Podcast: Play in new window | Download | Embed
Subscribe: RSS
Come on in and sit back relax, you’re listening to Episode 196 of the WealthTech Today podcast. I’m your host, Craig Iskowitz, founder of Ezra Group Consulting and this podcast features interviews, news and analysis on the trends and best practices all around Wealth Management Technology.
My guest for this episode is Sindhu Joseph from CogniCor. This episode was a deep dive on AI technology and who better to talk about the difference between chatbots machine learning and ChatGPT than someone with an actual PhD in artificial intelligence, and oh, by the way, six patents.
Sindhu originally founded CogniCor technologies in Spain in 2013. While she was running the company there, she won the European Union’s inaugural award for the most innovative tech company impressive and even gave a speech at the European Parliament. She then relaunched CogniCor in the US in 2018. You’ll get to hear all about what they’re up to in just a few minutes.
But before we get started, if you are an executive at a broker-dealer, enterprise RIA, family office or a TAMPs, your tech debt is holding you back. Your old software platforms are rusty and falling apart and they need either a complete overhaul or to be replaced entirely. Your disparate systems don’t communicate with each other and it’s driving your operations staff and advisors crazy with manual processes and other errors. If this describes your company’s tech infrastructure you should run, not walk to our website, EzraGroup.com and fill out the Contact Us form on the home page. Our experienced team can evaluate your technology ecosystem, deliver targeted recommendations, optimize your existing systems and operations, or run an RFP or RFI to help you implement new software to help take your firm to the next level. Please subscribe to this show wherever you listen so you don’t miss an episode. Now let’s kick this thing off!
Topics Mentioned
- Founding of CogniCor
- Augmenting Human Experience through AI
- Predictions and Potential of Large Language Models (LLMs) in Enterprises
- Use Cases from CogniCor
- Adoption of AI and Capabilities of Cognitive Technology in the Wealth Management Industry
Episode Transcript
Craig: It is the founder CEO of CogniCor, Sindhu Joseph. Welcome to the program.
Sindhu: Thank you, excited to be here.
Craig: I’m so glad you could find some time for us. I know you are extraordinarily busy and you are traveling right now.
Sindhu: Where are you calling in? I’m right now calling in from India. I’m actually visiting my team and spending some time with them.
Craig: Where in India are you?
Sindhu: I’m in the south part of India, which is called Kerala. It’s a pretty nice place with lot of backwaters and greenery, a very relaxing place.
Craig: I wish I was there. I’m here working. So I’m jealous. You’re doing combination work vacation?
Sindhu: I wish it was that, but it’s mostly work and hopefully a little bit of vacation.
Founding of CogniCor
Craig: Well, I hope you do get some in, but I appreciate you taking the time because it’s 12 hours later, so it’s 2 o’clock in morning where you are now so thank you so much for being up. But can you please share with us a 30 second elevator pitch for CogniCor.
Sindhu: Sure. CogniCor is an AI assistant platform purpose built for the wealth industry. It’s purpose built for the financial advisors. Financial advisors spend a lot of time in routine operational tasks and CogniCor AI assistant is here to kind of take over all of those routine tasks and guide them through the decision making process so that they can focus on the customer engagement and expanding their portfolio. That’s what CogniCor’s focus is.
Craig: I saw an early version of one of your products at the Pershing Insite Conference I believe in 2019, summer of 2019. That was a chatbot where Pershing has a proof of concept that could go in and say, my client lost their debit card, how do I get them a new one? It would go bring up the right form, fill out all the clients information for the advisors. All you have to do is fill out the last bits and hit print and get it signed and go. That was so cool. What were the reasons why you founded CogniCor, what gaps did you see in the market back then?
Sindhu: Let’s say, traditional, other industries like e-commerce or travel, if you want to buy something from Amazon, it’s in a few clicks you have that product on your doorstep within a couple of hours. Similarly, if you want to hire an Uber it’s again a couple of clicks. The experience is very, very seamless and instantaneous personalized ubiquitous.
