How Financial Planners With Zero Coding Experience Built Their Own Agents - Matt Ragone, Range Transcript, AI Summary & Key Points

00:01 [music] >> What I really wanted to focus on was how our financial planning team ended up writing the agents and skills that we use on a daily basis to make our team as efficient and effective as possible. And none of us have any engineering coding background. I mean, we are as fresh in this pool as as you'll quite get. But this is the build guide, how we did it step by step so you can take it back to your own teams and maybe present it to the non-technical folks at your >> [clears throat] >> companies.

00:42 So, as I mentioned, a little bit about me, um I'm a financial planner, right? I'm not an engineer. Um I've been in the wealth management industry for about 7 years now. I did have a 2-year stint in media during COVID, but then I found my way back uh into uh the financial planning realm. Um I spend most of my days talking with clients about cash flow, taxes, investment, insurance, really just like the most riveting topics you can imagine.

01:08 Uh but uniquely, I lead the AI adoption for the planning team at Range, which is an all-in-one wealth management firm. And this AI adoption team, which we call the AI Council, um it was an idea created from our co-founder David Kusaydis, uh who wanted to have every part of the company represented um in this council, right? So, think of the AI Council like a round table for the most forward AI users at the company.

01:38 And we share best practices because if a member success team creates a skill that the marketers can use, we don't want to reinvent the wheel, right? We want to make sure we are operating as effectively as possible. So, thankfully, the planning team trusted me with that responsibility for the planners. Um and I also love that it's called the AI Council.

01:57 Um you know, I feel like I'm in Game of Thrones when I say that, right? So, I'm either in Winterfell or I'm at like the Red Keep uh kind of just being like the warden of the financial planners, if you will. So, and then the last thing I'll mention here is that, you know, we work in a regulated business as an RIA. So, the output that we're delivering is financial advice to real clients and the review process, it's not optional, right?

02:20 This impacts real people's lives. Um so, this isn't like a vendor pitch. It's really a method of what worked best for us. So, let's rewind back to earlier this year. I mean, we started pretty much in the AI realm as much as everyone did last year. We just use AI as a chatbot, right? Claude Code, uh Co-work for that matter, wasn't really out yet. But, we were just using it to redraft our emails, um fact-check us for certain tax code changes, uh or parse through like large documents.

02:47 So, it saved us a lot of time in that regard. But, the one constraint was that we still had to manually do all the work. We got a little bit faster, of course, with the AI adoption, um but the big changes really happened a little bit later. So, here's kind of what we knew, right? The biggest constraint to how much I'm able to work with a client or member, etc., uh it had to be how fast I could essentially type, right?

03:13 And that's always going to be a really bad bottleneck cuz whether I type 300 words per minute, 200 words per minute, I still have to manually do all of the work, right? And another issue that arose was say I was on a call with a client 30 minutes, 45 minutes, I can't work with any other client during that time. It's essentially time wasted in a way from the rest of our client base.

03:35 Okay? So, we recognized that bottle. Then what changed is in late January, early February, right? Claude Co-work completely changed the game, right? Now, I could be on a call with one client and I can be having uh my Claude Co-work agent completely draft a cash flow analysis, an insurance analysis, a tax review, all while I am simply servicing another client, member, etc.

03:57 So, now what happens is when the call ends, instead of me just working with one client, I'll be able to do multiple at once. Um but the end product now is that I have to review, adjust, and really see what our agent built so that I can make sure it is bulletproof before it gets sent out. So now, right, the game has changed a lot in that respect. Um so how do we get here in a relatively short period of time?

04:24 Um it took a lot more than just a good prompt. Two things had to be true. First, the agent has to know where the work lives, right? In our systems, it's actual client data, not in a separate chat window that I'm copy and pasting it into anymore. And through our platform, individuals, we call them members, is why you'll keep hearing me interchange customers, clients, members.

04:47 Members connect all of their financial data, accounts, etc. right onto our platform. So we at least pretty much know from insurance documents, investments, home, uh you know, real estate purchases, mortgages, we have their entire picture fully connected. So the agent has to know how to pull that document correctly or or that data correctly, know what's correct data to pull, incorrect data to pull, etc.

