There's no debate that agentic AI has the potential to profoundly impact how financial advisors and wealth managers acquire, retain and deepen engagement with clients. What's often overlooked in the what-if scenario planning of the agentic AI era? Building an AI tech stack that works for financial advisors and their clients. In this panel discussion, industry experts will talk about why it's critical to align technology with specific workflows, compliance requirements and personalization objectives. As the talk on use cases heats up, here's what our panelists say that industry players need to consider in building their AI tech stack:
· Identify business goals and use cases first, including client engagement (personalized recommendations, behavioral insights, automated reporting), advisor productivity (intelligent CRM, smart task prioritization, etc.), compliance and risk (KYC/AML monitoring and trade surveillance), and portfolio management (dynamic rebalancing, AI-driven allocation, tax optimization).
· Core components of the AI tech stack, including the data layer (data sources, integration tools, warehouse or lake/lakehouse), intelligence layer (machine learning/AI models, agentic AI and model ops), and application layer (client-facing tools, advisor dashboards, conversational interfaces).
· Security, compliance and ethics, including data privacy, regulatory compliance (SEC/FINRA, GDPR/CCPA adherence in data handling), explainability and bias mitigation.
· Integration and interoperability, including open APIs to connect with wealth platforms, workflow compatibility, and third-party plug-ins.
· Governance and monitoring, including model governance, performance monitoring and human-in-the-loop needs for higher risk recommendations or compliance review.
· Testing and feedback loops, including advisor feedback, client insights on usage, satisfaction and conversion improvements, and testing for UI/UX and model performance pre-rollout.
· Scalability and future-proofing your tech stack with a cloud-native architecture, modular stack and the ability to plug into large language models (LLMs) and multi-agent frameworks as the field evolves.
Transcription:
Rob Burgess (00:14):
Hello, welcome to the third and final panel of today's virtual summit: Building an AI Tech Stack That Works for You and Your Clients. My name is Rob Burgess. I'm a Reporter here at Financial Planning. Today I'm joined by five distinguished panelists who are well-qualified to speak on this topic: Parker Ence, Co-Founder and CEO of Jump Advisor AI; Doug Fritz, Co-Founder and Executive Chairman of F2 Strategy; Dr. Sindhu Joseph, Founder and CEO of Cognicor; Oleg Tishkevich, CEO of Invent; and Timothy Welsh, President of Nexus Strategy. Thank you all for joining us here today. Thank you.
(00:53):
I'll address these first few questions to each of you individually, but please everyone else feel free to jump in as well. Also, if there are any questions that you have there at home listening, please send those our way, and we can ask those as time allows. So we'll get right into it. Oleg, we were talking a little bit before. You had mentioned how quickly Agentic AI is developing even in just these past few months. What does that look like on the ground? What is the cause of this acceleration, and why is it important for advisors to understand this?
Oleg Tishkevich (01:29):
Thanks, Robert. So, as I was saying before, there are significant enhancements happening just in the AI landscape recently, and it is growing with an escalating speed. We're all familiar with more of a helper type of a function of AI where you can ask a question, it comes back with an answer, or you're trying to create an email, you can paste what you want to write, and it'll rewrite it a little bit better. That's more in an assistant role that AI has been playing. I think what's happening lately is Agentic AI, which is really a different type of usage of AI. Instead of just providing a back-and-forth or question-answer type of a chat experience with AI, you can create these autonomous agents, AI agents. You can basically define who they are, almost give them a job description, if you will, and explain very thoroughly what they're supposed to do, and they become kind of your workers, if you are, your worker bees, if you will, your virtual worker bees, right?
(02:45):
The ability to do that today is not just have individual AI agents sitting there and doing their individual jobs, but you could actually orchestrate these AI agents where one could be the AI agent manager or a checker, if you will, and others are going to be doing certain types of tasks or perform certain types of functions. I think I read somewhere there were a couple of companies that started completely AI Agentic, AI-based startup, meaning there are no people; it's just the CEO is an AI agent, head of marketing is an AI agent. So it goes, of course, to significant extremes, but there are use cases like that that are happening. What we are seeing with Invent is we are really restructuring our company based on that trend. With Agentic AI, there are significant enhancements that have been done in software development's ability to leverage AI for the software development lifecycle.
