3 Steps to Law Firm AI Adoption, No Tech Skills Needed - #250

Most law firm AI adoption in the US stalls on floor one because owners think they need technical skills. 

But the 3 basics do not need advanced technical skills: prompting, skills creation, and tools integration.

We’ll show you how in this episode.

—

 ➡️ Register Here: www.RocketClicks.com/start-your-ai-factory   

📲 Subscribe Now: https://www.youtube.com/@karls.anthony

                             https://www.youtube.com/@TylerxDolph

📝 Schedule a FREE Family Law Firm Audit: https://rocketclicks.com/schedule-a-family-law-quick-audit/

---

📄 CHAPTERS  

0:00 - Law Firm AI Adoption Without Any Technical Skills

2:53 - Stop Typing: Five-Minute Prompts Beat One-Line Asks

10:11 - The AI Prompting Rubric For Lawyers: Payload, Classification, Output

13:29 - Skill Creation: Turning A Repeat Prompt Into Firm SOP Automation

17:43 - Skills Plus Tools: Legal Transcript Automation Into Salesforce

20:19 - Intake Call Scoring: 10 Calls A Week To Every Call, Every Day

----------------------

Ready to find the accountability partner you need to build your dream family law firm the same way we grew Sterling Lawyers? 

Follow these steps:  

1. SUBSCRIBE TO JEFF'S NEWSLETTER: https://jsterlinghughes.com/  

2. BOOK A FREE 30-MINUTE AUDIT WITH US: https://rocketclicks.com/schedule-a-family-law-quick-audit/  

3. CONNECT WITH US: 

LinkedIn: Jeff Hughes, Tyler Dolph, & Anthony Karls, 

Facebook: Jeff Hughes, Tyler Dolph, & Anthony Karls, 

Instagram: Jeff Hughes, Tyler Dolph, & Anthony Karls

4. TELL US WHAT YOU WANT:  
Tell us in the comments if you liked this episode and what other kinds of episodes you would like to see.

