AI for Wealth: Stop Losing Clients & Save Thousands!

Hello everyone. Thank you for being here. My name is Gokul and I'm a product lead at Altruist. Altruist is specifically in the wealth management industry. And I specifically lead the AI product, which is called Hazel. And I'm going to walk you through a little bit of what Altruist does, to begin with, what are the big problems that we're tackling in the wealth management space, and the kind of technology that we're building with AI to solve a lot of those problems and challenges for our core customers, who are financial advisors. Does that sound like a plan? Let's do it. Um so, this is a little bit about Altruist. Altruist is building the system of record in wealth management. Our typical users are financial advisors who serve retail customers. Altruist was founded in 2018 by Jason Wenk, who's uh repeat founder. He bootstrapped a $50 million wealth management firm, and then he we've raised $600 million to to build the system of record in this industry. And the system of record is called a custodian. And what that fundamentally means, to break it down, is if, let's say I'm I'm Gokul at San I live in San Francisco and I own 100 shares of Apple, this is my social, this is where I live, and this is how much I bought those Apple shares for, these are the realized losses, gains, etc. That whole piece of information is ultimately stored with this financial institution called a custodian. And that's what Altruist is. Altruist is the fastest growing custodian in the US by number of RIAs served since we began in 2018. We've moved from We've moved to become the fourth largest custodian by number of advisors served. Today, we power more than 6,000 advisors. And what we've built is not just a custodial layer or the system of record. It's the whole software stack that's built on top of the custodian. A simple analogy is Altruist is basically Shopify for advisors. Right? You just not get the system of record, but you get to run your whole operations and your business on top of it. Now, let's take a look at

what is the biggest challenge that financial advisors face today. This is the size of the opportunity in this industry, right? It's a massive industry, 104 124 trillion dollars of wealth. And there is a big intergenerational wealth transfer here that's going on. I'm sure some of you or most of you may have heard about it. There's wealth being transferred from the baby boomer generation to Gen X and even millennials and Gen Zs, right? And what's happening is 60 to 70% of these heirs leave the parents' advisor. Now, why is that happening? Because values differ from for subsequent generations, their investment philosophy, their goals in life drastically differ, right? Um so, a lot of advisors lose clientele and assets as they are passed through heirs. McKinsey had this report that they put together which showed that there's going to be a massive shortage in financial advisors by 2034. by 2034. So, this is a big problem where there's more need for better, high-quality financial advice. There is a shortage of advisors and there is a proliferation of B2C AI wealth management solutions. So, how is it that we can poise and position advisors for success here? That's something we think about every day.

Now, what are the more tactical challenges that advisors have to face? What comes in the way of them being successful? Is this chart. This chart is is a comprehensive list of all the different pieces of software that a financial advisor has to choose from to build their technology stack. There are over 550 different pieces of software here. The typical advisor spends 45% of their time working across eight to 10 different solutions or pieces of software, moving data across those pieces of software just to do their work. Financial advisors are typically asset gatherers. And And their best their time is best spent in front of clients. Either talking to existing clients, deepening relationships, understanding their pain points in serving them better, or growing their business by building pipeline, and effectively acquiring new clients. The more time that they spend on the operational side is less time spent doing that and growing their business or retaining their business. Now, if you take a step back

and look at what is the value proposition of an advisor? We are living in a world where AI is proliferating and can provide financial advice to a certain extent, right? Or at least suggestions. Now, so what is the exact value proposition of an advisor? To really answer this question, let's take a step back and and take a look at how financial advice has evolved over time. In the '80s and '90s, the only way you could access good financial advice was through wealth professionals and financial advisors. Then the mid-'90s saw the birth of what's called discount brokerage firms. And they are the ones who first democratized access to financial advice directly to the retail investor. You wouldn't have to go through a wealth professional to access high-quality financial advice. Right? At that point, financial advisors moved from just getting and providing access to investments and instruments to providing holistic solutions like tax planning, estate planning, optimizing cash flow, planning for retirement, etc. So, this was when financial advisory moved from being brokers to planners, right? And that trend just continued into the 2000s, right? Uh the 2000s is an interesting time, which is when the first robo-advisors came up, right? So, retail investors had better access to choice and instruments that they could directly use to make investment trades, etc., right? And then what you see is advisors' responsibilities move higher and higher up the chain, so to speak, right? Like now advisors help retail investors with goals, coaching, financial planning, etc. Now, in an AI-powered world, what does what does financial advice look like? Now, AI is going to optimize anything that you can put a number to. If you can If If If you can put taxes, risk-adjusted returns, portfolios, dollar Any Anything that has a dollar number to it can be optimized by AI. So, what's really important now? It's everything that's left unsaid. It's everything that's ambiguous. It's everything that's unclear. Right? And that includes what's really important for you in life. What are the goals? What is the legacy that you want to leave? What's How do you make trade-offs between the things that are important for you in life? And how do you really optimize your money, your investments, and all the work that you've put into your profession to to align the money with your goals and with what you really want to do with life. And advisors really help bring that to clarity for retail investors.

