AI in Action: Transforming Wealth Management from From Portfolio Design to Client Proposals

This high-impact session will showcase how wealth managers can leverage today's leading AI tools to enhance every step of their business process. John's session will feature live demonstrations of tools like Claude, ChatGPT, CoPilot, Perplexity, Gamma, and more to show how you can leverage off the shelf AI tools today to enhance your practice. Attendees will walk away with a handout of actionable use cases they can implement immediately upon returning to their offices. By the end of this session, attendees will be able to:

- Understand how generative AI fits into the daily workflows of wealth management.
- Identify at least five practical, low-lift AI use cases across the client lifecycle.
- Evaluate and experiment with leading AI tools suited for advisors and support staff.
- Design a basic AI pilot plan for their practice or firm.

Transcription:
Transcripts are generated using a combination of speech recognition software and human transcribers, and may contain errors. Please check the corresponding audio for the authoritative record.

John O'Connell (00:08):
Let's just dive right in. How is everyone? Hopefully you've gotten a lot out of today so far. I said no slides. So what I'm going to do is you're going to get two slides up front that are really setting the stage for what we're going to do. And then we're going to be doing artificial intelligence live from the stage. So why don't we dive right in? A couple of things you're going to see first is we're going to be demonstrating generative AI. So what does that really mean? Generative AI is a broad category of artificial intelligence systems designed to create new content. And what you're going to find is what we're going to do today is generate a bunch of brand new content based on some inputs that we're going to give it. Now, generative AI covers a variety of different things.

(00:52):
Those outputs can include text, which what you're probably familiar with with like a ChatGPT or a Claude. But it also includes audio. It can create audio sound for you. It can create video. It can create synthetic data. It can generate code for you. And it can create images. And you're going to see a good portion of that today in our demonstration. Couple of quick examples of what you're going to be able to see, not just today, but just in general. Video, SORA. If you haven't seen anything generated by SORA today, it is mind blowing. You should definitely check that out. Runway, Pika Labs are generating video live. So you basically give it a prompt and it will generate video for you. If you want to see yourself in Blade Runner, you can actually put yourself in Blade Runner by taking a look at any of these video applications.

(01:40):
Text to speech. ElevenLabs, Speechify. These are tools to be able to mimic your voice. With only 30 seconds of your voice, we can basically mimic you to the point that your own mother will not recognize you on the phone. With more audio that we can get, usually within the range of about five to seven minutes, we can not only capture and imitate your voice, but we can imitate the entire intonation of your voice. All right? We're not going to get into today some of the nefarious things that can happen with AI and what you're going to see around that moving forward. But know that that's out there. Music generation, you can literally put in any composer or any music you've ever heard in your entire life into some of those programs. And Suno, for example, will mimic that voice. If you're a big fan of the band Boston, you can't really mimic the guitar of Boston because he invented an amplifier to be able to generate that.

(02:35):
If you put that music into Suno, it will actually generate it for you. Images, DALL-E, Midjourney, Stable Diffusion are all really great image generation. You're going to see a little bit of Midjourney today. And then programming assistance. To give you an example, Salesforce today generates 30% of their code using artificial intelligence, not with a person. There are many wealth management technology firms today that are generating a large portion of their code using artificial intelligence. Right now, some people at Oracle are saying that they're going to be generating up to 40 to 50% of the code at Oracle via artificial intelligence. So those are absolutely coming and you're going to see more of those. What are we going to show you today? Monte Carlo simulation. We're going to generate a Monte Carlo simulation live for you.

(03:25):
We're going to generate an investment analysis. We're going to create a financial plan and we're going to generate the proposal for the client all live right now. So let's dive right in. Let's get into some use cases for a moment. I'm going to minimize the presentation. You'll all get to see Darby O'Gill with the little people for a second. So let's say, for example, you wanted to capture some data from your clients. You're a certified financial planner. You want to be able to generate some simple financial planning data for an emerging wealth client. How do I capture that? What data should I capture as a CFP? This is actually going to be running in ChatGPT-5. This is the off-the-shelf, $28 a month version of ChatGPT-5 that you're going to see. And it's going to generate what goals it should collect.

