Podcast Ep #69: Document Automation Secrets for Small Law Firms with Quinten Steenhuis

I think we've probably all had that infomercial moment at some point where we're drafting the same type of document for the umpteenth time, and we just feel like throwing our hands in the air going, there's got to be a better way. But if you've ever tried to turn one of your documents or templates into something repeatable, ideally some setup where you just answer a series of straightforward questions and the technology spits out a polished draft, you probably already know it is harder than it looks.

This week, I'm joined by Professor Quinten Steenhuis, who is one of the most experienced document automation professionals I know. He's worked with legal aid organizations, court systems, and private law firms to turn the production of everyday legal documents into productized legal services. And he brings a ton of practical insight into what works, what doesn't, and how to think about document automation in a way that will improve your practice and delight your clients.

You're listening to the Agile Attorney Podcast, powered by Agile Attorney Consulting. I'm John Grant, and I help legal professionals of all kinds build practices that are profitable, sustainable, and scalable for themselves and the communities they serve. Ready to become a more Agile Attorney? Let's go.

John: Alright, everyone. I am excited to have on the podcast this week, Quinten Steenhuis, who I'm going to let introduce himself because he wears so many different hats that I can't keep track of them all.

Quinten: Well, thanks, John. So, yeah. So, I have two hats I'll talk about today. One, one is that I'm the co-director of the Legal Innovation and Technology Lab at Suffolk Law School in Boston. And the other is I'm the owner of Lemma Consulting, which is a consulting firm. We work with legal aids as well as law firms around the world, helping almost entirely with document automation solutions, some of which have AI integrated, some of which use the front end of document automation and are mostly about the AI integration and not the document part.

John: Yeah. And so, we first connected, I think you were here in Portland right before the world shut down in the pandemic because it was the Technology Innovation Conference for the Legal Services Corporation. And it was kind of like the last big thing any of us did, I think for the most part, before we all had to go into to isolation.

But then, not long after the pandemic, you and your team at Suffolk started the Document Assembly Line project. And I would love to start with just having you talk about that project, including like what was the impetus for getting it started and how it evolved. And then we'll talk briefly about the small role I played in it, but I mostly am curious to know how it evolved and what you've learned and what maybe individual practitioners can learn from your experiences.

Quinten: So, that moment that you were talking about when the world shut down was the impetus for the Document Assembly Line. So, in Massachusetts, as around the world, courts were closed for people to go in and get help. But there was very little digital assistance that was available to kind of step in and fill the gap for that virtual assistance. So, what we did is we looked around and we, I had just started at Suffolk. When I saw you at the Innovations and Technology Conference, I was still working at Greater Boston Legal Services. My first day on the job at Suffolk was going to be early in March, and the campus was shut down. So I didn't see my physical office for more than a year, actually.

We decided that there were lots of things that people needed to help with, but there was one particular thing that we could help with really well in the Legal Innovation and Technology Lab, and that was helping come up with some digital solutions to let people solve the emergency legal problems, even, just the ones that the court was still open for hearing. They could get some way to get their documents filled in correctly. They would have gotten maybe help from a lawyer with before. And then actually get those to the court. So we focused on those two pieces, digital delivery and accurate completion for about like the 30 or so kind of emergency things that we saw were happening that people still needed help with.

John: And if I remember right, and correct me, but Massachusetts did not have e-filing in a lot of courts at the time or kind of rudimentary e-filing?

Quinten: That's right. And there still really isn't e-filing for every process and every court in Massachusetts. It's kind of case by case whether that case type lets you have a digital filing or not. So, that was another piece that we were we were dealing with, was that lack of digital access to the filing systems.

John: And so, in terms of what these documents are, right? They're court filings for the most part. And they're structured, they have certain rules about what the court expects to see in them. Often those rules are not what everyday people think of in terms of like how they want to write or create or convey information. And so, the assembly line team was basically tasked with figuring out ways to get information from people and then translate and I guess adapt that information into a document format that the court would recognize as a valid and effective legal filing.

Quinten: Yeah, that's a really good summary. And I'm going to have to borrow that phrasing in the future. But yeah. So, we think of it as, one phrase I've liked to use a lot in the past is we were trying to build a legal clinic in a box. You know, so imagine the best hands-on legal support you get, limited in a clinic setting. Someone's going to walk you through that form, make sure you've done everything correctly. You know when to check the box and maybe they're going to look it over and give you a little bit of information about what to do next. So we tried to package that up as a computer app that's accessible on a smartphone or on a regular desktop computer.