And what we're doing fundamentally with Hazel is empowering Hazel is empowering the AI native the AI native or the AI powered advisor. How can we use AI to not just give the advisor back a lot of time, which you saw earlier, like 45% of their time is spent on a lot of operational work. At the same time, how can you use AI to supercharge the advisor-client relationship? relationship? Such that advisors can develop a share like orders of magnitude better value to their client.

And this is exactly how we're thinking about building Hazel, right? Hazel is AI that's built specifically for wealth managers and financial advisors. It's vertical AI that's specialized for financial advisors and wealth professionals. Now, I'd like to share a quick 90-second reel with you to give you an overview of what the product does today.

So, that's an overview of what Hazel does today. Now, the first the reel started with showing how you could upload a few tax documents into Hazel and within a few minutes capture key insights about how you could save thousands of dollars for your clients. We launched this product, which is called Hazel tax planning, about 10 weeks ago. And since then, And since then, we've noticed that the financial markets have fundamentally repriced AI in wealth management. There was a $150 billion market impact across major wealth companies across the world when this happened. when this happened. The Hazel launch was covered in Bloomberg, Wall Street Journal, CNBC, etc. Now, how did we actually go about

building a product like this that that could actually deliver outcomes for financial advisors? I'd like to get into the meat of the technical challenge here, right? Like we look at the problem across four different categories. There are four hard problem spaces there. The first is data ingestion and integration. You're dealing with both structured and unstructured data here, right? Like you have information that exists in databases. The same time, you have information that exists in meetings, conversations between advisors and their clients. And there's key information present across both and both to put together provide a holistic picture of what's really important for the client. And then there is being able to search and retrieve across all of this data, right? In a safe and secure manner. And then there are specialized AI agents that you build on top of the search and retrieval infrastructure that essentially converts this data into actionable outcomes. And in this case, that means like minimizing taxes or improving your risk-adjusted returns, etc. The last part is proactive AI, which is being able to surface things before you ask for them. Right? And this is in some sense a progression of the sophistication of AI that's being built.

Now, here is how we think about building AI AI for wealth management, right? Um there are a few separate categories or layers I'd like to walk through. The first thing is we are an AWS shop. So, AWS is at the foundation of everything that we have built. Um so, you can see the horizontal section at the bottom includes like, you know, very foundational uh key parts of AWS, which I'm sure you're all aware of from firewall management to IAM, access management, to systems manager, to secrets key rotation, etc. The key part is um we have built an SSC-based um real-time streaming chatbot, real-time streaming chatbot, essentially, which allows advisors to interact with Hazel, right? And the engine of Hazel. Uh from here, it goes into our proprietary intelligence layer. This is the layer that's been built for wealth management. We have built our own harness that effectively takes the user query, breaks it down into multiple sub queries or tasks, if you will, and identifies the right model or combinations of models for a given query. And a lot of this is actually powered by