Transcript

0:00

Afraid to adopt AI because you're not a techie? The first step actually doesn't require any technical skills. Here are the first three floors of our AI factory blueprint at Sterling Lawyers. Welcome back to the Sterling family Law Show. The show for family law firm owners who want to grow their profits, multiply their cases set a clear roadmap on what they need to do to build the firm of their dreams. Our episode breaks down proven business growth, digital marketing, and revenue first strategies to help you scale smarter and not just be busier. I'm your host, Tyler Dolph I'm also the CEO of Rocket Clicks, which is a business growth consultancy that works exclusively with family law firms to drive more revenue and increase profits to full service development. Okay. Today we are going to break the false belief that you cannot adopt AI in your firm. We're going to walk you through the first three floors of our AI factory, where you're going to learn about prompting skills and then skills, plus tools to help you build real AI efficiencies in your firm. We are continuing our series on AI for family law firms leading up to our webinar. That's coming up in a few weeks, and we have Tony Karls who is the co-founder of Sterling Lawyers, with us today to continue our conversation. We are going to talk about the three floors we're calling the floors of AI adoption in family law firms. And today is all about the, you know, the family law firm owner or the practitioner who's like, this is way too technical. I don't know how to do this. You know, it's way too difficult. And we're here to dispel those rumors or those feelings because honestly, it's not that hard. And I can tell you from experience, I was in that same boat and have become a little bit of a I'm obsessed now because it's so easy, because it's just prompting and chatting and all the things. So, Tony, we're going to start with the first floor, which is prompting, right? For one is prompting. It's just telling the AI what you want, the same way you would explain a task to a new associate. So give us a little like background on this kind of first floor. This is the 101 of using AI. Yeah. So, prompting. Typically, when people are in this first stage, they're super afraid to use it. That's kind of the. They're like, I don't know if I want to break something is just going to take over my computer. Is the world going to end? That's typically where people are when they're in this space. They really don't know what the tool can do. So they're reluctant to give it enough information. And then they start kind of dipping their toe in the water. My advice to people would be. Stop typing. Use text to speech and give it as much information as possible. Typically when I'm prompting, I'm I'm talking for, I don't know, 3 to 5 minutes depending on what I'm doing, and I'm just giving it a ton of information. Typically, what I'm ending the prompt with almost always is recap what I just told you. Build me a plan for execution and ask me questions that I may not have given you information on so that I can fully complete my ask before you start doing work. So that's typically what I'm if I'm if I'm just using the, the interface and doing a prompt, I'm typically ending with that. I'm giving it a whole bunch of information, and I'm ending with the directive for it to explain back to me what it thinks I want, and then to ask me further questions so that it has a better idea of what I might not have given it, that it thinks I should give Yeah, I think we we lovingly refer that refer to that as like vomiting into the AI. Just all of your thoughts and feelings. Put it all out there. Give it all the context. And then I think the most important part of what you said is asking that at the end, ask me questions to make sure you're clear. when I'm building things with, I build in loops. So it's a little some of this is built in to to the process. So I don't always have to do that. But that's, that would be what we would call floor six where a lot of what I've a lot of what I'm going to be describing as we're talking today is kind of built in this process of, of a loop orchestration where it's it's going to do that first thing that I just described. And then it's also going to do what I would call the second step, which is okay. Now it delivered you a plan. Now I'm typically going to tell it to research that plan. Go do research on that. Plan on anything I've told you in the past, as well as best practices on that particular item, and then do an adversarial review on what you find to determine whether or not the plan we just described is the best plan for action. That would typically be like a second prompt that's built into my loop orchestration. So I've just done a whole bunch of review on the thing, and then I'm going to get a different version of that same plan that's a lot more comprehensive and a lot more complete before I start doing any work. The reason for that is how how AI is biased. the thing that I want our listeners to know is that, like, AI is built to be a confirmation engine for you. Oh great idea, Tony, you're so smart. Yes, of course, that's the right plan. And you have to get out of that loop. Yeah, well that's true for different models do do things differently. So I'm I'm talking specifically from the cloud experience. So Claude is very biased towards collaboration. It's a very it's very much a collaborative engine. So it's going to it's going to be more agreeable by default. It's not going to disagree with everything you say, but it's going to be like the friendliest coworker you've ever had. So it's not going to be abrupt and abrasive. It's going to try to agree with you. It's going to try to figure out how to agree with you while telling you the truth at the same time, which is not great when you need it to really actually give you a good answer. So that adversarial review thing that I put on the end of most of these, or I have a I have a skill called research critic. What I've done is like I've put I've put it on itself. So it's going to agree with me that the result that we just created isn't as