Transcription
Hello everyone. Um, thank you for joining us today. This is the LTH product briefing with Greg Mustard of Wexler and I'm excited to talk to you today because it's been a while since we've seen you on the product briefing. Welcome.
Thank you, Cheryl. Excited to take you through the updates both commercially and in terms of the products, of course.
Well, if you would introduce yourself to the folks watching in case they didn't see last year's and haven't had a chance to meet you yet.
For sure. So, I'm Greg. I'm the CEO and co-founder of Wexler. Um, bit of backstory about me. You've probably heard this one before, but I'm not an attorney. The reason I I do this is I have a personal connection. So, my dad was a litigator. He was a barrister in England and then a judge. He spent 42 years working on cases, um, usually divorces essentially. So, he did some big ticket money cases. Paul McCartney, Princess Diana, and a few others. And the thing that he he was sort of taught on his first day as a barrister was the power of having the facts at your fingertips. You know, if you crack the chronology, you crack the case. And he was struck that when sitting as a judge, the people appearing in front of him regularly didn't seem to know truly know the chronology of their cases. And so when sort of CH GBT came out, this kind of seismic moment in knowledge work if you like, I think we were we were thinking of of what we could do that would be a sort of more deterministic application of this rather than relying on a chat interface or the LLM can sort of go for this or go for that and and try to do every legal task. We thought what if we can break down these document sets into structured sort of discrete observable happenings if you like which we call facts and then use that to create this database from which you can create work product. So that's chronologies, that's inconsistency reports, that's drafts, but all powered by the core process of sort of extracting and enriching these events, these facts if you like from that data set. And that's what we call fact intelligence, which is the category that we're creating and the the category that we yeah, we're leading. So it's basically the exercise of establishing facts by extracting them, analyzing them and verifying them and then using them throughout the dispute, whether it's for early case assessment or for deposition prep or trial prep, hearing prep or any of the above.
So yeah.
Love that. And who are generally your customers? I've know you've been doing a lot of growing in the US and I've seen some big names on your website. So who are you working with these days? What types of folks?
So it's like big law plus boutiques plus sort of forward thinking mid-market firms I suppose. So you know most of our users come from big law firms purely because you know they're doing the bet the company disputes they're doing the you know really high stakes billion dollar plus arbitration or or litigation where you know quality is an absolute premium and you know we work with Clifford Chance HF Kramer Goodwin Shaw Godard Huntton various others and we're growing all the time in that segment. We're also working with some barristers chambers in England and also some kind of arbitration and litigation boutiques. So, you know, the smaller firms who may be taking on the larger firms and they don't have as many resources, but they need the system even more. But yeah, our bread and butter, our kind of ICP, if you like, is those big law firms who work on the bet the company disputes where you really need something that's going to give you an advantage in the kind of complex theater of litigation if you like and that that's jurisdictional agnostic. So yes, there are some nuances which we change in our product but broadly it's the same system. Obviously the way the law is applied is different but the facts are the same. Facts have no frontier and so yeah in the US it gets used a lot for deposition prep, trial prep, hearing prep, summary judgement briefing, motion drafting, all those kind of things.
Love it. It's such an interesting um area of growth and I'm excited that we've kind of finally gotten to this level. So what do you have to show us today, Greg? What are we going to be looking at?
So, we're going to take you through um basically the product, a sort of whistle stop tour. We're going to show you a sample matter, an opioid case. I'm going to show you how to very quickly ingest the documents, although I won't spend too long on that, but kind of how you give context to the system. And then I'm going to show you how to create a chronology, look for inconsistencies, and then use Kim, which is our sort of AI agent, you know, back and forth assistant where you can do iterative prompting, but it's all powered by that same fact bank, that same database, which you curate when you upload the documents.
Love it. All right, great. Well, let's jump in and I'll turn to you and then pepper you with questions as we kind of work through it.