(03:58):
The good news is you can, I think, build apps a lot faster. I live in Seattle, so there's a lot of talk today these days where big major companies like Google and Microsoft and AWS, all Facebook, all in Seattle, they're all restructuring their companies to use Agentic AI more and more in software development. It doesn't mean just coding. It could be creating specifications. An analyst could sit down and set up an AI agent that would be responsible for certain tasks. You'd still need human oversight and understanding, and you actually need somebody who's very experienced to be able to create those agents so they don't go off the rails, so to speak, right? You want to create very clear instructions when you're creating those agents and obviously tune it up so that they produce the right output. But that's really changing significantly with some of the latest releases. Microsoft just released a new coding agent for Agentic AI for code development in GitHub that's making some major strides. You literally could have an application built by AI agents start to finish, which is pretty remarkable. So these things naturally are going to be affecting the wealth management industry, and the ability to achieve something or build something or get something done is now escalated to a significantly higher speed, if you will.
Rob Burgess (05:47):
With all this development, I have to imagine that the regulation is kind of lagging behind. I can't imagine it's possibly keeping up with as fast as this developing. Doug, I know that your outfit there has done some research on that. Specifically, are regulators able to keep up with this pace of development?
Doug Fritz (06:08):
I wouldn't say keep up. I think F2—and just for the audience, F2 is the largest wealth tech consulting firm in the country—we do a lot of research on trends in this space, and two in the last two and a half years on the state of AI. To Oleg's point, this is taking off dramatically quickly, and the benefits of this are outstanding. That's not bluster. Everyday firms everyone's heard of are using AI now that's saving 90 minutes or more per call with clients on operational time. From a compliance standpoint, it's actually more about what regulations exist in the market that we're going to leverage, and then we're talking about regulations from 20, 30 years ago. In the SEC/FINRA regulated entities, this is going to be things like just making sure that the asset allocation and trades you're making for clients are repeatable, justifiable, and you can validate them.
(07:07):
Client communications is still going to be a third rail, not to have AI write things for clients that are not checked by a human. We're going to see a lot of firms avoiding natural language generation direct to client without having a second set of eyeballs of humans checking on it. It's great for compliance checking of things that we're writing to clients; that's been around for a decade or more. In the bank trust space, the rules from the OCC are a little bit different. Their concerns are a little different; it's more about process and procedure. So we're seeing broker-dealers and RIAs faster adoption of AI than the bank trust space. We're probably going to continue to see that the next five to 10 years, with the OCC being really slow to produce any kind of guidance that it's okay to start using AI for internal operations.
Rob Burgess (07:57):
Right. And Parker, you specifically, I think you have kind of a unique perspective on the provider side, given that your company is one that I write about a lot, and those note takers have been—I've seen a huge uptake from advisors on that. How has the regulatory environment influenced your product? What are you doing on your end to help advisors stay compliant as they use this technology?
Parker Ence (08:24):
Yeah, thanks for having me here today. It's so great to be with everybody. It's been really interesting for us. We are now working with almost 14,000 advisors. In order to do that, we've had to work with some of the largest broker-dealers in the country, some of the largest RIAs. I think what we've seen is because the SEC and FINRA haven't come out and given really prescriptive directions on "this is the right way or this is the wrong way," each firm is sort of left to their own devices to figure out how to interpret the existing regulations with new technology, and they're having to make their best judgment, which I actually think is good because you don't really want regulators to get too far ahead of technology; it causes problems. In fact, right now there are many states that are trying to pass AI regulations, and if we end up with 50 different AI regulations, it would really slow down the development of technology.
(09:24):
So I think that's actually okay, but in the meantime, we're left to interpret the existing regulations and make sure that we don't go afoul of those as we deploy new technology. As just one example, if you think about books and records and how it relates to archiving specific types of communication with your clients. We all know that we have to archive emails for a long time and then have those, there actually is currently not consensus in our category. When you capture conversation data—a lot of what we do is we capture conversation data, and we then summarize that and do other analysis on it and turn that into things like recap emails, compliance meeting notes, tasks, et cetera—if you're capturing data, suddenly we've got things like federal and state wiretapping laws. We've got books and records rules about what do we need to archive videos and audio and transcriptions the same way that we would archive an email, or can it be treated like a page of handwritten notes that can be disposed of?