Sindhu: That is what we are used to as a consumer. But when it comes to wealth, which is fundamental to our own existence, everything breaks down. The experience breaks down, the efficiency breaks down. Imagine anytime you want to kind of take a distribution from your 401k or even not change a simple thing as a beneficiary for your retirement account, I have no clue how to do that.
Sindhu: I think traditionally the wealth industry has been not focusing on the experience aspect of it as for an investor or for other stakeholders like advisors.
Sindhu: Behind the scenes, behind these broken experiences, there are lot of efficiency gaps because there are a lot of manual processes that’s been happening. This is where we believe that AI and automation can fundamentally change that equation. I approach AI from the point of view of how can we augment human efficiencies at scale.
Sindhu: That is the role of AI that I see. I have been spending a lot of time working with AI technologies much before the large language models, the foundational technologies has been a hype. A couple of decades of experience in building AI solutions. This is where I see the industry can utilize AI solutions in terms of both creating an experienced shift and building efficiencies.
Sindhu: It’s not just a nice to have, it is a must have at this moment. Because as we all know, there is a huge wealth transfer that is happening from an older generation baby boomers to the millennials and they are used to this Amazon Uber kind of experience and if they don’t see that same experience with the wealth industry, they are going to demand that experience.
Sindhu: It is a historic opportunity for all of us in the industry to kind of revolutionize both this experience and efficiency. That’s where it came from and I believe by doing this experience shift and making it seamless, there are a lot of like we are actually democratizing the access to wealth. There are a lot more people who can work with the industry and create wealth for them. That’s clearly the goal of CogniCor and why I was inspired to do this.
Augmenting Human Experience through AI
Craig: Something interesting thing you said was “augmenting human experience at scale”. I like the idea of augmentation because there are some fears around AI replacing humans. But I see it as just pushing people higher up the value chain. Technology is always going to get is going to always going to start automating manual tasks. It’s never going to stop from the cotton gin to the assembly line to the telephone, that automates things. It’s never going to end. This is going to keep pushing humans up in your concept looking for more value.
Craig: Let’s talk about the difference between different types of AI. I mentioned earlier the chat bot. Then there’s machine learning, then there’s natural language processing, generative AI. Can you do a quick 30 seconds on those and differences between them?
Sindhu: In general, AI is basically trying to create intelligence and create tasks that are normally need human intelligence to execute and we are trying to automate those using AI.
Sindhu: One of the fundamental building blocks of that is understanding human language or how we are able to communicate with humans. This is where the chat bots come in, where we can actually talk to a machine in natural language and communicate those instructions for what we want to get things done and how we can use AI to do that.
Sindhu: Initial set of chatbots were just a number of scripts that you would execute. So you would painfully articulate what you want to create with every single instruction and you would train the algorithms to understand that particular instruction in different ways.
Sindhu: That was the initial set of chatbots. It had many ways of building it a certain number of different technologies in machine learning like different algorithms that were used in machine learning was part of a chatbot building process before any kind of statistical processing was used. So that was the initial set of chat boards.
Sindhu: We then went into augmenting this technology with deep learning algorithms which was much more effective in terms of understanding the nuances in human communication or human natural language. It performed better because you know We could train and make the machine understand the human language better. Still, the rest of the workflows, how we want to get the task automated was again scripted.
Sindhu: In addition, the dialogue was also scripted. If you want to get any dialogue between the machine and the humans, you had to basically define each step in the process.
Sindhu: That’s where the large language models came in. One of the fundamental differences between all of the previous machine learning models and the large language models is when the large language models come pre-trained, which means that you don’t have to sit and painfully train how we interact with machines or teach it the advances in human language, it’s already trained. This removes one of the fundamental flows in AI deployment, which is training.