05:11 And then second, it has to understand how do we deliver advice at range, right? It can't just be delivering generic recommendations because we might be a little bit more conservative than what Claude CoWork or any sort of other agent would like to deliver. So we actually were ahead of this curve, luckily. Um you know, our team is about 20 planners. Before Claude CoWork really dropped, we standardized all of our processes.

05:39 How we deliver insurance recommendations, investment recommendations, retirement recommendations, so that if I'm a planner coming from a large wealth management firm and another planner is coming from a small RIA, you have some sort of reference document on how we deliver all of our recommendations, kind of across the board, okay? So that actually really saved us so much time when Claude CoWork came out because instead, again, instead of the wheel, it was "Hey Claude, just reference this document as part of your review

06:07 process before you kick off any of your workflows." Now, we had this data living in in these standardizations living in Notion pages, Google Docs, but we really just had it all kind of condensed in there, too. So, that's what really it took for us to get to a really good place with Claude Co-work. And then, we kind of identified five steps to build the skill or agent.

06:34 They're pretty much interchangeable I feel. So, one is pick, two teach, pack, test, and ship. Okay? None of them require you to write code, no code, and each one of them requires knowing the work really well, right? So, that's why the people doing should the work should be the ones building it, which we'll touch on a little bit later in the in the presentation.

06:58 So, step one, pick a workflow you repeat constantly. Okay? I'm talking daily, weekly. Your first instinct is going to be to pick like the flashiest thing, the most complex thing. I'm telling you from experience, do not do that. Okay? Pick what you think takes a lot of time, but can be automated very very simply. Okay? And and the great part about that as well is you will immediately know if the output is good, if it's bad, you'll have that judgment right away.

07:29 If it's more complex, it's going to take you a little bit longer to really reiterate with the skill or with the agent. So, for us, that was cash flow analysis, insurance reviews, and it was pretty much any recommendations I said that took a lot of manual work to do. Okay? Step two, I say show it one real case. I would end up now probably doing it two, three, four, five real cases.

07:53 Like, what does the final product look like, right? So, our standardization documents that I referenced, it just tells us how we do the work. Then, I will add to the referenced part of the Claude skill, what do you real cases look like of a cash flow analysis where it's an edge case, right? Or where I had to correct the agent's work. So that he can reference that again, so the next time it runs the skill or does the workflow, it looks much more uh efficient and understands some of those edge cases cases a little bit

08:27 better. Okay? And as I'm building the skill, um if you will, I am narrating to it why I'm doing certain things. So think like you're going to take a little bit more work up front to save you so much time down the road. For example, if I'm uploading a pay stub for a cash flow analysis or an insurance analysis, I'm letting the skill know why we're doing that.

08:49 Hey, in the pay stub, look if they have, you know, $3 deducted for groove group term insurance, right? Group life insurance for example. If I'm just uploading the pay stub, you know, you want to give the agent a little bit of direction if you can. You don't have to go crazy with it, but that additional narration has saved us so much time down the road as well.

09:11 Okay? Then three, right? Once you feel like it does a good job, pack it into a skill. And then we found two all the two best next steps to do as well. And I don't know if this is fully adopted across the board, but we've had a lot of success with it. When I'm prompting certain skills or agents, I have it ask me questions. Is this tax review a 2025 tax review?

09:33 Is it a 26 review? For the cash flow, you know, what tax code are we looking? What state do they live in? Now obviously some of this data they can pull right from um the profile on on Range, but there's some other things whereas is this member more focused on their homeowner's insurance or is it a disability focus that we're looking at? Do they have a whole life policy we want to review?

09:56 So when we prompt it, we have it ask a couple questions so that it can really do work in the most effective manner. And then second, what are the limits it has to respect, right? Again, we're regulated. Our recommendations have real-life impacts. So, every single skill is not allowed to send anything to a member, okay? It has to be reviewed by a CFP at this time.

10:21 And honestly, just moving forward, I feel like that's just going to be the case. Now, we do have a really cool built-in AI feature called Ry, which is CFP compliant and RIA compliant, which is very cool. That's built into it. But at this time, you know, when a member is asking for human interactions, we have to make sure that it is validated by us as well.