John: Yeah. Okay. So then, like through my weird lens of agile and lean and, you know, productizing legal services, part of what you're doing here is productizing a legal service.

Quinten: Absolutely.

John: You're taking something that was traditionally sort of a labor intensive, communication intensive process and trying to reduce it to something simpler that could be delivered using technology as opposed to quite as much human to human contact.

Quinten: Taking what might be a really intimidating court form. You know, imagine you have, for example, one of the first things we started with was the domestic violence restraining order. It's seven pages, 180 separate bits of information it needs you to provide. And that's for just the most basic one. Let's say you don't have children and you maybe don't need some of the other things like child support and maybe protection for a pet. Those are additional forms that are not counted there. Imagine just getting faced with that and having to fill it in on your own.

We heard a story of a woman who went to the courthouse in Boston early in the pandemic, and that was literally the situation she was faced with. So she went to the courthouse, no one really quite knew what was going to happen or the rules weren't necessarily that clearly communicated yet at that point. She was waiting on the courthouse steps for several hours, literally, until someone came out, a clerk, and handed her that stack of paper to fill in on her own.

So, what would she have gotten if she'd been able to talk, as she would have just a few weeks earlier, with an advocate in person? They would be sitting her down, letting her kind of tell her story and reflect actively back to her, hey, thanks for sharing that with me. Here's the thing I want to follow up on. How can I get more information about that? And helping her fill in that form. So it's a much less daunting process, something that's more bite-sized with each interaction. That's what computer systems allow you to do. They let you slow someone down to fill it in accurately. So it's less overwhelming, less traumatic potentially, for if we're dealing with a situation like domestic violence, and ultimately produce something that's just as high quality as you would have gotten with a lawyer doing it for that person.

John: Well, and so a few things that I feel compelled to point out for listeners, one is that you were very much leading with the needs of the client, right? The needs of the end user, not necessarily the needs of the lawyer. Although I think the needs of the court probably have to come into play as well in terms of getting the magic spells written in the right format in order to be acceptable. But the other is that you're, you know, again, you're using technology. And specifically, we're using document automation technology. So just to put a name on it. And if I recall correctly, the tool that this is all built on is Docassemble. Is that right? Or is it still called Docassemble? I sometimes lose track of the names of these things.

Quinten: Yeah, that's right. So we used an open source technology called Docassemble. It's built by a legal aid attorney, Jonathan Pyle, who's in Philadelphia. He uses it in his day job and he like gave this wonderful gift to the world. Two different startups were built on it that probably make millions of dollars a year. People have solved countless legal problems with it, and it's free for anyone to pick up and use and to change or modify or even build something commercial on top of. So it was that was a really amazing part of our story.

John: And for lawyers that maybe haven't dipped their toe in document assembly at all, what are the components of that process of actually getting information out of the user and then translating it into that sort of acceptable PDF or other document format that you can file?

Quinten: I think of it as involving three different pieces. So there's your document that you want to have as the output. We turn that into a template. So that we have blank spaces where the information from the user gets filled in. And if you think of a court form, it often already starts out looking that way, right? There are lines where the information is supposed to go. Then there is the user interaction, the interview script, if you will, where you kind of lay out how you want to ask each question of the user in a way that's appropriately tailored to what they need, giving them just the right amount of information that it's easy enough to use for them to answer accurately and correctly.

And then a third piece of it would be the logic that helps fill in all that information into the form correctly. Sometimes it seems very simple. There's just a checkbox, but that checkbox says that you have the right, this legal right that involves this very complicated statutory information. So we need to think about really carefully about that logic to make sure we're handling that part of it correctly too.

John: Right. So, different from a fillable PDF, right? Although fillable PDF is maybe a very rudimentary version of document assembly. In the interview itself, it's not necessarily just a barrage of questions that all show up in a wall of gray on the screen. You sort of have the ability to structure these questions in a more natural sounding way and a more adaptive way because part of what the logic does, right, is to maybe ask the correct next question based on an answer to a prior question.

Quinten: Absolutely. So if you could think of that restraining order example again, there are 180 bits of information you could provide on those forms. They're not all relevant to all people. So we're able to right away say, okay, we're going to skip this follow-up branch of questions because it doesn't apply to you. You can really streamline the experience that way by cutting out irrelevant questions. We could think through how to group them in a way that makes more sense.