Perfect. So when I would create a new matter and press create matter, we do have a plugin to relativity one and number a number of other platforms. They're not configured on this environment, but essentially it's either zip upload or it's pull in via relativity one. The document limit, which is an important differentiator is it's actually 250,000, but we put 200,000 product just in case. And you know that's vastly more than other platforms. And I think that's not just sort of uploading documents and tagging them and triaging them. That's doing in-depth LLM analysis over every single word mentioned on every single one of those documents. And in terms of file types, well, it's all the things you'll come across in your day-to-day. It's MSGS, PGs, PBTF, you know, TIFFs, all the rest. All the things that come up in eiscocovery. As long as it's got text in it, we'll we'll do it. And we'll also do images. For example, like if you take a picture of a sign, it will extract the text from that. Or if you take a picture of a sort of whiteboard, that will also be ingested as data.
So now I start to give it some mass details. So this is basically where I start to educate the system about the context of the case. And so this is not a prompt. You don't have to be a prompt engineer to use Wexler. It's not sort of something we want our lawyers to have to to become experts in because really they have enough on their plate already. We need to we assume that the system is intelligent enough to know a lot already, but we do need to know what matters in the case. So, this can be generated by uploading a complaint and it will fill out the information or pleadings or some other document, but otherwise you can actually draft a list of issues yourself. So, if you've spoken to your client and you know really all you care about is whether or not this ship delivered this sugar shipment on this day or something like that, you can put that in. It won't miss things out that aren't relevant to this, but they won't be tagged as relevant. So it's kind of the acid test we apply to each fact that gets extracted to determine its relevance to the case and also just to give it additional context that improves the output. Does that make sense?
That makes a ton of sense. So you're really kind of defining the universe at that point.
Yeah. But it won't miss things. So if you think you haven't drafted a good enough list of issues and it's not like oh god I I didn't know how to do that. It it doesn't that doesn't matter so much. It's just this is the kind of and you can update this. So let's say your client tells you something that actually they didn't tell you in the first call and you're like oh god I should have known about that. You can update that and it will then factor that into the relevance assessment.
Love it.
You might be thinking like when do you use Wexler? So you can either use it very beginning of a matter client self-produces a thousand documents and says you know get back to me by Friday. I've got three other law firms who are also going to get back to me by Friday. Who's going to give me the best you know understanding of the kind of quantum of the the battle ahead if you like? like how screwed up are we here? That could be that would be one very common use case. So that kind of initial case assessment where you need to get a really quick grip of the facts. And so what it's done here is it's pulled out for each of those issues that I supplied. So I basically added suspicious order monitoring is something I'm particularly interested in whether or not there were sort of suspicious ordering patterns in in the document set. and it's pulled out this incredibly detailed thematic summary sort of thematic overview of which everything is sourced back to the original section not just to the document but to the section as is kind of very common and t almost table stakes now but it can be done better or worse.
I actually had someone recently who's partially cited come back to us and said that our document viewer is super intuitive which was great to hear because he said most calls most sort of tech products he can't use because of the way the document viewers work but anyway I digress like you've got this kind of thematic overview view plus suggested follow-up questions all sourced back to to to each of those sections. And so let's say I don't really know about the sort of scope of the challenge ahead. I can quickly read this, look at the sources, and I'm already starting to get my head around the factual matrix.
And who do you see, Greg, who who lives in this in kind of this universe just in terms of I could see this being valuable to the the highest level relationship partner all the way down to, you know, a paralegal or a coordinator that's working on the case and supporting someone else. Who are you seeing working in your product?
Yeah, it's funny like some of our clients, we have senior partners who are using this as you say kind of the relationship partner to get a quick download of what's important in the case and use that to inform their strategy. I'd say the most common users of the associates, you know, the people who are time poor and and under tight deadlines and working extremely hard to get to build the kind of core of the case. You know, we've got, I think, 60 attorneys working on one matter at the moment, but also we have got paralegals, you know, we've got um eiscocovery support staff, we've got those kind of things who kind of work adjacent. They may not be actual fi owners but yeah I think it's like the benefits are are there across the board you know for partners about improving kind of recovery rate and efficiency there for associates you know helping you to find that information quicker so you can get home earlier or so you can build your you know spend other more strategic time elsewhere I think it's really important someone a partner recently told me that they said you know it's about the inconsistencies function which I'll show you shortly they said they probably would have found it but it might have taken three or four days to do that and they could spend the three or four days that wex Pixler found sooner that they could spend that working out how to use that inconsistency, you know, how to use it offensively and how to use it in their case strategy. I think that's exactly right. That's the right way to think about it. And I almost think about it as being smarter, right? Like your Iron Man, you put on the suit and you get smarter because you can get into the facts quicker, better, more deeply.