Sindhu: AI deployment always came with a lot of training time and large language models to certain extent, remove that training time. It was able to understand most of the nuances in human communication. That was the large language model based chatbots or assistance that was created.
Sindhu: However, there is still certain things that was missing, which was the context around every communication. Humans communicate with certain context based on the domain that we are working on.
Sindhu: When I say about retirement account, we inherently know that maybe we are talking about a wealth management domain. This kind of context understanding is still missing in large language models, but technologies like knowledge graphs and reasoning can bring in this context.
Predictions and Potential of Large Language Models (LLMs) in Enterprises
Craig: There’s been a lot of talk about large language models recently and one podcast I was listening to, there was a prediction that every private company of a certain size will have their own private LLM and it’ll become table stakes within three or four years because of the capabilities and the advantage it gives you if you can basically analyze every document, every e-mail, every PDF, every promotional material, every research report inside your company walls, then that would provide every employee with a tremendous resource. Do you think that’s true and if so, how would they go about doing that?
Sindhu: Not every large organization would have their own fine-tuned model, but what I believe is foundational models, as I said, come pre-trained with a certain amount of data already, which means it understands pretty much every domain to a certain extent. The vertical solution providers would provide that fine-tuned layer for each vertical. For example, a belt solution provider would train it for the belt domain. For somebody who’s focusing on manufacturing domain would provide that fine-tuned layer, which understands the nuances within that particular industry and provide that.
Sindhu: That is a fine-tuned foundational model. Within every enterprise you can customize it by feeding in the documents, the standard operating procedures, the data that is within that particular company and you can make it your own.
Sindhu: That is what enterprises will be doing. Once you have these models, you can actually use it in a way that is just working on your kind of data. I think enterprises are not trained to kind of do those customizations, but specialized industry. Three providers would be doing that from pretty much from outside. That is what I would think in the next few years. And that’s where the value you can extract.
Craig: I downloaded an LLM off of GitHub distribution that I can run on my MacBookPro to gather all my own data, all my documents, all my emails. Do you think people will have their own private LLMs as well?
Sindhu: Having their own private language model, I would say it’s fancy. It is not inconceivable. But the problem with where you would stop doing that is because there are no great frameworks available for fine-tuning these large language models. How do you kind of make sure the accuracy is 80% or 90%?
Sindhu: There is no models that tells you what the accuracy for a particular domain or a particular use case. If it’s your own use case, you how do you test that? I think there is a lot of infrastructure that needs to be built around these language models before it’s usable, widely available for this kind of use.
Craig: I would agree. I just think the infrastructure is being built so quickly. The tools are moving so fast. It reminds me of the internet in the early 90s or mid-90s. HTML spec was being modified every week. Now I put an image on my website. This is great. Now we’re seeing that with AI that we’ve got all these new features and functionality coming out all the time.
Use Cases from CogniCor
Craig: Let’s talk specifically about use cases. So, CogniCor is a number of very interesting use cases for AI, I know one of them is onboarding. Can you give us a little bit about suitability and client onboarding how AI can help scale that?
Sindhu: There is a lot of manual operations that operational workflows that advisors go through on a daily basis. Imagine you have your Alexa or your Google Assistant at home and you can simply tell your Alexa, turn on my kitchen light and it does that, but you haven’t programmed Alexa to understand or execute that particular skill. That skill is downloaded to your Alexa and somebody around the world has built that skill.
Sindhu: That is what we want kind of CogniCor and its ecosystem to kind of build for the wealth industry. If I want to change my beneficiary, I can simply say, change beneficiary or as an advisor, I can say change beneficiary for my client Alex. It just goes and collects the right form, executes the operational approval flows and then pushes every data to the custodian gets the status update.
Sindhu: That entire packaging is what we are envisioning. Imagine every single client servicing activities are managed through that then the entire industry becomes standardized as s result.