good as it could be, and it's going to try to identify how to make it better. And that's but now it's still agreeing with me. So it's no longer in an adversarial position. And it's really important to know which models are biased towards which things. At least that's what I found with Claude. If you're using grok and leveraging grok heavy, it's it's biased towards finding the truth. And you will see if you actually run a grok heavy session that it's going to dispatch between 8 and 12 agents all at the same time to to do exactly what I just described. And if you read the back and forth between the agents, they actually argue with each other constantly until they can come to a consensus with all of the data that they found on why the end result is the correct result, because it's it's biased towards truth. So each big picture, what's really good to know is like each model is biased towards something, and the ecosystem in and of itself is biased towards efficiency. And like this is everybody knows this, but they don't know it. So everybody's aware that we have a data center problem in the United States where everybody wants to these AI companies want all these data centers built and local communities don't want them built. And there's this like push and pull. So what does that actually how does that show up in the real world? It shows up in these models. So all of these models have a compute resource constraint currently that exists that showing up in this data center conversation that's happening nationally, and how it shows up in the code and in the use of AI is it's going to be biased towards efficiency. So the reason I have to force it and tell it to do research is it won't actually do research unless you explicitly tell it to it will it will use its memory. So AI has it's a huge model like it's a it's a trillion a couple trillion parameter model that runs, which means it has a lot of context for a lot of different things, and it has a pretty robust memory, but it isn't perfectly accurate. And you can improve it by telling it to do research. It won't do research by default. It will be lazy, what I call lazy because it's biased towards efficiency, and it's going to use its own memory to answer the question or use its own inference on what you've said in the past to answer the question. It's not going to actually go out and validate that. Validate what you've asked to be true. It's going to it's going to find what it thinks it might be true. I've my personal perspective is I think most of the answers I get from AI are about 70 to 80% accurate, just on a naked prompt. If you give it a lot more constructive feedback and force it to do research and force it to do an adversarial review, you're going to get way closer to a true statement than a biased memory, if that makes sense. So there's a lot of these edges that are important to understand, because you might end up getting frustrated because you don't understand what's actually happening within the system and how it's biased. So what are some of the things that new users must do when starting out? Prompting. The way I train people or when I'm talking to people. Like I said, give it a lot of information. So just fully describe what you're looking for. A good rubric for, for that is all of these, every single prompt or every single interaction you're going to have with AI is really three, three things. It's it's a raw data payload that you're giving it. So whether that's just your word vomit or your word vomit, plus a couple examples, a couple documents that you're going to give it, it's you're giving it a whole bunch of information. So think about like what is all of the possible information I should, should consider giving it so that I can do a task. Second is a classification process. So it needs to classify the information that you've given it against the real world. So it can take all of your raw, the raw payload, the information that you've given it and make it into an understandable structure. That's the second thing. And then the third thing is the output. So I also love to give it out. Puts examples of outputs that I want it to look like when I'm done, so that it has an inference of here's the raw payload, I need you to classify it in this way, shape or form. And here's what I want it to look like at the end. I want this in a printable PDF. That's eight and a half by 11. Yeah, yeah. Literally I will give it a give it a version of something that was completed in the past that, that I did basically like if I'm doing this for like one of our processes internally we do a. We after our consultation. So we have consult legal assistance. So non attorney salespeople do our consultations. Our attorneys want a specific output after the consultation so that if the client funds they have a dossier of what needs to happen and when. So our consult legal assistance take the take the consultation that they recorded. It gets transcribed and then it gets classified in the proper way. And then it has an output that it's given. So like when we created that we created the manual. We gave it a version of the manual end result that we used to have to manually do with Vas in Philippines. And we gave it the the transcript, we gave it the output. And we talked, we talked to it about what we wanted and then it now it just produces the results for you. Dump a transcript. So it works every time the same exact way. But it's because we created we created a skill for it, which is kind of where we're going next. But it started with basically just prompting, give it a give it a payload, tell it what you want it to do, give it an example of the output, and you'll be surprised the more information you give it and the more specific you are with that information. It's going to be almost exactly what you want at the end. And the more information you give it, the better. What if your family law firm isn't behind on AI? It just never started on the right foot. A prompt that someone wrote, a tool on a paralegals laptop and SOP that nobody opened. None of it survives. The person who made it. The work still roots through you because