end user, because ultimately it’s the end user that matters. ⁓ I think that there are some firms who have shown a willingness to partner with ⁓ suppliers like ourselves, but there’s also other firms who have gone the other route and actually either acquired it or tried to develop it in-house. I think that there’s no one…
size fits all model. I think different firms are going to utilize whatever model makes the most sense for themselves. And I think us as vendors, it’s our responsibility that we always keep the customer front of mind and we work with them where there is the opportunity to do so, to just ultimately develop the best products for their lawyers. Feargus, I don’t know if you have anything to add.
Feargus MacDaeid (30:06)
Yeah,
yeah, yeah, like the collaboration competition piece I love, you know, because when you look at the history of how things have evolved since 22, let’s say with chat GPT, it’s actually really, really interesting, you know, because initially you had this kind of flurry of a lot of law firms and others, you know, kind of putting together, you know, kind of
solutions that they believe they could use and generate and produce outcomes with, you know, and I think a lot of them kind of did start that build it yourself kind of route, you know, I think collaboration is how it’s going to end up though. I think that’s really good for building like a proof case or prototype or whatever, just to see if something works. But I think what you’re starting to see now, and I think that’s also why you’re seeing some of the firms acquiring kind of some deep ML experience. It’s because
When you’re putting like, let’s say, a rag architecture together, you know, using an LLM vector database, et cetera, I think what a lot of people really underestimate is the amount of maintenance expertise in terms of ML ops, ML engineers, et cetera, that you need internally just to be able to deploy that model and to maintain it, you know? And I think that’s where a lot of the firms that they…
are beginning to kind of see the cost ratio against the benefit because they’re very expensive resource to hire in just to do that work. And I think that’s why I think ultimately collaboration is going to be the only way. think it’s going to be like, you know, firms are going to like, you know, they’re going to smart up with people like Definely, and they’re just going to go, look, you provide a really good solution. We have other alternative kind of use cases also that we’d like you to be able to help us to solve for.
and let’s just go for it, let’s just build it, you know, because we have the expertise, like we have the ML engineers, we have the ML ops, we have the backend infrastructure, and we know how to support and maintain it. So once we could share the use cases and the problems and the workflows with each other, and I think another part that makes, I think, collaboration in our instance, actually, a really high likelihood in what we do with a lot of our customers is when you look at our core team, let’s say,
Like three of our C-suite, we’re all practicing lawyers. Like we have about 20 years experience between the three of us in the city, you know? So we know what it is to sit behind your desk for 18 hours on a deal, five days a week, trying to get something done. Because we’ve all been there. We’ve experienced it ourselves. So when people are putting problems to us, we know what the problem is immediately when they say it. It’s like I’ve always said to Nnamdi and we always say it. It’s like…
you know, when somebody sees our tools for the first time or our tools being used, you know, like oftentimes the difference between a lawyer who has worked as a lawyer seeing it being used and somebody who’s, let’s say, in an innovation office but never actually practiced, like the difference in the reaction is like, it’s incredible, you know, because the second you double click on a definition,
and you show the meaning in context and you now start navigating around the document and interacting with it. Like the second you do that, lawyers immediately, you don’t nearly have to say anything else. They just immediately kind of go, they get it. They understand exactly what it is you’re solving for, you know? And likewise with pulling up precedent documents or pulling up clauses from the vault, you know, they know because they do it every day of their lives. Like,
Hey, have you got a clause from this kind of agreement? Let me try find it, et cetera. And to be able to search that in a data bank and pull it up in the context of what you’re drafting is super important. So yeah, think collaboration, I think, is the way forward. There’ll be elements of competition, of course, but you
Marlene Gebauer (34:07)
Yeah, I have a follow up on that. So are you seeing a trend and maybe you’re experiencing this yourself of, of, um, vendors collaborating, like, you know, you have, you, you have a product that does something really well. They have a product that does something really well. It’s all kind of in the same workflow. Um, are you seeing sort of collaboration that way or say, you know, integration with tool, other tools that firm has other than the DMS.
Feargus MacDaeid (34:39)
me personally, I’m not really seeing much vendor, vendor, you know, ⁓ collaboration. think though there could be elements where one, like I, there’s definitely some use cases, I think that Nnamdi and I have thought about before where actually getting an API from another firm, ⁓ could be as efficient and more beneficial than maybe doing something ourselves because what they’re solving for is in itself its own solution.
where it’s a well established technology. It would take like a large amount of resources to build it ourselves. So maybe an API call is the better opportunity, you know? But go on. No, no, go Namte, go.