Yeah, exactly. I think that's me on. It's about freeing up your time to think more strategically and and allocating resources on where you know you can have the most impact.
Cool. So I'm going to show you the facts. So the facts are the the oil that power the rest of the product. They're the oil that you can do various things of fact with workflows for creating chronologies by filtering facts. You can batch out things and say those facts aren't relevant etc. But actually the fact is the kind of as I said previously the way that we distill complex information across a messy data set into something structured and ordered which has a date which has an event which has relevance which has entities assigned to it and it creates this really amazing queryable database which is what improves the quality of the output. So we call that the fact extraction pipeline. So this starts in the 1800s. It's not relevant to the outcome of the dispute unsurprisingly. It's background information. But what I can do is I can filter just to the relevant facts and then I can filter to everything related to specific entities. So I might filter to everything related to Floyd Ratliff here. And now I've got 128 facts where previously I had 28,000. And so I can create a chronology just based on that one person. Let's say it's a witness. And I can even view it visually which is nice. And so now I'm going into a deposition. I've got the key timeline of Floyd Ratliff, all the things, the facts that relate to him. I've got the ability to view it visually, and I can even export it into tabular or written format. And you'll also see that some facts are dduplicated across multiple different documents. That's to say there's not a onetoone relationship. Even if it's not written exactly the same, if it refers to the same event, we'll ded that which is really important because you know you might get 20 documents which repeat the same critical information and what you really care about is what happened, not what document they appear in.
You know, where is um where is the platform getting the things that are filtering? Is it extracting that? Is that part of your kind of analysis when they upload the documents and they're pulling that out of the the documents itself?
Yeah. So there's a few things. So relevance obviously comes from the user supplied information the list of issues. So we have four different categories for relevance relevant maybe relevant not relevant and key and then other things are inherent to the data set itself. So you know entities that's defined by the system document type key facts and so on and you can even do date ranges.
And that's all preset up automatically right so that's not I know in other systems I've seen where you have to go in and you have to add the entities and you have to that your system is doing that automatically for the user.
Yeah. Yeah. It's extracted. So I can go to the entities now and I've got all the entities that have been pulled out. So if I'm doing a cost of characters, you know, I want to get the key players in this litigation, I can pull out all of those things. So the FDA has been pulled out there. It's got aliases and then I can view related facts, relevant facts related to the FDA. Could be a really useful chronology. I'm going to build that now. So this is like the kind of more deterministic, some could say boring workflow, which is a bit less kind of jazzy, but it's nonetheless the oil that powers the rest of the product. I'm now going to show you the the kind of AI agent. If you like the AI assist, I mean agent is an overused word, but I think like it's agentic in that it looks at different data sets. It will kind of correct itself and it will do iterative tasks to produce this output. It's not going to go and send the email to your client, but I think most people want that with the current state of AI.
Yeah. So, you know, what's the strongest docu documentary evidence that Malenro had knowledge of diversion? And so it's pulling out this really detailed output with all these different sources which are all sourced back to the underlying fact not just to the document as you see there. And so I've got here 33 different sources. I can then view the references in the sources and then bounce them into a cron. And so where where this is really cool is I can use the like AI. So I've got like all of the facts mentioned in this query which is about the strongest documentary evidence of Malen Crot. So now I can then take that and I've now got a chronology of all the documentary evidence related to Malenrot's knowledge of diversion. But then I can even go back to the facts table and search diversion. And in this way you build up like a really comprehensive understanding of the facts. I've got the AI's version. I've got I've gone back to the facts table and I've gone through every single one. And now I really understand the case.
That's amazing. I think how long this used to take to do.
Yeah, for sure. I think it's.
Quite a hack.
Yes. And then I'll give you another example. So, did any internal communications contradict Malen Crot's position that it lacked knowledge of downstream diversion? Yada yada yada giving you a really good detailed answer. But then I've said turn this into a deposition outline with questions. And so now I've got it's created this sort of work product piece which you can then go through and it's given you questions sourced back to each of the facts based on that initial query and then I can use those to create a chronology as well. So you use the different parts of the product in complimentary fashion essentially to sort of build up your holistic understanding. I've got all this work product which I can use I can share and collaborate and so on. So that's facts that's Kim. Any questions on that?