(04:16):
I've got to keep remembering my stuff's on this side and not that side. So these are some of the things that it would recommend that you capture. Well, that's pretty interesting. So it's still generating some things that you want to be able to capture to be able to create a CFP level plan for someone. Now, for those of you that are CFPs, you know that the seven steps of a CFP recommend you capture about 40 points of data to create a financial plan to the CFP level. And if you want to create what's called a complete financial plan, that's usually about 60 data points. So what this is generating right now is just some of the data points that you can get. While we've got those data points, why don't I say this: I'd like to create a web form that actually captures those.

(05:01):
So why don't you go out for me and create a web form to capture this. I want to capture this on my website. I just want a form that I can pump into my website really quickly to capture this data. Now you'll notice what I said is don't worry about the database for now. If I didn't say that, it will actually generate for me not just the web form, but it will generate a database for me and it will put that all together in a Docker container that I can just deploy. For those who are technical in the audience, you could literally just go right from this to deploy it. Now, it's writing the code for me. If you take a look here, what is this doing? It's actually writing all the HTML code to generate the form. While that's running because of our time constraints, why don't we do something else in the meantime?

(05:45):
So if ChatGPT is going to run for that, let's do something else over here. Let's say in Perplexity, a different AI program. In Perplexity at the moment, let's say I'm an expert financial analyst and I want to analyze the financials. My client calls me and says, "I work at Oracle. I've got a pretty solid concentration of Oracle stock at this point. I'm thinking of buying a beach house and should I sell that Oracle stock right now or should we do something else?" To do that, what you'd want to do is take a look at what's the performance of Oracle stock right now and is now a good time for that client to sell some of that Oracle stock. It's not out of the question for you to be able to pull the Oracle 10K and then to ask a couple of things.

(06:33):
You're a certified financial analyst. I'd like you to analyze the attached financials and provide me with this: What's the company's performance? I'd like you to calculate the common ratios. For example, everything that I would need for a fundamental securities analysis. Also, I want you to outline your expectations for the company's performance of the future, outline any concerns you have around their business operations and provide me with a final assessment. So we'll kick that off. It's already read the document. It's already read the Oracle 10K and it's now coming through with the analysis. It's created all of the fundamental ratio analysis that you would create. Here's your operating margin, your net income margin, earnings per share with dilution included, the return on assets, the current ratio, debt to equity ratio. If I wanted the Sharpe, I could put all that in there. So it's actually continuing to generate my analysis.

(07:29):
And here it is. In summary, Oracle's well positioned for continued performance with moderate risk to investors that could benefit from its stability. Let's say for an example, that's good, but why don't I say at this point, how can you modify this analysis to be like a CFA level? I want you to tell me what would a CFA do differently from what I just asked you to do. And now what it's doing is it knows everything that a CFA would need to do and it's now running through and generating for me what's the difference between what I asked it to do and what would a CFA ask? Here's some CFA level prompt modifications. Here's what the CFA prompt could look like. Now I'm not that great at AI yet. So let me ask it: Can you generate a CFA level prompt for me to use in the future?

(08:37):
So now what the AI is going to do is it's going to go in and create a new prompt for me that's better than the one that I gave it in the first place. And it's creating it for me as a template. So now I could use this anytime I want. Here's a CFA level prompt, suitable for professionals in financial analysis. Using the attached financial statements, perform a comprehensive CFA level analysis that includes all the following. Now you're going to notice this prompt is a little bit bigger than what I had before. When we were talking about prompt engineering, if I wanted a real prompt to do this, this is the type of prompt that I would use. Notice that prompt is a lot bigger than what I gave it before. But let's say, for example, I wanted to do that.

(09:25):
I want to use that real CFA level prompt and now it's going to go off and it's going to generate for me a formatted 20-plus page financial analysis on Oracle stock and it's going to do that using notice what I've done with this prompt. I'm using things like role and context. If you tell AI a role that you want it to play, it will significantly improve the result set that you receive from the artificial intelligence. In this example, what I've done is I've given it the role. You're a CFA charterholder and you're preparing this sell-side level equity research report for an institutional investor level. I've given it the inputs. I've told it a methodology that I want it to follow. I've told it how I want the report to be structured. Notice the boundaries that I've provided for artificial intelligence here.