(10:34):
We've seen firms take actually very different positions on this. Some have said, "We think that if there's a video or an audio, we have to archive that the same way as an email. Therefore, we either need to push all of that into our archiving system, or we just don't want that," right? "We don't want to deal with that." Others have said, "Well, we don't think we have to archive it, but we also don't want it discoverable for a long time if there's a claim. So we want to kind of auto-purge that after a certain amount of time." Others have said, "Hey, we actually like having this data around. We feel like it actually protects our advisors. We think they're doing the right thing." This is a roundabout way of coming back to your question, which is thinking about building product for all of those different scenarios to make all those different chief compliance officers happy. We kind of came in at the beginning saying, "Yeah, they're going to be different takes on this." I think if you look at products that have been specifically developed for our industry, you're going to see a much wider range of compliance options and configurations, which is good because it means that we can speed up deployment, we can get practical tools in the hands of advisors, and still make the chief compliance officer happy, or at least they'll at least sign off on it.
Rob Burgess (11:46):
Right. Shifting the conversation just a little bit, Tim, we had spoken earlier, and you had mentioned that you were particularly watching the developments in the marketing technology space. How does Agentic AI affect aspects like lead generation and scoring and business development, things like that?
Timothy Welsh (12:05):
Yeah, I'll kind of go back to Oleg's opening remarks about being an assistant. I think the two use cases in our space are meetings and marketing. Meetings definitely having a big use case, and marketing also. If you think about, there are really four aspects I think that advisors are using AI in marketing. The first one is in just pure content development. If I'm going to send out a blog post or an email or some sort of content, AI is fantastic, maybe that first draft. Obviously, to Doug's point, you need to have eyeballs on it, the human involved, to make sure it is correct. Maybe a very tricky Roth IRA conversion tax law post would definitely warrant some significant oversight just because we know sometimes the AI makes stuff up or maybe pulls from the wrong place where there's no source. So there's a lot of, again, back to the compliance rules.
(13:00):
So I think one, that content development saves advisors tons of time. There's nothing like looking at a blank screen and saying, "What the hell am I going to write about?" when all you have to do is just type in a one-sentence prompt, and boom, comes back some amazing first cut, and then you could iterate on that. That's a fantastic case for the marketing content development. I think what people adopted to the second one was what you mentioned there, which is all about leads, the funnel management. Let's start at the top of the funnel. There are some really spooky databases out there that I'd love Parker's and all interpretation of this information that's out there about us. I think they've got 300 million Americans and all 3 billion data points of what they know about us, and being able to harvest that and being very prescriptive, such as, "Give me all the physicians in Northern California who are thinking about selling their practice."
(13:54):
Can you then text them a message saying, "Hey, we're wealth advisors, and we specialize in this. Would you like to set up an appointment?" That is really precise. I think really what the platforms like Finny's of the world are doing is to help advisors source those leads, but that's just filling the top of the funnel. So then the next level, I think, is really all about optimizing those leads and scoring them. Will they convert? Tons of platforms there like Catchlight's of the world who can then score those leads for you and say, "Hey, you have a hundred names on this list, just call three of 'em because the other 97 are never going to convert." Think about how awesome that is in terms of optimizing your sales funnel and your sales pipeline. The third component of this marketing area, actually the fourth one I mentioned, was about advisor matching.
(14:40):
We're starting to see some really cool platforms who can actually then take the data about an advisor, what their specialty is, what they care about, and then matching them with leads that come in and say, "Hey, they like big dogs and RVs, too. You happen to have a big dog and you drive an RV, so you might share a lot of personal aspects that fit together." So when you think about the marketing stacks, sort of the genesis of what this conversation is all about, if you want to think about what you want to fill in the gaps there in terms of the business development, the growth, the marketing, I think those four areas you want to look at, particularly if you're the enterprise large advisor matching, when you've got tens of thousands of leads coming in and you've got a hundred advisors, how do you vector those leads to the right person?
(15:21):
The AI is amazing at doing that. As Oleg said, we could turn these into bots and the agents and really turbocharge what advisors can do on the growth side. So the sky's the limit here. I think we're still—we haven't even started the first inning. We're still in the warmups before catching fly balls out in the outfield before we even start the game. To me, that's very impressive, and I think it's a greenfield. Obviously, the compliance overlay and the fantastic questions you're asking, Rob, in terms of how does this all intersect and come together. So lots to talk about. I'm sure we could spend an entire hour on just that, but those would be sort of my high-level ways to think about what's in your stack and what you want to build going forward.