nothing lives in a system. Because AI isn't something you buy, it's something you build. We built ours at Sterling Lawyers, which is a $20 million family law firm, and we're teaching the journey behind family law firm AI factory, starting with the four AI first that you need to know pricing policy, platform and progression links in the show. Notes below to know more. so prompting you got to talk to it. You got to give a lot of information and best practices would be to ask, what am I missing? What questions do you have to make sure we're clear and then make sure you have an output. Yeah, I was going to say it's like when you prompting, think about the three steps. What's the raw information you need to give it? What's the classification that you need to do. What's the output. And then the final piece of it on any given prompt is ask me questions and clarify and confirm. You think you think you know what I'm talking about. And that's that's probably a good setup for any prompt on anything that you're going to do is just those simple, those simple things. Moving on now moving up to floor two, which is skill creation. Now this is a little bit harder, but I think it's really important to think about it. Like you're onboarding a new team member. You have to teach AI to follow your firm's SOPs so that you don't have to read the task every single time, because if you just use prompting, you're basically starting from scratch every time. So if prompting is asking AI something, once a skill is writing a template for the output so that you get it the same way every single time you command it to write, a good prompt can actually be repurposed into a skill. Talk to us a little bit about that, and maybe use a Sterling SOP as an example. Yeah. So I'll just use the one that I just, I just mentioned. So our, our non attorney sales folks are constant legal assistance. They are required to do what's called the console notes in a certain way so that when, when and if a client funds, the attorney has the correct CA for onboarding the client. That used to be done manually, like I just said, by a VA of the virtual assistant in the Philippines. Now that's done with a skill. So basically what we did is we talked we we talked to it, gave it the raw transcript, gave them that, gave it an example of what an output looks like, and told it to create a skill for that exact thing. So now every time, yep. You just tell it to create a skill. So like when you if you use it and that's if you're using Claude. So Claude's very good at skills I believe I haven't, I haven't used them in open AI. They, I think they recently released the, the functionality for that similar similar functionality to work in OpenAI more recently so that that might be there and you can do some of the same things in Gemini. They use a different terminology for it. And the same thing with grok. So these are call it a little bit different, but it's the same thing. It's an SOP for the AI to follow. So now for us every time we we load a transcript a consultation transcript and tell it we want the consultation notes, it knows exactly the output, and it's going to go through that whole process and just dump it, dump it out for for our team And so just to be clear, when you go into Claude and you put in the consultation notes, do you say use your skill to execute this task? Yeah. Once you get used to it, you're not even going to have to give it that much explanation. You're just going to you're just going to say, I need consultation notes, dump the transcript, and then you're you're going to be done. It's going to be it's going to look like a really lazy prompt, but it's because everything's built into the the skill. That's going to be where all of the information is. And it's really easy to create a skill because by default Claude comes with a skill to build skills. So you just tell it when you're starting this, if you have like an idea of something that you've done over and over again, or you want to copy what I just talked about, you're going to open it up. You're going to say, I want to create a skill for for this, and then you're going to explain it the way that I just mentioned before, what's all the raw information that it's going to need? How are you? How does it need to classify it. And what's an output example that you want. Then tell it to ask you other questions to complete to make sure it's complete in its information. And then it's going to run a process to then create a skill file that you can save. And then the next time you open a session, you start a new chat, dump a new transcript, and you're going to see that it works exactly the way that the one you just built was. So typically when you build a skill, build it separately and then use it in a different session, don't smash it all together. You're probably going to end up with a weird skill that's Uber specific to one one specific use case, which is not not the functionality of skills, typically. it. Okay, so now we have prompting. We now have. If you're going to do the same prompt over and over again, you're going to build a skill. And then I think if floor three is skills plus tools. So now we're getting out of the AI and we're leveraging it or connecting it to other things in your business like your CRM or your file storage or your billing an accounting software. Talk to us about how we leverage this with our CRM, which is Salesforce. So when we do consultations they're recorded in Google Meet. And so there's a transcript file. There's a video file and a transcript file that gets loaded into the the agent's Google Meet folder. Where we started was all this was manual with the VA. Then we went to okay, now this is going to be done via a skill, which means the the VA needed to take the transcript and dump it into the. Dump it into Claude to then execute the skill and then copy and paste it out, put it into Salesforce, which is faster than the first version, but it's still that's still kind of clunky. So the third version is now you can do that. You can do that with the tool. So basically you can have a a Google Drive tool. You can also have a Salesforce