No, I mean the Kim is very impressive. I love it.
It's not just like asking a sort of chatbt style model to one shot like chat to some documents and here's the answer. It's querying using a really complex graph rag system this factual sort of database that we've built up and all of the context you've supplied as well. And so we do hear that the answers are very detailed, very accurate and really grounded in the sources rather than sort of arbitrarily coming down on one side of the argument. you'll know about context like running out of context windows. Those kind of things don't occur in Wexler. Obviously, you cannot can't say it's 100% accurate and anyone any vendor saying it's 100% accurate. You definitely need to to double check the the T's and C's there. But um but I think it is highly accurate based on this grounding in that that sort of curated fact bank if you like.
Absolutely.
Before we wrap up, a couple more features. So subsets really powerful. So you can actually use the relativity integration to batch out documents into subsets or you can do it through the the sort of Wexler native characteristics if you like. So use case for this is like I just want to look at the other side's data. I just want to look at my side you know our data or I just want to look at everything related to Cheryl or everything related to Greg or everything in a certain time period or maybe everything related to a certain issue code. And so I can create a subset using all of these different characteristics and now the analysis is super focused just on this data and it's not going to go out exterior to that. So what I could do is I could have people asking at this matter level. Let's say there's 100,000 docs at the matter level and then I've got 20 subsets with a thousand documents in each and I can build really detailed granular analysis and then I can get a more sort of global analysis at the matter level. much more um sophisticated approach to issue coding than just taking a document and putting a label on it.
Huh.
Yeah, for sure. And I think this is really useful and we get a lot of good feedback for this because some you know people want to if you're doing a decadel long class action or let's say you're just doing a class action is 100 plaintiffs you want to know the chronology of each of those users and you also want to know be able to ask sorry of each of those plaintiffs but you also want to be able to ask about patterns. You know if there's a fraud claim for example and you know loads of people are claiming the same thing you want to know what each of them is claiming. You also want to see you know what are the common topics that are coming up. whether those kind of things are very common.
Finally, two more things. We've got inconsistencies, which is a really powerful feature. We had feedback that, you know, it picked out something that forensic accountants had missed, which was used throughout the rest of the matter as a kind of key, you know, I don't want to say smoking gun, but let's say very hot gun. Um, which was like which was like very very important throughout the rest of the matter. And so this is looking for discrepancies, you know, contradictions. So for example, if someone says something in their deposition and the underlying data suggests something else, that will pick that will be picked up. And so really useful for impeaching testimony. You can upload data at the end of the end of the day. You can upload the transcript and it will add that to the data set and then you can even hero certain documents like certain depositions. So that's really useful.
Can you choose what the what you want to compare to? Because like you said, you may have a it's an expert. He's been deposed 50 times. So you have many and then you have the one in this case, right? Like can you choose one to one or filter by someone? I'm just curious.
Yeah. So you can anchor to specific documents. So if someone's got like an expert report and you can also those one didn't have any existence but you can also filter by other things. So the level of inconsistency and the source and then but what you could actually do is ask Kim to pres prepare you inconsistencies you know and then it will be it'll be retrieving from the same data set. So again, it's the same thing. This is like the full raw, not raw, but like this is the more sort of baseline data we use and then Kim will reason over this information and pull out what it thinks is most relevant based on your query. So yeah, we've also got the ability to do it real time. I'll show you an example. So we can do real time analysis. So this is obviously very interesting and it gets used a lot for uh like deposition preparation, let's say scenario planning. So I want to put my witness on the stand and say, you know, let's run through some scenarios here. What are you going to say? What are you not going to say? And so on and so forth. And so I've got Richard Sackler's deposition here and it's pulled out inconsistencies based on the state of Kentucky and about whether they were marketing Oxycontton before they knew of its ill effects. So this is only possible because of the fact bank that we've created that it's able to compare it against but it will transcribe and fact check in real time against the same record. So this is really useful. I would say it's like a bonus feature but it still gets used regularly but it's it's you know the core intelligence is powered by the underlying fact bank that we've built up.
Well and I suspect for based on how you're building it and kind of structuring the data and massaging things there are almost endless kind of edge case things that you'll be able to rail down as customers ask for it because you'll have that data in a way that's very accessible and your team can use it. So, seems like you have a lot of opportunities coming forward.