Rob Burgess (16:00):
Definitely. And Sindhu, this kind of speaks to something that you wanted to talk about, I think, with all these various uses that people have for this. Now I know you're concerned about tech stack fragmentation and especially when it comes to these new Agentic AI tools. Can you go a little bit more in depth about what you're looking at, and how can people avoid that fragmentation if possible?
Dr. Sindhu Joseph (16:28):
Yeah, first of all, very excited to be here. I think we are no longer debating whether AI is significant in the wealth industry or any industry for that matter, but how effectively we can use AI to empower advisors and clients. One of the big silent barriers, I would say, is the tech fragmentation that exists in the industry. Advisors today use eight to 10 tools, switch between the planning, custodian platforms, the portfolio management, market insights, and compliance tools, just each one holding one piece of the puzzle, but none the entire picture. So this is where the industry is. The paradox is as we are excited to adopt more and more AI tools, we are kind of multiplying the fragmentation. We have the AI fragmentation. This would start to, if you look ahead in five to 10 years, advisors, instead of switching between eight to 10 platforms, they would be switching between 15 to 20 platforms.
(17:38):
That's not something anybody wants. So how can we create and make use of AI in a way that is avoiding the fragmentation or unifying the fragmentation? I think the solution is creating an intelligence layer that uses the existing technology tools, that uses the data from different disparate sources, but combine that to provide a holistic level of intelligent insights as well as intelligent action. This is where we could really drive advisor productivity and also provide that highly personalized personalization for client households where you are looking at a 360 view of the household rather than kind of piecemeal offerings. So I think we should look at that intelligence layer above all of the current tech stack, current fragmented tech stack, and this can really help in the long term providing high efficiency as well as putting it the right way, a system of intelligence about the system of records and between the system of engagement. That's how we would look at the AI tools and AI layering. That's what Cognicor is looking at. We are a copilot platform that kind of uses data from all of these sources and generate intelligent insights as well as actions.
Rob Burgess (19:12):
Great. And I do a series on that same note. I do a series here at Financial Planning called "Show Me Your Stack," and I talk to advisors about how the different pieces of their tech stack work together. That is one of the main concerns is the integrations or lack of integrations, and I'll ask this question to anyone who wants to take it on, but how can people make sure that these different pieces are talking to each other properly as people are adding all these new tools to their practices?
Oleg Tishkevich (19:45):
Well, if you don't mind, I'll take this one because that's kind of what Invent does from the beginning. That's how I started the company. I did want to solve this little problem called integration in the industry, and it's definitely way more complex than I had anticipated when I started the business, but I think there are a number of different ways and paths that we need to take in order to solve it. Definitely, an intelligence layer is a great thing to have, but I think it all starts with the data. The challenge is that you have all kinds of intelligence in the world, but if you don't have a golden source of truth, like a single source of truth for all your data, that golden record, it's very difficult for any AI agent or intelligence layer to figure out. What if I have my—I mean, the simplest use case—if my client is in those eight or 10 or 15 different systems, and that client has different address information and the first name and the last name and a bunch of other stuff, sure, AI can figure out what's probabilistically possible, but at the end of the day, if you don't have, it's probability. When you deal with probability, those of you that—I'm sure advisors—spend a lot of time with financial planning and explain to clients what the probability is.
(21:07):
On the flip side, you don't want to be in a situation when your CFO comes in and says, "Well, our revenue is—there's a 90% chance that our revenue is about 1.3 million a year." That's not something you want to have. You want to have a precise output, 100% correct data. In order to achieve that with any kind of AI overlay, you need to have AI-ready data. That's another big thing that's happening right now because, sure, AI can source information and tell you where the data is coming from, and then it's up to you to Tim's point, like, "Hey, you need to still have a human to look at it and say, 'Yeah, I guess you got it in the right spot. It kind of looks right. Let's send it to the client.'" But if you want to be sure a hundred percent, that's a whole different situation. You need to create essentially data that's specifically designed for AI consumption, that is AI-ready, and it's already cleaned up so that all that information is delivered 100% correct.