Nnamdi Emelifeonwu (35:23)
One of the things that I
I think one of the things that’s actually pretty interesting and is one to watch given sort of the developments in this space is, you know, as more and more ⁓ firms and vendors think about, you know, agentic workflows and how they leverage agentic workflows into their existing product suites. And it’s one of things that we’ve been working on over the last few months. And one of the things that will be very interesting is that, you know, that really only works if you have the existing sort of underlying products to, you know, allow the agents to sort of call into.
if you’re a smaller vendor that needs to ⁓ reach out to maybe a larger vendor’s database, it will be very interesting to see whether that vendor will actually allow you access into it to be able to utilize it. So I think it’s definitely one to watch, and it’s a really interesting sort of development.
Greg Lambert (36:14)
Yeah. nominee back to, back to you on, on this one, at least, uh, well, actually we’ll probably get both of you in on it. Um, you know, you were talking about, uh, agents using, using AI, um, and, uh, you know, the, the legal tech adoption report, uh, that you came out with in, uh, recently talked about a significant AI trust gap where there’s like 75 % of
legal professionals are concerned about AI accuracy. And ⁓ I guess that ranges from the news cases where people are citing to hallucinated ⁓ case law ⁓ to just misinterpreting, I would think, within contracts or changing language midway.
Marlene Gebauer (36:57)
again. ⁓
Greg Lambert (37:08)
So, and then on top of that, I’ve read where Feargus has a strong statement that says, you know, 80 % accuracy just isn’t good enough in the law. So do you mind expanding on what you guys are doing with on the specific technical and operational safeguards there Definely, and kind of how you use that to enhance, but also keep the human in the loop on this.
Feargus MacDaeid (37:41)
I ⁓ kind of a geek head this way, you know, so I do actually love these kinds of questions. So what are we doing, you know, to mitigate the risk? think, well, the first thing I think is like, you know, having actually quite a deep understanding of how the technology itself works. know, knowing that the
Greg Lambert (37:41)
Ha
Marlene Gebauer (37:41)
Hahaha.
Feargus MacDaeid (38:08)
the hallucination, you know, Nnamdi and I were talking about hallucinations like in January 2023, you know, it was like, was already becoming obvious that this was one of the fundamental kind of weaknesses of like the LLM as a technology that could be used in law, you know? I think, you know, as soon as we saw like, you know, some of the first kind of research papers around rags started coming out in about like March 23, you know?
And as soon as you started seeing that as a solution, you kind of started realizing actually there’s a really strong way you can mitigate a lot of the hallucination risk if you use the data itself as context that the person is using. So I’ll give you an example. Like when you look at, let’s say the system that we’ve built, know, one of the examples would be when a person runs a query on the document, let’s say, let’s say they’re on
particular clause or whatever it is and they want to understand something or they want to understand a question about the document itself that they might have. When that vector query gets passed, when the retrieval is done, that retrieval is done only over the information that’s in the document itself. So what we’ve done is we’ve generated embeddings of the entire document, not just like, you know, the document as a whole, but we’ve chunked it up.
basically really, really well. And the way we chunk is quite unique because as Nnamdi said earlier, we have all these tools already built and they’re built prior to this. we had a really, really good way of segmenting the document by clause reference, definition type, all of these different things already built. So we were able to use that and leverage that to be able to create embeddings within the document or of the document itself.
we know that our retrieval accuracy is very, very high when we’re doing that retrieval part of the query. Then that itself gets passed to the language model, because you pass that as the context and you get it to bring back your answer for you. But the thing is, obviously you set your temperature to zero, like everybody knows that, but you can also then build in particular guardrails along the way so that, you know,
you can actually really, really mitigate any hallucination risk. ultimately, what it boils down to is like we’re not asking the language model for any of its general information at all. We want none of that. All we want is its linguistic ability to take the context we’ve given it and give us an output. You know, so I think that’s that’s a really interesting point. I think the trust gap, I think, is something that will narrow. think initially, I think, you know,
A lot of people had issues with it because it was actually quite wild, know, like some of the hallucinations that you’d see coming back in the way. But, you know, that’s been such a known problem over two years. And a lot of people like Anthropic and OpenAI and others have done a lot of work on how to actually reduce that and bring it down more more and more. And I think also people are just becoming more comfortable with it. You know, I think
Soon enough, people would just chat GPT or perplexity at the way they say Google it. Sometimes too comfortable. actually, you know what though, actually you bring up a really great, just that comment alone is actually a really great point because another part of my background is recently I’ve done a masters in cyber psychology. And you learn a lot about like Eliza syndrome or like the uncanny valley and
Greg Lambert (41:34)
Yes, sometimes too comfortable.