Court forms are designed for the court. They're for clerks and judges to look at and see if they can understand the case at a glance. They're not designed for litigants. With a guided interview, we're focusing purely on the litigant experience and what they need out of that process. We don't have to ask questions in an order or in a grouping that's really just fit for the court's purpose.

John: Yeah. Okay. And so then, I'd love to hear a little bit more, and I think this is where I had a small role, but in terms of you, you know, as you said, there were hundreds of court forms, I think eventually that you have that have gone through the document assembly line process. And, you know, as I recall, the assembly line, in the early going, it was sort of an intention and a clever name, but it you still had to figure out how to actually build the assembly line itself. And so, you know, and it's been a few years now, but I'd love to hear from you kind of what your experience was.

Again, because I think it's helpful for private practice attorneys as they're going through the steps of building their own tools and systems, and whether that's all the way into productized legal services that you're pushing forward to the client or not, there still is a lot of benefit in creating repeatable systems and processes that can work inside of your own practice, much better than say, grabbing the last piece of work product you did that was remotely relevant and deleting all of the case specific information and starting over, right? Which I think is still a fairly common practice in a lot of law firms. So, I made that a long question, but.

Quinten: Yeah. I don't want to totally like dismiss that idea, right? Maybe sometimes not everything needs to have a document automation solution built for it. I do a contract like maybe once a month with my consulting business. I don't have a document automation solution for that because the process, it's unique enough each time. And I don't do it enough times to justify the investment.

John: I don't either. And it's and every time I do it, I ask myself the question is should I actually automate this? And so far, almost 10 years in or over 10 years in, I haven’t used the same document for 10 years, but, you know, I'm at least four or five years into this version of my engagement stuff and I still haven't automated it yet because it just it's not necessary yet. But let's say it is. Let's say that I it's something that I'm doing a few times a week as opposed to a couple times a month. How should I think about it?

Quinten: Yeah. And I would say there's two kind of things to consider. One is like how often do I do this? There is maybe how complex is it? Right? Something you can get with a document automation tool is kind of a self-enforcing checklist where each step of that process works exactly the way that you've designed it to work each time. If you try to do that by doing search and replace, maybe you forget to update the pronouns in your estate plan, one of the times or in one of the places. Or you do a search and replace and someone's name is contained inside another name, and so you have this weird, awkward misspelling halfway through your document.

If the stakes are high enough or the document's complex enough, document automation might be like the only responsible way to do that when you're working in a really document and form heavy kind of area of law. If you can automate it, like maybe you should, just so it comes out right every time. If it's a simple document, you can read the whole thing over, maybe you don't need it for that reason. But maybe you need it because you're doing it several times a week and you just want you're getting back that, you know, half hour each time you use it, it starts to add up.

So, with the Assembly Line, I mean, the reason why we have that name is we had to figure out, we have a bunch of things people need right now. They need access to these legal processes. They don't have any way to do it right now. How do we decide what's the most important thing to start with? And how do we divide up the work? You know, right? Like pushing back against that mythical man month. Can we make it so we get the benefit of having hundreds of people's help to divide this up so we can help solve each of these processes at a time or as quickly as is realistic for us to launch them.

So we had more than 200 volunteers over the course of the Document Assembly Line project that each contributed very varying amounts of time. Sometimes it was just a few hours. Sometimes it was foundational work like yours, John, helping us set up some of the systems to divide the work in Kanban boards. So it was really easy for us to piece it out and apply project management strategies to figure out who's doing what.

And that was a really big part of it, that process of what's the right size work someone can do that's helpful for to move this project a little bit forward with the amount of time that person has. Some people donated months of full-time work and some people just a couple hours. And so that was a really interesting part of it as well.

John: Yeah. Well, and I think what I remember about the project planning setup and the Kanban system that we set up was that a big piece of it was being able to signal to volunteers which items were sort of ripe and ready for their attention. And then once someone gave it attention, make sure that it was claimed by them and that, you know, you didn't wind up with three cooks in the kitchen working on the same thing. So getting that sequencing right.

But then also making sure that work was actually moving all the way through to the completion process, and that we weren't just starting lots and lots of new documents and having them sort of get stuck in the middle, the messy middle of, you know, whatever it happens to be, right? Getting that logic coded correctly or making sure that it actually is producing the right PDF image that the courts are going to recognize, et cetera, et cetera, right? So we worked a little bit on the Kanban notion of stop starting and start finishing and trying to get things all the way to done as opposed to just getting new things started. Which I think ultimately seemed to have worked reasonably well.