Parker Ence (22:19):
I think that's the ideal situation: to have a beautiful data warehouse or data lake where all of your data's in one place and all of your applications are interacting with that. I think there's a lot of progress there, and in the meantime, there are still a bunch of applications that aren't ready to do that. So integration becomes really important, and we know that advisors care a ton about this. We actually just this week did a webinar to announce Jump's integration with eMoney, and the use case there is kind of fun. We basically extract all of the financial planning data that comes up in a conversation, and then we organize it, and then we put it into the same format that the eMoney data schema is designed for. Then you can kind of update all the financial planning data with one click, and there's a human in the loop there.
(23:11):
Doug mentioned how important that is earlier, but there's a human in the loop there so that the advisor can check that data before it gets loaded in. But I bring this up because this was—and by the way, we have this same integration with Right Capital—but this particular webinar with eMoney was by far our most attended webinar that we've ever done at Jump, with almost a thousand people that signed up for this webinar. I think this just speaks to how much pain there is with moving between different systems. If you can get a really nice integration, whether it's a direct integration or whether it's via your own data lake, yeah, it just saves a massive amount of brain damage to try to move across systems.
Doug Fritz (23:55):
I can offer just a little bit here on top of what Oleg and Parker mentioned. I think it's a big industry. First off, we've got folks I'm sure on this call are listening in that are from massive UBS, LPL, Cetera kind of firms, and we've got smaller firms as well. If you are a very large firm, the risk of federating two dozen different AI tools that don't talk to one another and potentially create data themselves that then may not actually tie back to some of the AI-ready data that Oleg was talking about is a big risk. Even for smaller firms down the road, like five, six years, there probably is going to be a next new thing. You want to be with a partner that's innovating and integrating. You may not want to sit around and rely heavily on one vendor that is not going to integrate with your CRM, your planning tool, your portfolio construction tool, trading tool, et cetera.
(24:53):
So it's tough, just noticing that these are difficult decisions for anybody to make. I think right now, we're seeing in our client bases, we're weighing the operational, human capital, client experience, organic growth, marketing, branding benefits of some of these new technologies versus maybe the hangover you might feel in 12 to 36 months down the road of going with a vendor that's early stage or may not integrate or build too much. In other words, there isn't a perfect answer to this question. I think there's the weighing of immediate benefit of these tools versus long-term complexity. The firms we work with generally are leading into some of these functionalities. They're writing contracts that are maybe 24 months long, not super, super long, just so that they're able to try a little bit for a little while before they go all in and dedicate all their resources and move from VCR back to Betamax or something like that and then realize that they bet the wrong horse long term.
(25:54):
So if you're not fielding the phone or not using these tools right now, I think the type of technology that Jump deploys is by far the easiest to adopt and sort of low-bar entry of just immediate benefit to the organization, and then layering in as you get better and better at AI and get more comfortable with it over time, layering in other things. As you layer them in, the universe changes, Oleg. The universe changes every two months in terms of what technologies are coming in and how these things are going to impact our industry. So go now, don't wait.
Rob Burgess (26:31):
Yeah, go ahead. Sorry, go ahead. Go ahead. No, no, no, you're fine. Joke.
Timothy Welsh (26:36):
So Doug, what's a VCR?
Doug Fritz (26:39):
Exactly. I'm aging myself by even mentioning VCR.
Oleg Tishkevich (26:44):
Yeah, I thought about the terms a few times, maybe I'd say AI related, so I'm trying to sort my brain what that was about. Thank you, Tim. One thing I wanted to add from the universe perspective, the solution to this really from a technical perspective, what's happening in other industries is really the move to the next model, which is called the data ecosystem model. That is where you have your data centralized in one massive data lake, data warehouse with AI-ready data, and applications are actually—and agents—are built on top of that data. So the data doesn't move. Integration problem goes away completely. There are a number of firms that are actually building right now those types of applications that are essentially future-proof from a standpoint: they can run independently, but because they can be simply dropped into your data lake with a click of a button and then operate on all the workflows, user experience, but you actually get to keep all your data in one place, and any other application, they want to request that data, it's already in your data lake.