Sindhu: As an RIA, you don’t have to figure out how these operations are handled within your organization. It just becomes part of the standard workflow and you can focus on declined engagement and your expansion of the AUM. That’s what in a very high level, what CogniCor is looking at.
Sindhu: Some of the top use cases that we have been working on is as you said, onboarding, client onboarding is a great use case for this purpose because in the process of doing that, collecting data, there is a large form that normally needs to be filled. But in this case, you can go through a process of AI enabled data collection, which looks at what is your process.
Sindhu: This is your network can be provided some similar kind of situations where a similar profile has taken certain decisions and understand your risk profile by asking certain questions that is generated by AI.
Sindhu: Through this interactive process we collect your data both risk profile, your investments, your assets and so on. Once we collect all of these data we again push that to the CRM and also to the custodial platforms for both for onboarding purposes and opening an account as well.
Sindhu: That is one of the workflows that we handle. There is another use case which is around client review meetings that the advisor goes through again on a daily basis. Basically handle that entire end-to-end process where you can get reminded for an advisor that this client is up for review and we scheduled the meetings automatically, prepare the agendas and then also capture meeting notes to extract action items from there and map them to cases to drive certain workflows from there. That end-to-end experiences built using CogniCor. So these are some of the exciting workflows for RIAs that we have seen capturing their key pain points in an exciting way and creating efficiencies and experiences for advisors.
Craig: Sindhu, can you share a little bit with the audience, what is your background and your training in the AI space?
Sindhu: I have been working in the AI sector for a couple of decades now and the background story is that I was excited about human intelligence. I was fascinated by how we make decisions and how we understand and think through a decision-making process. Then I heard about a field called artificial intelligence and I was fascinated by it because I realized that by trying to experiment and trying to build this kind of artificial intelligence, we can actually understand our own nature of intelligence, how do we make decisions and so on.
Sindhu: My interest in AI came from that point of view of natural systems behave the way they do and how we can have a deeper connection, deeper understanding to that.
Sindhu: I did a PhD in and spent a lot of time working on building these intelligent autonomously decision making systems. I used a theory called the theory of coherence, which is a cognitive theory and humans use that to make decisions and I made a computational model of that for AI systems to make decisions.
Sindhu: That’s my background in terms of the AI world. Using this expertise, I have spent a lot of time working with natural language processing because I believe if artificial systems are able to understand human language, that is a holy grail of AI, because once it understands and is able to communicate with humans, it can execute past what humans command. Combining these two things, the autonomous decision making and understanding humans were some of the areas that I was focused on.
Sindhu: After my PhD I also co-authored a number of patents in machine learning and natural language processing. I realized that there is a lot of potential for AI in these applications, domains, I was fascinated by the use of AI and how we can create this kind of platforms to support efficiency and leadership and that’s why CogniCor was created.
Craig: That is a fantastic story. You have a PhD in artificial intelligence.You know what you’re talking about, you have many years of experience in this space.
Craig: We do a lot of research in this area and clearly, every industry is rushing headlong into AI of course at different speeds. We’ve seen a few surveys that show, in terms of the biggest wealth management firms across the board, 80% are either deploying or scaling client or advisor facing AI power technology. Do you think these firms are moving too fast, should they slow down? What’s your recommendation?
Adoption of AI and Capabilities of Cognitive Technology in the Wealth Management Industry
Sindhu: The ChatGPT and OpenAI has created a hype, a lot of hype, not just in the wealth industry, but throughout rest of the industries as well in terms of adopting AI technologies. However, the AI platforms, AI solutions has been there before. Although the mainstream has not been fascinated by it, I would say to certain extent. In terms of adoption, I believe there are certain use cases that are mature for adopting.
Sindhu: For example, in the wealth industry, I’m not talking about building an AI advisor per se that would look at your client’s household, the portfolio performance and then turn out automated recommendations for that. I’m not talking about those kinds of use cases. I think they are pretty far fetched even today.