Quinten: I will say that's a challenge for any organization to do. With document automation tools, something that can happen is that there's really no place you have to stop it until you want some people to actually use it. You can look at it and say, I could improve this. I could make this better. I could make this simpler for the person. I could do this little bit of extra step where it's done for them and they don't have to do it themselves. That was really hard for us to get. Like, how do we get this project over the finish line and say it's good enough? We got better at that over time. But definitely having those systems where you could kind of visualize and see the amount of work that was in different stages was helpful for us to be able to do that.

John: Well, and so then where is the project today? Does it still exist?

Quinten: It does. Yeah. I'm really excited what kind of evolved with it. One of the key differences I would say is that we no longer have 200 people who are, you know, kind of at home looking for a meaningful way to engage with the outside world. So we don't have that massively distributed set of work anymore. But what it has evolved into is a support system for legal aid programs and courts around the country, primarily. There are people using our tools in other parts of the world as well. We focused really on, okay, we're building a interview for a court form. How much work can we save somebody when they want to automate the next court form? What are the parts of that process that are exactly the same every time that we can kind of set up for somebody so there's a consistent look and feel, so they're not reinventing the wheel with the next project, so they can redo it.

So we pivoted to spending a lot more time thinking about the framework. And that's a lot of where our time goes at the lab. And we help offer support to other legal aid programs and courts around the world to make use of that framework and weekly meetings now with about eight different legal aid programs around the country where we help offer them support on their own document automation projects and help them get their tools live. We take what we learn from that back and add it to the core. They get to share their ideas and offer peer support too. It's a beautiful thing to see when you can just shut up on a call and people help each other solve their problems.

John: Right.

Quinten: And that's kind of where we've focused.

John: Well, and that in some ways is the natural result of developing an effective process, right? And a process built around producing work to a certain quality of output. It actually, it brings to mind a question I meant to ask you a few minutes ago that I didn't, or maybe a statement I meant to make a few minutes ago, which is, you know, one of the benefits of standardizing work, whether it's through the documents themselves, right? Because part of what you're doing when you create these interviews is you're saying, okay, we're going to do this in a repeatable way. And yes, there's time savings that comes from that, but that's often not the primary benefit.

The primary benefit is the ability to produce a work product of consistent quality in a very repeatable and very predictable way. And you can do that through the technology innovation, right? Which the Docassemble tool allows you to achieve because it's built by smart people that were keeping those things in mind. You can also do that with process innovation by having quality standards and good communication practices and good tools and training and onboarding and documentation, et cetera, et cetera, that allow humans, even if they're not using a computer system or depending on the extent to which they're using the computer system to do the work, to achieve a predictable quality outcome.

And then once you have multiple users familiar with that same standard of quality, they can work with each other because now they're speaking the same language, they're referencing the same ideas, the same standards. And you really do get some amazing, not just economies of scale, but I think sort of powers of innovation. When you have people speaking that same language, then you can discuss problems more effectively and come up with solutions in a more interesting way.

Quinten: That's definitely been a big focus of ours. You mentioned documentation. So, we build out a lot of documentation covering things like how to use our tools and how to use our systems, but also how do you ask this exact kind of question the right way? You know, the usability testing that we've done that I've done through my consulting work and kind of contributed back, that has inspired a lot of really great information to help other people replicate, not just the technical part as you said, but also the user experience and the other what you might call, I guess, soft aspects of the project, which are really critical for it to be.

John: I was going to say, yeah, it's funny because I do you do hear them talk about as soft skills or soft aspects and at the end of the day, I think the soft skills are 90% of the game, you know. You can have a technically effective output with a terrible human experience that is not going to generate a good result because if the human experience isn't there, then it doesn't matter how technically proficient the whole thing is.

Quinten: Absolutely. I mean, so one of the things that I think about with that is we're all lawyers. A lot of the people on our team are lawyers too, who are working on this. And it's so hard for lawyers to let go of their jargon, which feels like very specific and it's like we have to be exactly legally accurate. So let's use that word.

John: That's why I referred to them as magic spells earlier, because it is sort of like that sometimes.

Quinten: So we try to think of push on the plain language aspect of things a lot. And a mentor of mine, Caroline Robinson, who's now at the Alaska Court System, she was at Massachusetts and a key part of the assembly line project for a long time, still involved. Now we're doing that work in Alaska.