Marlene Gebauer (41:37)
Yeah.
Feargus MacDaeid (41:59)
all of these other different things that are happening in the mind cognitively without you necessarily being conscious. But I did learn a lot about like, you know, confirmation bias, trust bias, et cetera. And I know you mentioned something earlier in terms of like keeping a human in the loop, you know? And I think that is actually critical, you know? I think there’s a huge part of this where, you know, and there is like an agentic debate, you know, like in the world where people are kind of going,
or what is an agent, you know? I think like the CEO or whoever was from Google, I think at Google I.O. said yesterday, you it’s a combination of intelligent language models with access to tools to produce an output, you know? And I think that that is one version of an agent. There’s other people who believe that like, you know, agents should be kind of fully autonomous, you know, that
It’s only really an agent if it does it without any interaction or interference. I don’t really adhere, I don’t think we adhere to that school. think we very much believe in the human in the loop as kind of a validation layer that you need. And that goes back to the 80 % thing.
Greg Lambert (43:14)
Well, since you brought up ⁓ your master’s degree, me dive in there for just a second. ⁓ Because when we talk about the human in the loop, one of the things that I think we’re kind of doing is, yeah, we’ve got the human in the loop. We may have created a lazy human in the loop in that how do you picture
Feargus MacDaeid (43:23)
dear.
Awesome.
Greg Lambert (43:43)
the ability to come back with a good quality output from the product, but at the same time, still make sure that the client is not necessarily paying for, know, defiantly to answer their problem. They’re paying the attorney to answer that problem. so how do you ensure that the attorney is not
Feargus MacDaeid (44:07)
Mm-hmm.
Greg Lambert (44:12)
you know, being cut out of that other than just reading your output.
Feargus MacDaeid (44:18)
That’s a fascinating thing, isn’t it? Because you could see a situation, can’t you, where you create your own kind of catch-22. We provide you with these really great tools that can provide these really great answers, but in doing so, we de-skill your young lawyers because they don’t learn legal reasoning. And then what happened?
Marlene Gebauer (44:40)
And people get, they
get compliant, you know, they, just assume everything is right. And they just sort of, as Greg said, they get, they get lazy. Like we need, we need, ⁓ you know, skeptical auditors and how do we have that?
Feargus MacDaeid (44:52)
Yeah, the only way you have that actually, the only way you can be sceptic is if you actually have knowledge, you know? Like there is a reason why, you know, legal reasoning is called legal reasoning. There’s a reason why people go to school for three years and then do the bar or they do JD or whatever it is. There’s a reason why you do trainee contract and you build your experience. think.
There’s reason why people always say to you like, God, just stop thinking like a lawyer, because it is actually a way of thinking. It’s a very particular way of problem solving and thinking your way through in a very semantic, logical, linguistic way through the words and what they’re saying and what they’re pointing to and what they’re relating to. And if you’re not training people to do that, you’re going to definitely end up where basically how do you backfill
the senior associates once they become partners because you don’t necessarily trust the quality of knowledge of your junior attorneys. I think it’s a huge problem. think schools are starting to address that now. know like I think Georgetown has some program, I think I saw one as well, I think at Yale, where they’re kind of introducing some of this stuff into the actual fundamentals of the academics that they’re teaching people. But I think the way we do it, like…
So like I can I can speak for us at Definely, know, like I know Nnamdi like had his analogy earlier, you know, in terms of, you know, ⁓ the sculpture or the statue, you know, and I kind of use the analogy of Ford when it comes to looking at a lot of these things, because I think if you if you kind of reconsider what Ford did with the assembly line and manufacturing, you know, like ultimately, what did he do? You know, he saw
a vehicle and then he started to decompose all the elements of what put that vehicle together. And he created these kind of different stations along the assembly line and each station was a specialist at their task. And then they would do that task and pass it on and the next person would do it and the next person would do it. And in the end, you got like your car out at the end, you know. In a lot of ways, I think agents are kind of like the fortification of digital work.
I think what you’re going to end up with is a lot of agents that are very specialized at doing particular tasks, whether that’s a review task, a definition search, proofreading, finding information in your vault, et cetera, like what we’re building, because we already have those tools. And at the end of the day, though, even when you still got that car, you needed somebody to quality assure it. You needed somebody to test it to make sure that the thing wasn’t going to fall apart on you and crash.
And that person is the lawyer, you know, and they’re the person who’s still going to need to be able to test the robustness and the quality of their answers and what they’re getting. Because if they’re unable to, they’re going to look at an answer and assume correctness when something might not be correct. So unless they can be skeptical, unless you’re teaching your lawyers, you know, yeah, this gets you.