Yeah, for sure. I think we're increasing scale. So, we're going to get to millions of documents by the end of the year, which I think from an AI I've not heard of any other AI native platform being able to ingest and process that many documents. And there's loads more there's loads more things that we're developing at the moment. So, more integrations. I think you know we're doing an email add in so you can kind of keep the context from your emails and the kind of the pingpong you get back and forth in litigation plus other things like procedural timelines so you can know you can upload the timeline you know when things need to be filed and it will suggest to you proactively what you need to be across you know let's say 7 days beforehand or 21 days beforehand you know after the complaint has been filed and give you suggestions of things you need to look for just increasing that context I think is so important because that's really what people want. They want to know before they even thought of it what they have to do. So, yeah.
Yeah. I mean, that's amazing. And Greg, you know, one of the things that I used to always think about when I was buying was like, where does this sit in my workflow or my lawyer's workflow today, right? And so, I think, as you said, this is something investigations, early case assessment. You can use those very, very, very early in the cycle before it's even a litigation matter.
And I can see use all the way through trial. So do folks step out and go to a relativity at some point and then bring it back or how does that like what do you see the workflow being for your users?
So we don't do first level review. So basically we don't do first level review. We don't do legal research. So everything else in litigation we we cover. So early case assessment, investigative work, second level review, depo prep, trial prep, summary, judgement, briefing, you know, everything. Ad hoc drafting queries, arbitration is obviously very useful as well. and across jurisdictions. So that's a significant portion of the day-to-day work of a litigator. But yeah, we don't do the core like first level review issue coding, but we bring in a lot of that intelligence into Exerta and we don't do the law, but we, you know, we leave the law to the lawyers. They apply the law. They apply the law to the facts that we help them to to develop. Um, so yeah.
I love it. Yeah, I can see I can easily see lawyers working in this non-stop.
Um.
Yeah.
The other thing I was thinking about is, you know, this is a really interesting tool even for firms where they have a lot of experts testifying. So like an FTI or decision quest or something like that. Folks that, you know, might be the types of people who could get caught up in a intelligence battle of facts, right? They've testified 50 times and they're getting ready to go to their 51st deposition. So it sounds like there's lots of really interesting applications for the product.
Um.
100%. Yeah, a lot. We do we we work with a few expert firms I think.
Absolutely.
For that.
Yeah, 100%.
I think look I mean if you someone told me said to me and I agree litigators tell stories tell stories backed by facts hopefully and so yeah you know we help get that bring that information in a vastly shorter amount of time and help them be the best lawyers they can be backed by you know the power of the documentary evidence that supports their arguments and their hypothesis. So yeah.
Love it. Well, as we close things out, I want to give you the opportunity, of course, to kind of share anything around security, privacy that's important to share for folks. I know we've talked about that previously, but we'll throw that out. And then we'd love to kind of um give you a moment to promote anything or send people to your website or whatever you want to whatever is whatever's hot right now.
So, security, you know, we we don't take this lightly. We're SOCK 2 type 2 approved. We have ISO271. We do regular penetration testing, abuse monitoring exemptions which you know is usually the preserver of much larger companies. Your zero data retention hosting in the US, UK, Australia or we can do private cloud. We can also do BY if necessary. So yeah, that's pretty much everything I can think of. I would have expect to to tackle any surveys you can throw at us. And then in terms of promotion, so we just released that new version of Kim which I showed you with the kind of chat iterative prompting the artifact or sort of work product creation but all powered by the same facts rolling out across the US. We've just been at legal week last week. Yeah, I think if you have heard of us then you know get in touch if you haven't. We'd love to set up a demo. You can easily sign up for a demo on our website or via you know legal tech hub. So yeah, I think you know we don't do every legal task. We do not try to boil the ocean, but what we do do, we do with great depth and we do it really well. And when you're in a when you're in a very adversarial litigation, like I was just speaking to someone yesterday where they say people are playing dirty these days. When you're in that, you need to be backed with all the ammo you can get and Wexler can give you that ammo that could really help you gain the upper hand. So yeah.
That is a beautiful way to end this. Thanks all for joining us. It was great to have you here, Greg, and looking forward to seeing what you do next. So thank you again.
Thanks very much.