She's like, sometimes the reason why lawyers don't want to drop that legal jargon from their writing is because they don't know how to explain the concept using regular words. And actually, it turns out maybe they don't fully understand that concept that they're hiding behind the jargon for. That's just filling in the place for, like, here there be dragons. We don't know exactly what that is, but it's got to be this exact legal term.

John: Yeah, well, and it also, I think there's a function of adherence to precedent that comes into play, which is when a particular magic spell has been effective, we perceive risk in deviating from the exact language of that spell. I'm probably overusing that metaphor now, but the specific legal language, because lawyers are trained to be risk averse, we think, well, gosh, if we make that into plain language, do we actually risk losing the effectiveness of it? I think the answer sometimes is yes, but it's probably yes a lot more infrequently than most lawyers fear it's going to be.

Quinten: I think you're totally right. I we know this is right and it can convey the thing we want to say. And they're more focused on that risk of changing it than maybe than the benefit that could come with it. And it's not invalid. I mean, sometimes writing the very long explanation that covers every scenario is going to help somebody. It's just at the cost of burdening everybody else maybe with something that's difficult to understand and kind of slows them down and maybe makes them even give up from finishing the process if they're not sure what to say at a certain point.

John: Yeah, yeah. Well, and I can't remember if it's a Mark Twain quote or whoever, but “I'm sorry for writing you such a long letter. I didn't have time to write a short one.”

Quinten: Yes.

John: That comes into play as well. But maybe this is a good time to maybe transition a little bit too, because I think one of the key differences between the work that the Document Assembly Line project was doing back in 2020 at the start of the pandemic and today is of course the rise of generative AI and the increasingly widespread acceptance of it in the legal profession.

And I had a conversation with Damien Riehl a few weeks ago where we agreed we're not going to talk about all the pitfalls of AI because they get beaten to death. But it is interesting. I mean, one of the things that it does, number one, in some ways generative AI is only adhering to precedent because it's trained only on precedent, right? It has to look at existing examples. But it is recombining words and, you know, phrases and even ideas in ways that are increasingly interesting to me and, you know, sometimes surprising in ways good and bad, right? It's not an unvarnished problem either way.

But one of the things that I know that generative AI can be really good for is what I'll call translating, but translating legalese into plain language, translating legalese into, again, 8th grade reading level, 5th grade reading level, maybe not getting 100% of the way there, but at least giving you a really good starting point to think about language. Is that something that you and the team are doing at the lab or on the project?

Quinten: Absolutely. I think that exact use case is a is a great one to help. Helps I actually do an exercise with my students in my seminar and in our clinic where people build these tools as well, where they kind of compare their own approach to doing the plain language to that of a tool like ChatGPT. And then they come back and make an assessment over which how it works and how they're able to approach and tweak it. So we try to help them train them to be able to use the tools effectively for that task.

But yeah, I think that's a great one. I think what's also really interesting is that generative AI tools kind of help with the bedside manner, right? So, there are studies that show exactly this. You think computer can't have the human touch. It has a better human touch than a lot of doctors where studies that have talked about this and found this. And I think the same is true for lawyers. We have limited time. You go into a lawyer's office, how long do they have to meet with you? You go to a doctor's office, how long do they have to meet with you? These tools are tireless. They can be available any time of day or night.

So that's why we see people turning to use them as maybe the first line of defense for when they're experiencing a legal problem now in some cases. And increasingly, I think we're going to see more and more of that. I think it can play an important role in doing that. And what we're trying to do is get ahead of the risks that come with that. I will say the risk word at least once and think through, okay, well, let how do we take that benefit and integrate it into the safer tools, the rule-based tools like our document automation solutions to smooth over the rough edges and to give the user a better experience without taking away the safety of the tool.

John: Right. Okay. Well, so that's the perfect segue then into the next thing I want to talk about, which is where are the right places? And again, my audience, I think is mostly private practitioners, mostly smaller firms. How should they think about document automation, document assembly when they now have access to, you know, certainly ChatGPT and Claude and Gemini and maybe Vincent or some of these other, you know, legal specific AIs.

What are the sort of use cases for saying, okay, great, even though I can say, you know, hey, AI, write me a TPS report and it's going to get pretty close versus, oh, the TPS report is something I have to do twice a month and I can turn it into something that is going to be a guided interview that lets me sort of answer a handful of questions and is going to punch out the TPS report. I know there's not a bright line, but sort of directionally, where's the line between the right use cases for each of those things?