(27:58):
That is the future state. There are not a lot of firms that are doing it today, but I think those that are going that route are going to be in a position where they're going to be way more adaptable to any of those changes. So to Doug's point, you don't need to be signing contracts for two years for your platform and then figure out how you need to redo it next year. If you've got the right infrastructure, the right foundation for your business, anything could change, and then the right companies would be able to be simply a plug and play, just like you have on your iPhone. Imagine that same experience essentially in your enterprise where you're no longer like, "Oh, does anybody ask a question, what's your tech stack on your iPhone, Robert?" Do you ever ask that question to anybody? You don't ask that question to them easy, right? Everybody's got some apps, but we don't even think about it. That's the way it should be in our business. This is the way where we need to go as an industry, I believe. I think that if we start realizing that, we can provide significant benefits to our clients, we can service more clients more efficiently, we can be more efficient as an RIA, we can be more efficient as a business themselves and servicing more clients. So just my two cents.
Rob Burgess (29:23):
Right. And you guys have talked so much about what I was just about to ask about, which is kind of owning your own data and these data warehouses, data lakes, data lake houses. I know that not everyone out there is going to be somebody that owns their own data, but we've talked about the benefits are obvious, so I was wondering if you all could speak a little bit more about that process. It's not cheap, it's not easy, but the benefits, like we've been saying, are pretty great, right?
Oleg Tishkevich (29:54):
Well, Doug, I don't know if you want to say—I'm happy to continue on that topic, but
Doug Fritz (30:02):
Yeah, Oleg's entire company does a great job of exactly this, and it's really valuable. We work together on clients, and I've seen it be very, very valuable to have your organization data organized. So two things back to my previous point: size matters. If you are a $500 million emerging RIA, thinking about your data is important about where your data is, how you have access to it, how you can overlay analytics on top of it. That's important. Most likely, it's going to be within your CRM like Salesforce. That's great. When you get into the $2 billion, $10 billion range, that's when we really start seeing a need to bring data more in-house so that you can use it across all your different platforms. It is, as you said, not inexpensive. Five years ago, there weren't really Invents out there in the world to do this for you. You had to roll your own.
(30:59):
It was much more expensive. The barrier to doing this is actually much lower now. When you get to owning your own data, a couple of things happen. One, operationally, a lot of the headaches of quality assurance before you talk to clients, your data, that goes away. Your data is correct when you're pulling it out of any different technology because all of your surrounding tools are running off the same set of data. To capture something like a name change, address change, best objective change in one system, it can feed back into that database, and it's updated everywhere. This saves operational time; it saves the embarrassment of coming to your clients with wrong data and sitting at client meetings and saying, "Hey, I told you about that new 529 account that I set up for my grandkid. Why is that on here?" The other thing that it does for you besides operational efficiency and the payback internally is as you think about business valuation, and we do a lot of work with confirmatory diligence and PE firms looking at buying advisor businesses, especially in that five to 10, five to 20 billion range.
(32:00):
Those that have architected and organized their data well are actually prepped to be the 30 to a hundred billion wealth firms in the future. They get a higher valuation, and if they don't, they're sort of a federation of different tools, and they've got a great brand and great clients and great growth, but honestly, they're not really ready to go to 20 billion, 30 billion, 50 billion. The PE firms, we got to tell 'em this like, "Hey, you got to figure factor in not just the cost, but the time of getting your data right before you can start really going through proper M&A and bringing businesses in." So there are two huge benefits of having data organized. They're probably more than that, but those are the ones we see that probably aren't talked about enough.
Dr. Sindhu Joseph (32:41):
I just want to provide an alternative view. I know that data lake and data warehousing is a beautiful solution, but we are kind of discounting the power of AI. I have been working on AI for around 15, 20 years now, and I see that AI has a huge—it's like running a Ferrari, and people still kind of think that, "Okay, I am using Ferrari, but driving around the school playground or around." So I think today AI can read from unstructured to semi-structured to structured data. Using a schema can go and pull data from planning software, from portfolio systems, from custodian platforms. So while probably the audience is thinking about, "I cannot look at a data lake or data warehouse solution," the intelligence layer, the AI layer can, I think, has the potential to do this in real time. So it is kind of how do you enable technology that exists today versus how do you disrupt the technology that today and build a beautiful new world? I think you can be that enabler to augment on the current technology stack and not touch anything that is working today, but still deploy AI in a wonderful manner. So I would encourage people to kind of think of whoever cannot afford or does not like the fact of data lake and data warehousing—taking it's a big project, multiple years, often fails in many cases—there are alternatives because AI is very, very powerful today.