Sindhu: But the use cases that are mature in the industry are, again, augmenting human efficiencies. How can we delegate certain administrative routine tasks that would enable humans to perform better, humans to make better decisions.
Sindhu: Those are the use cases that is matured, the platform companies have been working on it for quite some time. I think in terms of adoption, a couple of things that large firms as well as the RIAs need to be aware of, understand there are a lot of firms out there, everybody offers AI. You need to understand how do you differentiate between a provider with certain capabilities versus somebody just using an API and providing some service.
Sindhu: This differentiation is key because someone who is providing all of the infrastructure to fine tune deploy and expand these use cases will be needed as you deploy and roll out your cases.
Sindhu: Those kinds of infrastructure you have to check whether the these capabilities are there available. That is one checkpoint that companies have to do in terms of either bringing in external vendors or like internally if they are developing as well, they need to look at that.
Sindhu: The second is there are a lot of people who are directly using chat GPT. While it is a fantastic tool, if you are calling chat GPT with your customer data or your own information. Unfortunately, that information will be used to train ChatGPT’s next generation, which means that your data will become a permanent part of its training, its model.
Sindhu: You would never want to do that at the capacity of somebody who is handling customer data. You should look at models that can work in a closed space where your customer data, your data is isolated and it is not used to retrain those models. You should look at open models versus the closed models. That’s the other differentiation that you should be focusing on.
Sindhu: Third is, of course, the use cases. As I was saying, in some use cases are matured and some use cases are not. If you kind of start where your pain points meet those use cases that are matured, I think you are in a perfect place to roll out the scientific solutions.
Craig: Sindhu, we’re kind of running out of here. We’ve covered a lot of ground. Before I go, I wanted to ask, what are some items on the roadmap of CogniCor? What can we expect to see in the next six to twelve months?
Sindhu: So as I was saying, our vision is to create this standardization in the industry and create that experience shift, just like how you are calling an Uber or how you are buying an item from Amazon. That kind of seamless experience if we can create within the wealth industry, that is our end goal in terms of that AI assistant doing that for you with a very seamless, conversational interface and behind the scenes automating and stitching together different systems. that is what CogniCor is focusing on.
Sindhu: Once we have these basic infrastructure pieces. For example, in our own use case as the meeting assistant, we are actually automating certain like routine activities like scheduling preparing the agenda, capturing meeting notes and converting them into action items. That becomes a just a base for us and imagine now the possibilities are endless through that pipeline that we have created.
Sindhu: For example, we can analyze the context of the household. Let’s say there is a kid in the household, five years old, no 529 plan. Suddenly the AI can start to create these next best actions. It becomes part of the agenda. It is in front of the advisor, as they are going to meet with the client, they can propose we notice that there is no 529 plan for your son, maybe it’s time to put in put that in place.
Sindhu: Likewise, we can look at portfolio performance and suggest these actions and a lot of these analytics becomes part of this enriching that basic infrastructure piece. I think that’s where the value is so that the advisor becomes a very powerful decision maker who’s understanding the context of the investor client and providing really value added advice to the clients.
Craig: Sindhi, you’ve said it all that’s where I’m really looking forward to seeing some of those capabilities coming from the roadmap, very excited for what’s coming for advisors, broker-dealers, asset managers, other wealth managers in space with this kind of technology. Where can people listening to find more information about your firm?
Sindhu: Our website is a pretty good starting point. Anyone who wants to look at our demos can reach out to us and schedule a demo meeting, we would be glad more than glad to walk you through some of the top use cases our firm is configuring for both the RIA market as well as for the broker-dealer, wire houses.
Sindhu: Also there are a lot of customers who are excited to talk about CogniCor. You can also reach out to some of your peers who have implemented CogniCor and they can narrate some of their experiences of working with CogniCor.
Craig: CogniCor.com.
Sindhu: That’s right.
Craig: So thank you so much for bringing the program. Really appreciate your time.
Sindhu: Thank you so much for having me.