80, 90 % of the way, but you are that remaining layer of safety. You are the person ultimately that de-risks this company. This is an element of de-risking because it can find information quicker than you can, or it can point you or get things and do things quicker than you can on the documents themselves. But ultimately, the lawyer is still the de-risking factor. And that’s the way we’ve built everything. We’ve built everything like human validation.
is in the loop, you know, I think maybe the way to look at it would be a human being becomes one of those tools that the agent accesses in order to be able to accomplish a task. maybe in our case, the agent will kind of use the draft tool in the background or the proof tool or the vault tool or other tools that we’ve built. But along the way, as it’s presenting itself towards a solution,
It’s validating where it’s going and what it’s doing with the human. And the human is one of those tools that says, okay, it’s okay. You can continue, keep going. You’re in the right direction. You’re doing well, you know?
Marlene Gebauer (49:13)
Okay, Feargus and Nnamdi , we have come to the end of our interview, but we always have a crystal ball question for our guests. sort of in the next few years or few months, what do you see as big changes that are gonna be impacting the legal space, whether that’s technology or something else?
Nnamdi Emelifeonwu (49:37)
Yeah, ⁓ I mean, I think we’re going to see, and I think we already started seeing, I think a few years ago, everybody was very much excited about the promise of generative AI. ⁓ And people were very much ready to try it because they were very much excited by it. I think now people are being a little bit more informed and a little bit more.
just doing their research and really understanding ⁓ the workflows that these technologies can solve. think that that trend is going to continue. I think you’re gonna go from a very sort of approach where it was just get the technology in and then sort of see what we can do with that technology to actually, how does this technology actually help us solve predetermined sort of workflows that our end users have? Because that’s ultimately how you’re going to, and how firms are gonna sort of achieve ROI.
going to continue to see that trend where people are going utilizing the technology which is super powerful but actually you know go in with predetermined and predefined workflows that they’re using it to solve.
Feargus MacDaeid (50:43)
Nice. From my side, maybe I think on a technical kind of piece, I think it’s definitely going to be agentic, know, that’s obviously already there, you know, so it’s definitely going to be the layer that people use in software. In terms of how that’s done, I think you’re going to see a lot of MCP stuff like the
model context protocol stuff and servers coming along the line with tools and access to tools and interaction of tools. I think you’re probably also going to see a lot more come on small language models that are going to be very specialized at specific tasks and agents using those SLMs as opposed to just being heavily dependent on an LLM. ⁓ I think that’s because
A lot of people are going to start seeing inference cost. They’re going to see the cost, how much it is to run these things, you know, when they’re really in production. And people are going to want greater efficiency that way. I think ultimately the last thing I think where you’re going to see a real revolution, I think in the long run is going to be UX. I think UX and UI design. I think people are only really starting to come to terms now with how it is you design.
with an agent declare in software. It’s a completely different way in many ways of interacting with the technology that’s sitting in front of you. It’s no longer just a point and click with a mouse deal. There’s a lot more to it. And I think the UX is going to be highly important going forward.
Greg Lambert (52:23)
I totally agree with both of you on that. So, well, Feargus MacDaeid and ⁓ Nnamdi Emelifeonwu I want to thank you both for coming on the show and speaking with us today.
Nnamdi Emelifeonwu (52:33)
you go.
Marlene Gebauer (52:34)
Thank
Nnamdi Emelifeonwu (52:40)
Thank you so much. I really appreciate it.
Feargus MacDaeid (52:41)
it’s been our pleasure. Yeah,
thank you very much. Really enjoyed it.
Marlene Gebauer (52:46)
And of course, thanks to all of you, our listeners, for taking the time to listen to the Geek in Review podcast. If you enjoy the show, share it with a colleague. We’d love to hear from you, so reach out to us on LinkedIn and Blue Sky.
Greg Lambert (52:58)
And Ferguson and Nnamdi we’ll make sure that we put all the links on the show notes. But if the listeners want to learn more ⁓ about you or about Definely, where’s the best place for them to find you?
Nnamdi Emelifeonwu (53:13)
Sure, Feargus and I can both be found on LinkedIn under our name, so namdemahefun.org, Feargus and Mike did. Ultimately, also you can ⁓ go to the Definey website, so that’s just Definey.com, D-E-F-I-N-E-L-Y.com. You can find us there.
Marlene Gebauer (53:28)
Thank you very much. And as always, the music you hear is from Jerry David DeCicca So thank you, Jerry.
Greg Lambert (53:35)
Thanks, Jerry. Bye, everyone.
Marlene Gebauer (53:37)
Bye.