Parker Ence (34:47):
I want to build on that. I actually think this is one of the most exciting areas that AI can help us with. I think a lot of the impact of AI so far has been around productivity, saving time, helping advisors be more prepared or helping them execute after client engagements, helping them identify next actions. One of the things that we see that's been really fun is the point about generative AI. This last wave of AI, the magic of generative AI, is it can deal with unstructured, semi-structured, and structured data, just like you just said, and this unlocks all these very cool opportunities to not only automate what previously was impossible to automate, but also to get insight into areas that were completely black boxes in the past and be able to actually gain more success as a firm and as a team.
(35:43):
Just one example, we've got an insights program here at Jump where we'll take an anonymized version of some of the conversation data that we've processed, and we've processed over a million conversations at this point, so there's a lot of data there that was previously not available. You can start to ask questions about what's actually working, what's actually on the minds of our clients. One kind of fun example is if a client brings up tariffs or they bring up "the big beautiful bill," they're three times more likely to tell the advisor they want to deviate from their predetermined plan and move things into cash or move things into a money market. That's probably a good thing to know if you're walking into a client meeting that they might bring something like that up. Another example is we've looked at some of these prospect conversations where the advisor's goal is to turn a prospect into a client.
(36:39):
There are different behaviors that an advisor can do in that initial meeting, and those behaviors have different probabilities of success in completing that conversion. One simple example is if an advisor brings up some kind of low-hanging fruit, some kind of early win, maybe pointing out, "Hey, we can actually help you save a thousand dollars on your taxes this year." If they can bring a very concrete early win, that vastly improves the probability that that client becomes—or that prospect becomes—a client. I think there's just this massive opportunity to take all this data that's just sort of been floating around in the ether and actually analyze it with AI in a way that helps every firm do better at what they're doing and grow faster.
Oleg Tishkevich (37:21):
Yeah, 100% agree with Parker. AI is really good at understanding language-related stuff and being able to read unstructured data, but the reality is our business is all about structured data, except for the documents. I mean, that's the only thing that's technically unstructured or Parker, to your point, any conversations you're recording and through Zoom and then transcribing. Everything else is all structured data, and AI is not very good. It's getting different structures of structured data and actually pulling it together in a way that humans need to see it. You got to have a structure to do it. To your point, Sindhu, I got to argue because, yeah, maybe two years ago it would take a two-year time to be able to build a data lake. We bring in a data lake, stand up a data lake for our clients, for the firms within seven days with all the feeds that they need into those data lakes. Know what happens next? Once we actually bring it in, every single firm—
(38:25):
There's not been a single firm where we don't find a lot of problems with data. If you let AI loose on all that data, it's going to lie to the client; it's going to provide wrong information because the data itself is wrong. We're finding it in every single case. There's not been a single client that we bring in with all the feeds. Everything's beautifully connected, but then there's like, "Oh, there are rep codes that are sitting up there and no assignment to it." There are accounts that are actually lingering in. They're not live accounts, but they're showing up as live accounts in the system. There are a lot of problems with the underlying data that firms are not even aware of because, "Oh, I'm using this tool for performance reporting. I'm using this for CRM. I assume that everything is right." But I tell you, we work with hundreds of firms, and we've not hit a single one where we installed the data lake within a week, and then we can just turn it on, and everything works perfectly.
(39:21):
There are always issues with the underlying data, and that's the state of the industry. That's a true state of the industry. So you definitely want to use the AI, and the way you guys described it is very transformational. It's extremely helpful, but unless you clean your data, unless you get to the state where it's reliable and from all the sources, you are running into a significant liability for your firm because what it produces either needs to have a human reviewing it very thoroughly, or you might have a situation where you may be saying something wrong or you may be providing wrong information or wrong assumptions based on the data that you're reading, essentially. That's just what we're seeing out there.
Dr. Sindhu Joseph (40:13):
I don't know how far you want to take this debate, but one of the interesting point around regulation that we mentioned previously, maybe one point that I want to add to the regulation is while the regulators haven't focused on putting out a set of concrete rules, one thing they might be looking at is providing evidence or having a trace of whatever insights or recommendations or any actions that AI has taken, whether there is a trace that is available. AI systems that are structurally strong need to provide this trail of evidence for every decision that has been taken. Some of the problems that Oleg mentioned couldn't go away with this providing evidence, but maybe we can take this discussion offline, but I just wanted to put it out there.