Quinten: I would say that the first place to think about it is, okay, well, how closely can I read this to make sure that it's doing the right thing for me? If what you're doing doesn't necessarily need to be exactly 100% correct and you just need to make it a little bit more human or add a little bit more detail, generative AI is okay to use to draft that. But often, I think you do already have a document that's your precedent that you just want to make easier to use. So why not just use generative AI to help you build that automation?

So we've been experimenting with that in a couple of different ways. One that we have that works very well right now is to build a template of a document. Imagine you have a Word document that's your precedent. You want to make it automatable. The old way would be like you go through and you add all of the document template syntax by hand. Now we just have you upload that document to a web page. It uses generative AI to read through it and mark all of the language that it thinks should be template language and puts that in for you as well. So it's a huge time saver.

John: And by template language, you mean the effectively the variables that are going to exist inside of the document.

Quinten: That's right. Yeah. All of the special syntax that a tool like Docassemble uses to put the information into the right place, to make lists and tables and whatever other things that you need to do to make it a more live document. We can do with generative AI as a first pass. It helps save a lot of time. It's not always all of what you need, but it can work as well with a 100-page document as a one-page document. And so the time savings can be really big.

John: Well, and that's huge. I mean, I know I have a long-term client, that estate planning practice has been using HotDocs for document assembly for at least as long as I've known him. So at least at least a decade, maybe longer. Um, but over the years, he's probably invested hundreds of thousands of dollars in custom work in order to templatize the, you know, and he does complex, high net worth estate planning, so there's a lot in it, but he's also spent a lot of money doing that work of basically going through his templates and a lot of his own personal time and effort as well to sort of quality check it and make sure that it's all right and getting the variables right and et cetera, et cetera.

So, it sounds to me like that's something that is the incremental cost of doing that is dropping rapidly if we can use generative AI to get a credible first pass or maybe even first and second pass at that work.

Quinten: Absolutely. Yeah. And it can help generate the questions too. So that side of it is something that you can use generative AI for.

John: That was going to be my next question.

Quinten: Then you take the interviews as well. You take those drafts and you have to revise them. I don't know what your experience is, John, but there's no end to how much a client is ready to tweak a product that you build for them. So you're not getting rid of that part of it. But at least you get to that prototype stage where they can give you feedback to tweak it a lot earlier with the help of generative AI. It might not cut the project cost down by half, but it's going to help make a significant efficiency gain for you.

John: Yes. And to quickly answer your question, I try to talk about minimum viable products. So let's figure out, you know, what's going to get us to a first usable draft and then actually put it into practice and figure out where we need to iterate and be intentional about iterating as opposed to having 20 rounds of iteration to get to the exact perfect thing right out of the shoot. That's not always practical, right?

Sometimes it is so important that minimally viable is actually quite a high bar to get over because viability is complex and specific. So, people hear MVP or minimum viable product and they often focus on the minimal part, but the viability part is also really critical. In terms of your, I guess, consulting practice, or maybe both, you know, your teaching and your consulting practice, where do you see it going?

Quinten: One of the areas of work that I'm really interested in right now, and we have a few projects. One of them actually is in Oregon, John. So I'm working with the Oregon State Bar on a project to help them with triage and the bar referral system. So, we know people don't always know what their legal problem is. A lot of existing tools where you try to get connected to a lawyer or to a legal aid program, you get to say, well, my problem is a family law problem. My family is a housing problem. And then you'd kind of drill down from there with very specific questions that are rigid and maybe hard to answer.

John: Sort of decision tree style logic.

Quinten: Yes. So, what we want to do is to replace a lot of that decision tree logic with the help of generative AI in more of a triage function. So someone can use their own words to describe their problem. We help them drill down to the right problem automatically with the help of the AI tool.

John: Yeah.

Quinten: And we can also ask follow-up questions there, which is important too. Because you have someone a great form. These are the things we need to know. Well, are they going to write one sentence or are they going to write a novel? You don't know. And generative AI can help be responsive to that to make sure we get the relevant information and it's enough to completely describe the person's problem.

John: I don't know anything about the specific project, but I do have familiarity with the referral service and Eric and his team. And I know that at least historically, one of the issues they have is that being a part of the lawyer referral team is a very stressful job, right? You're often talking to people who are in crisis or experiencing very stressful life situations. They're desperate for help. Oftentimes, by the time they've called the lawyer referral service, they've been told no by a couple of lawyers already. It's often not their first stop.