tool that it can you can say, all right I need to do the consultation notes for XYZ client. It's going to go into Google Meet. It's going to find the folder. It's going to identify the transcript. It's going to pull the transcript. It's going to classify the transcript. It's going to create the consultation notes. And then it's going to save the save the consultation notes in Salesforce field that you would save it in all in one process. So you don't have to do anything other than tell it to do the thing. So like that would be the floor three, where you're starting to automate some of this stuff so that you don't have to manually copy and paste and do all this back and forth, which is faster than doing it faster than doing it manually, where you have somebody else listen to the transcript and create these consultation notes. But there's like versions of like how do you automate this? And like that's, that is a skill plus a tool. It's just a more advanced version where it's pulling the information and then pushing the information to where you do it through tool connections. So give our listeners the actual real life story here in the fact that we want to coach our intake team on a regular basis. Before we were doing this, maybe monthly or weekly. Now with the use of AI, how is that going? So we used we have a team quality assurance team that listen to our phone calls. So we have 12 agents on the phones six consultation, six council legal assistance that do consultations. So we had a QA team that reviewed ten phone calls from each agent per week, and then ten consultations per agent each week and scored them. So and then we would take that scoring and give the feedback to the agent so that they could improve their process. It was an iterative improvement process, which, you know, most firms aren't doing that at all. So that was that gave us an edge on the sales side on how do we how do we keep our lead to set rate really high? How do we keep our shows to funding really high? But now what we can do is with with AI is we score every call that comes into the call center and every call that is outbound made to our prospective clients. And that's every single day. And all of those calls then get scored, patterns get identified and a report gets delivered to because now we don't have a QA team, we just have a trainer. We have a dedicated person that's now sitting in a trainer seat. And they get that report every morning and they see the pattern for all 12 agents. They see the pattern for the team. They bring that pattern into the the huddle every morning, and they do a little mini training every day for improvement. That's based on like, what are the patterns we're seeing across the team? And then that trainer meets with the individual agents and works with them on their own personal improvement plans. So that's now in place. It's way better backed by data because it's it's all of the information, not just a really small sample size. And the same thing is happening with our console legal assistance. What they do every morning now is they log into the tool and they see where they how they did yesterday on. Typically they'll do 5 or 6 consults each and they'll see like how how did they score, what could they improve. And they're all super. They're very competitive. It's a very competitive sales team. So they all want to get better. So it's fun to see them use the tool and get engaged and like really get into the self-improvement process because they they want to win on kind of their the scoring apparatus. Because when they when they win there, they see higher results from a closing perspective. And obviously that impacts their compensation. Incredible. They're getting real time, basically real time coaching able to improve so much faster. And everybody wins Yeah. Like we've we've been able to put in some other things in the QA process because these calls get scored as soon as the call ends. So when a call comes in, as soon as the call hangs up, it, then it, then the application grabs grabs the webhook that is sent and transcribes the information. And one of the things that we've we're currently working on implementing is a real time, what we call refer out. So if you refer a client out because we didn't, it doesn't qualify for our service. What we see on newer agents is they do that wrong. Like maybe they're not familiar with all the areas that we service. And they like screw up. They they don't. They misremembered that Aurora is in our service area down in Illinois and they refer it out. Now we're building a tool that's going to catch that in real time so that a we can correct the agent today, but B, we can call them back right away and be like, oh, I screwed that one up. We actually do service you. Let's have a conversation. So that's going to happen in real time. As soon as the call hangs up it's going to get scored. And then any of those notifications are going to go right to the intake supervisor to do immediate follow up. So those are those are other things that we're going to be able to add in to keep our quality up and do really fast coaching on what I would call like emergent issues if we if any of those arise. incredible. So if you're listening to this and you haven't dipped your tone AI yet, this is where you start. Our next episode is going to talk through the other three floors. So we're getting really into the weeds and all the automation coming up. But Tony, really appreciate your time today. Appreciate you walking us through this and I'm excited for the feedback. All right. That is the first three floors prompting skills and tools integration. Each one is built on what your firm already has and doesn't require any pro level technical skills. Next week we climb the rest projects, orchestration and the AI factory itself, where it all runs as one system. That's the last step before our live webinar called AI first for Family Law Firms, where will hand you the full blueprint of everything we've discussed in this series from start to finish. It's free, so the slots are quickly running out. Click the link in the description to save your seat. Today.