Oleg Tishkevich (41:15):
No, evidence is definitely important, but the fact that you provided evidence that you bring the data from a place where the data is wrong, that doesn't solve the problem. It just provides a record that, "Hey, we just served up wrong information to the client." At the end of the day, an advisor looks at it like, "Why is your system wrong?" No, your system is not wrong; it's just the underlying data is wrong, but nobody's realizing that when they see this information. We see it day in and day out. We have multiple AI agents and applications working on Invent, using Invent AI-ready data. They're having no issues. So all this stuff beautifully works, but if you were to connect these things directly to all these different sources, I guarantee you they're going to be issues, and you will show where the information came from.
(42:04):
But humans, we don't have time to check every single source and where it came from. "I have a client meeting in 15 minutes, just prepare all this stuff." I need to make sure the stuff is right, and if I can't a hundred percent be confident that the information that looks right because AI is very believable—it looks right, it's going to find its best way to provide best information-based outcome with a great trace record—how do I know it's right a hundred percent? That's where the probability thing doesn't really fit when we're in a regulated environment like financial services, in my opinion. Again, we can take this offline, but I love this conversation. It's a great talk.
Rob Burgess (42:49):
Here I was worried we wouldn't have enough to talk about to fill the time, but we've got two minutes left, so I'll throw out one question we did receive from an audience member. They said they agree with Oleg that we need to avoid a garbage in, garbage out situation. To that point, for a solo practitioner without the resources to properly vet the guardrails in place, how should one approach the adoption of agents while staying compliant?
Doug Fritz (43:18):
I can offer a solution on this one. I think solo practitioners and firms that are in the earliest stages of growth should look to your custodian, be it the traditional RIA custodian, so your broker-dealer. The amount of AI and Agentic AI, especially on operation support and then also client analytics that are going to be coming out of these vendors the next 12 months is significant. I wouldn't say they're bleeding edge. They didn't lead with that two years ago, but they understand thoroughly that having Agentic AI from an operations and client analytics standpoint, even like latent cash or just wallet share accretion type services, a lot of that's going to be coming from your custodian, which is great to some extent because it's at Fidelity. You should be thoroughly trusting—at least it's pretty good—or Schwab or Pershing, wherever, and those services will be coming from your custodian likely at a lower cost. As you get bigger, you want to differentiate. That's when we start thinking about expanding out and owning your own data. You get into the $500 billion range; the suite of tools out there for you to use is significant.
Oleg Tishkevich (44:35):
But even today, if you think about leveraging AI tools on top of a data lake, data lake base cost is what, like 2,000 bucks a month for a firm. So even for certain solo practitioners or people, folks with a couple of people in a team, if you're really serious about growth, are you serious about how you want to become more effective? I mean, I think there are talks about having a solo practitioner be a billion-dollar RIA leveraging AI, leveraging right data. There are already conversations about if somebody's going to do it. We just don't know who, but somebody's going to do it. One-person shop over a billion dollars. I feel like we're going to see it. It's technologically possible with the right structure. So sky's the limit, in my opinion, and things are becoming a lot cheaper than they were even like 12 months ago. That's just what's happening.
Rob Burgess (45:38):
Great. Well, I believe we're about one minute over our time here, so thank you all so much for this great discussion. I really appreciate it. I always say a good interview, I'm left with more questions I want to ask than less, so I could definitely talk to you all for a lot longer about all this, but thank you again. Thank you everyone for watching. Thank you for submitting questions and have a great rest of your day.
Doug Fritz (46:01):
Thank you. Thanks so much.
Brian Wallheimer (46:05):
All right, and we are wrapping up our AI Virtual Summit today. I want to thank everyone, the panelists, the audience, and all of the folks who pulled this together, the tech folks who kept us running. Thank you so much to everyone there. There's a lot of information that came out today over the last couple of hours. I know that it has me thinking of a million other questions. If you have more questions, if you want more information, if you want to dive deeper into what AI can do for the wealth management industry, what it can do for you, what it can do for your firm, please join us this fall at Advise AI, Financial Planning's AI Conference. It's October 28th, 29th, in Las Vegas. If you want, go to financialplanning.com, click on events at the top of the page, and you'll find all the information you need. From everyone here at Financial Planning, I'm Brian Wallheimer, editor-in-chief. Thank you so much for your time and hope to see you this fall.
Building an AI Tech Stack That Works for You and Your Clients & Closing Remarks
August 4, 2025 3:49 PM
47:04