And so, I know it can be a very challenging emotional environment. I'm purely asking, is this something that is mostly happening over text through the website for now, or is it is it something that could eventually even sort of help agents as they're on the phone with people? I say human agents, right? Human people responding to those calls to be able to sort of use augmented intelligence, I guess, instead of true standalone AI to sort of help them with dealing with these people who are experiencing these challenging problems in a way that maybe is less traumatic for the workers.

Quinten: That is an area where we've been getting a lot of interest. I do think it's a bit of a future goal to have that kind of live intervention, but that is something that AI systems can do. So the trick is, can they do it quickly enough to intervene in a timely way in a way that surfaces actionable information? I think it's a really interesting problem. There's a colleague of mine, Hannes Westerman, who's at the University of Maastricht in the Netherlands now. He used to be in Montreal.

And when he was in Montreal, he built a tool called LLMediator, which is essentially just that, right? It surfaces interventions for a mediator to use in online dispute resolution to say like, "Hey, you might want to interject here. The people are getting really heated." And offer these suggestions to cool things down or to redirect people to solve their problem more constructively. So, I could imagine something like that in a call center as well, where it pops up with suggestions on the fly for the, in this case, the intake worker to offer to help improve the interaction.

John: Okay. So I just took it to an imaginary place. Help me understand what the actual project is for now. So I assume it's mostly text based.

Quinten: We're going to start with text because I think that's where we can have the most quality control. It is amazing how easy it is to do things like add voice modalities now, like out of the box, relatively affordably. It still does raise the cost of each interaction by like a 100 times. So, I think we budgeted this might cost us like a couple hundred dollars of AI usage in a year. And I we’ll have to see, but I'm pretty confident in that number having run the envelope calculations.

If you have a couple thousand or a couple tens of thousands of dollars to do the voice mode of like you've I don't know if you've tried the voice mode in ChatGPT. It's like a dollar a minute or something. So if that's an affordable number for you, which is not impossible to imagine that it is, then we can get that kind of interaction now for very little engineering time. It's still a bit expensive to use for every interaction for a big system like the Oregon State Bar though.

John: Well, and yeah, and the bar being, you know, someone who pays my bar dues every year, I want to make sure that the bar is being a good steward of that money. But as someone advising a private practice that maybe is a pretty profitable place and has an opportunity to use these tools, I don't think it's out of the realm of affordability to say, yeah, a dollar a minute or even a couple of dollars a minute in order to, you know, as long as the output's right. I think that's maybe the biggest concern is, you know, the AI is going to make a mistake that will cost me a client, cause me a problem somewhere down the road.

Quinten: I think that's a really important part too, right? Because there's this uncanny valley you enter when you start to have a voice agent that you're talking to. It's just like a little bit too slow to respond or you interrupt it, it interrupts you. Whereas when we have a text based system, people can use the voice to text feature of their phones so they don't have to type a lot necessarily. But because it's a text modality, they kind of recalibrate their expectations so you don't get that kind of weird uncanniness that you might with a voice system.

John: There's a middle ground too, which is I think what I started to mention in my imaginary scenario, which would be maybe the AI is listening and prompting a human, again, assuming it can present those prompts in time to be effective and be useful, that I think can help bridge the uncanny valley problem a little bit as well.

Quinten: That's a really good point. And yeah. And actually, I spoke with a couple of different legal aid agencies that are interested in trying to get funding from LSC to explore that. So we'll see what happens with those projects and how they get keep going. I think there are some people in the commercial call center space who are experimenting with some of those ideas too. It's just okay, are they tailored enough to what lawyers need and especially the needs of legal aid? These out of the box, not quite yet, but I think it's where we might start too.

John: Yeah. Well, and the added question of whether, you know, depending on who the agent is on the other or handling the call, is the AI going to venture into legal advice territory versus legal information territory? There's all kinds of fun things that we get to think about, right, in terms of the ethics and the and the regulation of it and all the rest. Anything else, you know, again, for the benefit of a small firm practitioner, probably likely in the people law sector. I mean, I'll say out loud and I think you and I have talked about this in the past, right?

Part of why I'm so passionate about smaller practice lawyers is that I think that they're an important part of closing the access to justice gap. And I think the more we can empower even full price lawyers, but certainly everyday private lawyers to be able to offer services a little more efficiently, hopefully a little more affordably, that's going to help sort of close that donut hole of people that don't qualify for legal aid but can't afford current price full scope legal help.

So, kind of with that market in mind, what suggestions would you have for folks that are maybe thinking about this combination of automating some documents, maybe using some generative AI, otherwise pulling technology into their practice.