our evals and benchmarking, which I will get into next. So, the key three parts of this intelligence layer is the multi-step reasoning part of it, uh the context management, and the multi-model routing, right? We always try to find the best model for a given task at a given point in time. What we're trying to do, because we're building vertical AI, is bring the best of general-purpose intelligence with domain knowledge. with domain knowledge. Right? So, we effectively have to combine this general-purpose intelligence this general-purpose intelligence to the key tools or the key domain expertises that's available in this industry, and that includes, in this case, tax planning, like custodian information, email agents, and a bunch of other tools. On average, Hazel is connected to 30 to 40 tools for advisor that it has to query across. And for semantic search, we use uh our pipeline is fundamentally based on AWS. We use OpenSearch and a combination of Titan and Cohere models for the embeddings. From here, From here, the data is fundamentally fetched from our data stores, which includes, like, AWS S3, the Postgres layer, and then and ElasticCache uh Redis sessions for caching and session management for the different streaming sessions or chat sessions that advisors have with Hazel. Now, how do we know that this is something that works? something that works? How do we know that it does what's important for advisors? How do we know that it actually can deliver on outcomes for advisors and their clients? And the heart of this is eval's and benchmarking. And that's absolutely crucial to ensure that the AI ultimately delivers outcomes and meaningful outputs with high accuracy, right? And this comes down to, like I said earlier, combining domain intelligence or domain expertise with the general purpose intelligence. We work closely with advisors. We have in-house advisors as well who really list out and painstakingly build out the data set that reflects the top use cases that advisors engage in across tax planning, financial planning, investment analysis, etc. Right? And this becomes the heart of our eval's process. And then, as we iterate and experiment with building out our harnesses, we score each harnesses performance against this data set. And until it hits a certain bar of accuracy and a bunch of other scores that we've built internally, it never gets shipped to production, right? So, essential part of this is making sure that your eval's is a part of your CI/CD pipeline. Right? Like you earn the right to be deployed only if you hit a certain benchmark score that effectively guarantees that you maintain a certain level of accuracy and fidelity across the suggestions that Hazel makes for advisors. And this comes down to repeatedly being very diligent and rigorous about deploying this process. So, for instance, to give you practical example, a few weeks ago there was a great foundation model uh company which launched a new version of their model. We actually put it through our eval process and within minutes we were able to find it actually regressed on a few benchmarks. Latency was half the previous model version, but accuracy and completeness was actually lower and we didn't really go through with deploying it in production.

Now, this is a regulated industry. So, how do you ensure that the AI that you deploy is enterprise grade? The first thing here is Hazel is built by Altruist. And Altruist is a regulated entity in this space. It's a custodian. It is effectively responsible for holding your assets. Shyam. Shyam. It is non-trivial to have Hazel deployed within Altruist itself. And today hundreds of Altruist users use Hazel. So, the fact that we managed to convince our compliance, security, and infrastructure teams was itself testament was itself testament to the quality with which the product was built. Beyond that, we have zero data retention with all key AI model providers that we work in our stack. And of course, we're SOC 2 certified and we have transcription both rest and transit. We have online evals and scoring that's happening real time on advisors using the product. So, part of it is scored samples, so we're always looking for regressions and to catch any issues that might be happening in real time, which is then, you know, queued up to our alerting mechanism and DataDog. We have a more robust role-based access permission setup that ensures advisors can set up access to Hazel as they see fit amongst their team members. Now, we've walked through quite a few aspects of aspects of how Hazel was built. Now, what was the

impact of Hazel? Right? In the last 10 weeks that we have launched, the typical advisor the typical advisor saves $5,900 saves $5,900 in taxes per client. The typical advisor serves about 100 households, and that fundamentally means half a million dollars in savings per year across all the clients that they manage. And you get these savings while saving 5 hours per week, which comes down to a full calendar month, 30 days a year. And all it takes Hazel to generate a tax plan is 10 minutes. So, that is the impact of being able to build high-quality AI that actually delivers value and outcomes in a regulated environment.

We have a couple of testimonials here from advisors. from advisors. A lot of this can be found on social media and LinkedIn as well, where advisors upload a tax return into Hazel to test it out, and they're practically blown away by what it can do. This is an interesting example of an advisor who runs an outsourced tax management business. That is, his fundamental profession is to do tax planning for other financial advisors and their clients. So, on average, he does like 50 to 100 tax plans in a month. And something that he had set aside, a very complex tax return for a client with multiple businesses and entrepreneur themselves, he had set aside a full day to actually work through that client's tax situation, and within 5 minutes he was able to get it done. This is real-world impact of something that we've built. But,

But, the harness that we've built is so general purpose and foundational that that we're launching our next AI agent, which is going to allow advisors to do holistic financial planning in July. Thank you.