(10:21):
I've said to it, I want an executive summary between 150 and 200 words. I want to highlight the financial strengths and weaknesses, summarize the revenue margin and earnings trends, and then I'm going to go through the things like I want the financial performance overview, fundamental ratios, and the cash flow. What this is now doing is this structure ensures that you meet the needs. Okay. Run this. Run that analysis for the Oracle 10Q.

(11:03):
It's going to include all the graphs. It's going to run the full analysis. Here's the executive summary. Does this answer the client's question directly? No, but it probably gives you a lot more information that you need to then get on the phone with the client and explain to them this would be a good time for you to take a look at reducing that Oracle position or a bad time for you to do it. That analysis, how many people in the room think that that analysis would take you more than three or four hours? Right? Three or four hours, you're really good. This analysis is generated here now in just a few minutes. And if I wanted to, I could ask. By the way, this is pretty big, right? Now let's just do one other thing here. I'm asking you to provide me a confidence level that this is a CFA level analysis. Tell me where I can improve it.

(12:12):
The AI will now go in, analyze this, and provide me with a confidence level to tell me whether or not this fits a CFA level. Notice, the confidence level is moderate to high here. It's going to tell me here's areas where I can improve it to reach a full CFA level rigor. It will tell me exactly what I need to do. It will tell me down here it's going to probably ask me to connect this with some other data. Sometimes it'll say, "Can you connect this with other data sources so that I can do more of a momentum-based analysis regarding Oracle's stock?" That would mean that what you can do is FactSet offers connections to ChatGPT. If you're a FactSet subscriber, you have these connections for free right now. You would literally just point this to connect your ChatGPT to FactSet and it will bring in all the data you have access to in FactSet.

(13:16):
So all the historic data would come in and would provide you with a much more detailed historic analysis other than what you just find in the 10Q or the 10K.

(13:26):
So interesting, let's take a look at where we are on our Monte Carlo. So what we have here, oh, here we are. Now what I've done is let me show you the first where is my GPT. Here we are. Okay. So it created for me my input form. Let's preview that for a second and see what that looks like. Here's the client intake. It generated all the HTML that I need to be able to take this, send it over to my marketing team or whoever runs my website and say, "I need you to just pop this on the website for me and connect it to a small database behind it." Your web team would be able to do this very quickly. So here you notice it's capturing their name, date of birth, full name, right? Some of what they want.

(14:16):
If I hit next, oh, so it had required on this form. I didn't take the required field off. I could take the required field off and then it will go through the five levels of the form and you just go next, next, next, next to capture all the financial data. If you didn't like the way that it did this in GPT, we could also do that in Claude to generate the form. But first, let me show you in Claude one last big demonstration I want to show. It's multi-part. Here we are: You're a financial planner of the wealth management industry, you carry a CFP designation. Can you do me a favor and run Monte Carlo simulations with these capabilities to it? I've interviewed a set of clients, Harry and Ginny Potter, Harry's 47, Ginny's 43. Life expectancy, Harry's pretty much going to die at 90. Ginny's probably going to make it to 92.

(15:12):
Here's some financial inputs that we've got. This is basically what you would get out of any client interview that you've done. So we're going to run this really quick and it's going to generate the Monte Carlo simulation.

(15:25):
As we're doing this, notice I did for only a thousand iterations. I only did that for brevity while we're on stage. I could do this for 10,000 iterations with any of the financial planning software it does today. 10,000 iterations on a financial plan when I'm running a Monte Carlo simulation reduces any variables to what we would call statistically insignificant. In other words, if I got 10,000 iterations, I'm not going to get any more different types of iterations coming out of that. So 10,000 is normally the number that you'll find in all of the financial planning software. I just put it at a thousand right now because we don't have all the time in the world for this. So it's now running here in the background. It's going to generate the Monte Carlo. It's checking for available Python skills. I don't write in Python. I'm an old C++ developer.

(16:12):
I don't know how to write in Python, but AI does for me. So it's going to generate the Monte Carlo now and it'll generate the Python code. Why is this valuable to you? It's a time saver. What you want to be able to do is leverage artificial intelligence in a way that's going to save you a lot of time and not burden your team with one-off questions that you may get from a client. For example, with that client that called up and said, "I have a concentrated position in Oracle," you don't know if they're going to sell that position in Oracle. You don't know what they're going to do with it. What you do know is you need to be able to answer the question for them. So an ability to basically do a really quick analysis of Oracle that's factually correct is going to save you an enormous amount of time.