Quinten: So, the tools of the trade that I use are really inexpensive. If you want to run a Docassemble server on your own, you can do that with your own systems administration skills for like $25 a month. You want to use ChatGPT to help you play around with doing something like pulling out case information from a document so you don't have to review it yourself manually to do that or my favorite use case for a CLE is usually I say like, hey, that nasty gram that you got from the opposing attorney, how do you like write a measured and polite response to that and tone down maybe your own knee jerk reaction to it so you can keep that relationship.

ChatGPT subscription, $20 a month. And you can use that and turn on the settings so it's safe to use with your client confidential information as well. If you're paying for that subscription, you're not using the free version of it. So now you're at $45 a month for really powerful tools. I help a lot of small law firms learn how to use Docassemble on their own. So we book like a couple of weeks, maybe a couple of months of once a week hour sessions where they're just learning to use it. And it's amazing what they go off and build on their own to solve problems because that's the resource they have is that time. And they want to make that trade.

There are other folks who hire me for the whole solution, and that can be expensive. But if it's at the right scale for their law firm, they're saving and making money on it right away as soon as they can cut like a 10 hour estate plan down to one hour of generating and reviewing it. So, because it's so cheap, I would just encourage folks to give it a try and see what they can do on their own. And maybe you're going to learn you don't want to do it all by yourself, but it's not very expensive to give it an a try.

John: I love it. Well, how do folks get in touch with you if they're interested in maybe working with you, learning more about how this could happen in their practice?

Quinten: I'm on LinkedIn. That's a good way to stay in touch and you can send me an email, Quinten@lemmalegal.com. So, Quinten, Q U I N T E N, and Lemma, L E M M A.

John: All right. So, and presumably Lemma Legal's also got information on the on the web page.

Quinten: That's right.

John: Awesome. Well, thank you so much for spending this time and contributing your expertise and for all the work you do around making the legal system work better for everyday folks because that is certainly what we need.

Quinten: Thank you. It was so great to catch up.

John: Okay, so a few quick takeaways from that great conversation with Quinten. Number one, systems design work, and whether that is workflow and process improvement like I tend to do, or technology tool implementation like Quinten works on, they're all there to help you create consistency, to ensure quality, and hopefully to work towards predictability in the deliverables around your legal work. When we create those standardized approaches, we're not just doing it to save time, we're doing it to ensure that our clients and our team members are getting the same sort of high quality experience from as many aspects of our law practice as we can.

So, the second one is kind of a personal one for me, and I've long maintained that lawyering is ultimately a caregiving profession, and I believe that's true. But I've been assuming that human to human interaction is going to be the way that we generate that caregiving experience. But both Quinten and then a few weeks ago, Damian Real have really challenged that assumption. They keep highlighting these real world scenarios where technology assisted care or computer assisted care is actually outperforming straight human interaction in some scenarios. I think that's something I certainly need to sit with and we probably all need to consider and probably is going to cause me to rethink when and how to create more of an augmented human touch for our practices.

Number three, I'm just going to straight up plug Quinten, right? If you're struggling with where or how to implement document automation in your own practice, maybe you're trying to get off of an outdated tool onto something newer, or maybe you're looking to adopt it for the first time, I really encourage you to check out Quinten and Lemma Legal. As they say out in Boston, he's wicked smart. And as you just heard, he's got this really thoughtful, measured approach where you just know he's going to take care of you and your practice.

I'm also going to put in a quick plug for next week's podcast episode where I'm going to air my recent interview with RJon Robins, who is the founder of the company How to Manage a Small Law Firm. And you've probably heard of RJon, you've almost certainly seen his marketing. A lot of lawyers tend to have strong feelings about RJon and how to manage. But he and I actually have a lot of overlap in how we see the problems facing small and mid-sized law firms and their owners. Now, that said, we have pretty different approaches on how we want to work with those law firm owners to solve those problems. As for which approach resonates more with you, I'm going to leave that for you to decide.

Okay, that's it for this week. If you have any thoughts or questions about any of the stuff Quinten and I talked about or for your law practice operations in general, please don't hesitate to reach out to me through my website, agileattorney.com. As always, thank you to the amazing team at Digital Freedom Productions for their podcast production help. And our theme music is Hello by Lunara. Thanks for listening, and you're definitely going to want to tune in next week.

This podcast gets production support from the amazing team at Digital Freedom Productions and our theme music is Hello by Lunara. Thanks for listening and I will talk to you next week.