10:40 So, those are the two, I would say, things maybe people don't think about when packaging a skill that have really helped us out tremendously. And then step four, and this is this is again, this is before you're shipping it out to the rest of your team. I would run it on five to 10 real cases. Some should be like your standard cases, but you really want to test a few edge cases as well, so that when those cases arise in the future, the skill, agent, etc.

11:08 can really pick it up and know how to work with that specific issue. Now, what we do as well is you might have one or two other individuals kind of working with you to kind of stress test this skill as well. And then every time after I run it, I say, "Take what you learned, repackage that into a skill. Keep iterating on your skill. Keep making it better."

11:33 And then before I I fully ship it, you ask Claude or the skill, the agent, one final question is, "How could you make yourself better? What would you specifically do?" Again, update the skill, keep packaging it out. And then once you feel like it's strong enough to go, um it's time to ship it. Now, that's not where the work ends. I wish it was that easy.

11:56 I wish. But, tax code's changing, a lot is changing in the industry in general, but you need to ship it and it needs to have an owner and a reviewer, right? When you're a small company, it's probably going to be the same person doing both of those things, but as we're We're our team, we have about 20 planners now in the team and growing. Um you know, if I ship it as the owner, I might have my colleague Joe review it as the reviewer because he might have a little bit better subject matter expertise in insurance, for

12:22 example. >> [snorts] >> So, where are we now? You know, we're about 6 months into this uh experiment, if you will, 6 to 8 months in, and we have about 24 skills across the company, 17 just in financial planning. And the really exciting thing was we have been able to build off each other across the firm, right? A lot of the skills agents were building, um you would think it's really the engineers running with it, but they're they're so good at what they do, they understand it in their realm.

12:51 We have been able as planners, marketers, team success, uh member success team, right? We have been able to work off each other and create even more skills to make our work as effective uh as possible. And our next evolution, by the way, we have a uh we created an agent named Ranger, right? We're Ranger, so this is Ranger, um that assists everyone at the company with tasks right now.

13:13 It lives in Slack. Uh it can go on to our our our dashboard to assist with um kind of other member task, but we're building that out actively, which is very, very exciting. >> [snorts] >> And what we've really found as well, and I kind of touched on this a little bit with having an owner and a reviewer, you really want to have the experts in that area build out that skill, right?

13:39 If someone is an insurance specialist, they should be the one iterating on the insurance skill because they will pick up on stuff that myself as kind of a general practitioner might not uh might not know right off the bat, right? Um the way I think about it, certified financial planners, we can kind of be like a quarterback. Sometimes you got to air it out 50 yards.

13:58 Other times, you know, you got to ground and pound, chew the clock out. And that's where sometimes you got to hand the ball off to the CPA, to the investment team, and then they can assist even further, right? So, again, we have our CPA team, our tax team really help with the tax skills that we've built in the agents, um but then I can kind of come in as just the subject matter expert with really how to build the actual skill.

14:22 They'll be the subject matter expert on the actual content the skill needs to know in that matter. Okay? And now the bottleneck has completely moved, right? The bottleneck used to be, how fast can I type? How much can AI really draft for me, rewrite my words, you know, fact check me as well? Um Now we're starting to ask the question is, you know, when everything gets shipped, uh you know, who is making sure all these get tested, updated?

14:49 How do we know that every skill is fully up to date? We're almost going to start having planners, I would imagine, almost be AI agent managers, right? That's how I see the future of this industry going. Of course, you'll still be able to access certain planners on calls because the human's touch is still there. It's going to be there for a long time, but how can we make sure instead of providing advice for three to four people a day, right?

15:15 We're providing advice to 100 people a day per planner. That's really the goal. And what's super exciting is all this stuff is moving fast, and you know, you're either a part of it or, you know, we're seeing a big divergence right now in firms who are adopting it and who aren't. But you know, this has kind of been our adoption uh policies, if you will, or our success guide with AI adoption.

15:37 So, I hope you guys can take this across your firms and and again, you know, this is coming from someone with no engineering background, no coding background. If we can get AI adoption as strongly across our company as you all can at yours, um I think anyone can can kind of implement this as well. Well, so thank you so much. My name is Matt from Range. If you want to join an awesome team, we are hiring as well. But thank you all for joining. Uh I really appreciate it.