(16:58):
Notice here, this is now generating the Python code behind me for the Monte Carlo simulation.

(17:07):
Once this Python code is completed, it will actually run the Monte Carlo simulation. And then what we're going to do is I'm going to show you how to take the output of that, make it a much stronger financial plan, and then we're going to actually generate the financial plan using Gamma AI. This is now running in the background. We'll let that go for a few minutes. Basically, the next step I'm going to do is ask it to get to a CFP level plan. In the interest of time, let me show you what this does. If I actually said using my first prompt, just so you know, there's my first prompt right up here, hasn't changed. If I ask it now to do me a favor and tell me what would it take to get this to a CFP level?

(17:50):
Now notice, I didn't have all the information. Like many times when you interview a client, you're not going to have all 40 points of information that you need for a basic CFP plan. You're probably not going to have the 60 points you need for a comprehensive CFP plan. What you're trying to do is get something together that's a little bit faster and then determine what else am I missing. So here, what I've done is I said, what would it take to get to a full CFP level? And it says, "This is what the analysis includes. Here's my strong points. Here's what you're missing from a full CFP level plan." I'm missing a lot from a full CFP level plan. Then it does a comparison for me. It'll tell me this is how your prompt compared to what the CFP really needs.

(18:34):
This is what I would need to generate a full CFP plan. The discovery meeting, two to three hours. That's a lot. I don't know if I'd spend two to three hours on this couple, but certainly I'd want to be able to gather more of the information. Now, if I used AI to generate the form and I've met with the couple and I've said, "I really would love to generate a full financial plan for you, but right now we're still missing a little bit of data. So I'd love for you to go to my website and go to this form," the form that we created using artificial intelligence, "and I'd like you to fill that form out so that I have the rest of the data that we need to generate a full financial plan." That would save you a lot of time on that discovery meeting.

(19:17):
So here, if we run down, this would be the comprehensive prompt that you need for a full financial plan. To give you an idea of what that prompt looks like, here's what we've got. Now notice this prompt for a full CFP level financial plan is extensive, as you would imagine, with the rigor that you'd put into a full CFP plan. This will generate the prompt for the full CFP plan. Now, let's say, I'm probably not going to get a lot more data from you. Notice this prompt's really big. Prompt engineering matters. For anyone who thought that there's no jobs for prompt engineering, there isn't, but prompt engineering really does matter. If you go through this entire prompt and you say, "I do not have all the information that I need," you still can run this by saying, "Hey, let me generate a CFP level financial plan for Harry and Ginny." And this will go off and it will make assumptions on the remaining information that you are missing from the financial plan.

(20:24):
Those assumptions could be capital market assumptions, they could be investment mix assumptions, right? So on your basic 60/40 type of portfolio, but it will fill in the assumptive data that it's missing. Once it's filled that in, you can now download what the comprehensive CFP plan looks like. So if I download that and show it to you really quick, here's the comprehensive financial plan that it generated for them.

(20:55):
So here's your executive summary. Here's your current financial position. Let's make that a bit bigger for everyone. Here's your comprehensive scenario analysis: If you work longer, delay Social Security, combined, early retirement, reduce your spending. Here's your scenario results. Here's your portfolio distribution, your Social Security claiming strategy, this was all generated from that one prompt. Now let's say as an example, this is all great, but they're probably not going to want to look at all that documentation. I could download all that. I could also say, "Do me a favor. I'd like you to generate this financial plan in a format that I can send to Gamma." Gamma is a presentation software, another set of artificial intelligence that will generate what that output could look like. So what I've done here is I asked it to give me a format that I could send to Gamma.

(21:56):
Let me just grab that format really quick. This is the Potter format for Gamma. This is actually going to be done in markdown language. Markdown language is just a set of notation that you could utilize. Now you'll notice this is a pretty extensive financial plan. Let's say for a moment that I wanted to put this into Gamma and I wanted Gamma to go generate for me this financial plan. So I'm going to say I want it to generate a presentation. I'm going to throw this in here. That's the financial plan. Let me edit that prompt. Let me do this for the optional instructions: Generate the first 20 slides using the Oasis Group template.

(23:02):
So now Gamma is going to take that output from the first artificial intelligence, the financial plan that we created, and it is now going to create the presentation for the client using artificial intelligence. I asked it to use the Oasis Group template. For any of you that have looked at any of the research that I've done in the Oasis Group before, you're going to notice a couple of things about the Oasis Group. Number one, we use a mountain motif. Notice the mountain in the background. Notice the colors are all the Oasis Group blue colors. I told it the hex code colors that we wanted. I told it that the Oasis Group likes a mountain motif and it is now generating the entire presentation using my template essentially. Here's the opportunities that they've got. Here's their net worth statement. Keep the mountains in the background.

(23:57):
Here's their savings discipline. It's generating all of this. It's generating all of the graphs using the real data that came out of the first AI prompt to generate everything for this financial plan presentation. If you are trying to run a financial plan for the 20-somethings or the 30-somethings that are still in the wealth accumulation phase, they don't have a lot of complication associated with their financial planning at this point. This can save you an incredible amount of time in just generating the initial thoughts on the plan. Do I think you should put this directly in front of your client? Absolutely not. The reason I say that is you need a human in the loop. You have to have a human in the loop to avoid certain things called AI biases and hallucinations. A bias is when AI has a heterogeneous data set, it can create a bias.

(24:54):
I'll give you a really good example. If you fed all the financial plans of your existing clients into an artificial intelligence engine and all of your clients are in their 60s, and you said, "Can you generate a financial plan for me for someone in their 40s?" It's going to skew that towards your existing client base in their 60s. You have created an AI bias. Hallucinations. If you don't give AI a strong prompt, artificial intelligence will make up the gaps. Try to fill in those gaps and it will make assumptions based on those gaps. Sometimes those assumptions are wrong or they are fabricated. Here's a really great example: There's a court case right now. Anthropic, the makers of Claude, are being sued by the Record Association of America. The Record Association is basically suing them saying they've got all their lyrics.

(25:50):
The lyrics have been out everywhere forever. That's not what's interesting. What's interesting is the attorney for Anthropic used Claude to generate part of their statement back to the court. Claude hallucinated a court case. The attorney did not keep a human in the loop and provided that to the court. You've now provided false information to the court. That can get the case thrown out. So you need to be really careful when you're dealing with artificial intelligence to avoid hallucinations. But this output that we've created here is pretty good. There's Harry and Ginny. I could replace that with their pictures if I wanted to. They should be in cloaks, quite frankly, if anyone gets the reference. But here's the scenarios that it tested. This is the first 20 pages. This prompt, I will admit, because there's 80 points in a full CFP plan, it generates 80 slides.

(26:50):
A slide for each one of the points. Now, if I had provided a template that said my financial planning template are these 25 slides, and I need you to summarize the data into these 25 slides, I could easily load that just like I loaded in my regular prompt into the generative AI. I could tell it to limit this to 25 slides. It'll create the output for Gamma in 25 slides. I cut and paste it just like I showed you, cut, paste into Gamma and bingo, bango, you've got your presentation for the client. And it's a pretty decent presentation. Now I only asked you to do this slide, so it stopped here, but it's not bad. I've got about a minute left. I just want to give you a couple of quick points. We do an enormous amount of research at the Oasis Group.

(27:41):
We've done research on AI prospecting tools. We've done research on wealth platforms. We've done research on AI note takers. The way that we do our research is we ask the providers to provide us with what capabilities they have. On the AI prospecting tools, of 162 questions we gave them across five different levels, we rated them on a rubric. Then we did demonstrations to determine what they said is what we saw. And then we talked to some of their clients to determine what their experience was. We produced this research on a fairly regular basis. Those research reports are absolutely free. Go to my website, you can download them. We also produce an AI wealth technology map every month. There's 87 firms on the map right now, and that's growing every month. That you can find on my LinkedIn and our website every month. So I want to thank you all.

(28:43):
The one last piece that I'll say is if you have any questions on anything that I showed you, if you want those prompts, if you have questions on the technology that we leveraged, by all means, please reach out to me. Those are the QR codes. One is for my LinkedIn, one is for my website. You can get us on the website. If you liked what I said today, my name is John O'Connell. If you didn't, my name is Eric Stratton, Rush Chairman, and I was damn glad to meet you. Thank you, everyone.