The story of Frankenstein, written by Mary Shelley in 1818, is a cornerstone literary work that contemplates humanity’s relationship with a form of sentient technology. The book’s relevance has only increased over the past 200 years as the relationship between humans and technology has become more intertwined.
This relationship is also the core of one of my favorite films, Young Frankenstein, directed by Mel Brooks. Once Dr. Frankenstein builds his monster, he must reckon with the power of his creation and the dangers it possesses—as well as the risks it has to the good doctor’s own identity.
The rise of powerful artificial intelligence tools, such as ChatGPT, that can generate new text, images, audio, and video based on past learning has permeated our daily lives and raises questions about how it can be safely implemented and managed. How do we gather and organize the massive amounts of data needed to feed AI systems? Which human jobs will AI replace? How can wealth management firms incorporate natural language models without exposing their private corporate data?
All these questions (and more) were discussed at the BNY Mellon | Pershing INSITE23 Conference panel Tech Forward: Is Your Firm Ready to Leverage Emerging Tools? The panel was moderated by Kiran Nair, Director of Technology Product Management Integration at BNY Mellon | Pershing, and featured Craig Iskowitz, CEO of Ezra Group; Sandeep Kumar, Managing Director and US Wealth Management Technology Consulting Leader of Ernst & Young; and Christopher Napoli, Industry Principal of Financial Services for Snowflake.
The four gave their thoughts on the current technology landscape, theories about the future, and practical advice for firms looking to merge AI solutions into their tech stack.
Technology Trends
Underlying all wealth management technology is the question of how these tools help improve the advisor and customer experience, Kumar posited. Clients expect more from their advisors, and technology is at the heart of how firms can keep up with those demands. AI functionality has quickly become the fastest growing addition to software for advisors, with two-thirds of fintech companies already incorporating it to some extent.
Personalization is the next most popular trend with programs such as direct indexing being made available to the mass affluent due to the rise of fractional shares trading and tax-sensitive trading. Kumar also observed that clients are now willing to share more of their private data than ever before to receive more targeted experiences. Ernst & Young’s recent survey found that 72% of customers are willing to share data with their advisors and are more willing to share data with them than they are with their doctors.
The Kitces/Ezra Group AdvisorTech Map, a resource which aggregates information about software products and vendors across the industry, has recently added a new category based on a rising wealthtech trend called Advice Engagement. Iskowitz explained that the category includes tools which provide ways for advisors to better communicate with clients and build stronger relationships with them outside of their financial lives. Applications like LUMIANT, AssetMap, and Elements illustrate client life events, fears, risks, and priorities to encourage better connections and conversations with clients.
Napoli explained that what has made all this exponential growth possible was migration to the cloud over the last decade or so, permitting us to develop solutions that are able to leverage the numerous GPUs, CPUs, storage, and external programming now available. The cloud has enabled the creation of functions such as personalized algorithms, next best action systems, and micro-segmentation, he said.
Snowflake builds on this cloud technology, reaching across the three hyper scalers on AWS, Microsoft Azure, and Google Cloud to allow for the normalization and curation of data, Napoli noted. That data can then be transmitted to develop and build applications in the future. The company is currently positioning itself to become the fourth cloud provider, where data management, development of applications, and usage occurs. (See Running Up the Score: How Predictive Analytics Gives You an Advantage Over Your Competitors)
Data Management Challenges
Just like Frankenstein’s monster was only as good as its brain, a natural language model is only as good as its data. Napoli described the AI engineer’s relationship with data by comparing it to three dragons where the data dragon is often an idiot.
Artificial intelligence and machine learning may be powerful, but what goes into your system is what inevitably comes out of it. “If your data isn’t controlled and curated to generate the answers you need, none of the other elements matter,” Napoli emphasized. To this end, Snowflake recently purchased NEMA which sources high quality data for their system based on peer-reviewed scientific papers.
While some companies have started placing data front and center, many others continue to leave it behind. Ernst& Young’s same survey found that only 13% of wealth managers are correctly equipped to handle the data they collect, and a study by Forrester looking at Fortune 1000 companies and small businesses found that less than 0.5% of data collected by companies is ever analyzed or used at all. (See The Journey Towards Data-Driven Wealth Management)
Another of the major data-based issues facing wealth management firms is inconsistent records across different sectors of the same business. Kumar recalled working with clients who have five different records for the same customer, and pointed to this missing single source of truth as a common gap holding companies back from modernizing their platforms.
Iskowitz echoed Kumar’s concerns around duplicated data, focusing on issues around data redundancy in large broker dealers and RIAs. Ezra Group offers a data optimization service, where they investigate all the vendors a firm is using and identify overlapping data sources. In many cases, different departments brought in their own technology solutions which provide the same data which the firm winds up paying for multiple times.
Resources like Snowflake are making it increasingly simple to bring in data from different sources, but Iskowitz cautioned that data analytics isn’t useful without a focus on business outcomes. The wrong question to ask is, “what data can we get”? The right question is, “what are we trying to get out of this data and where are we going with it?”
Another pain point Iskowitz highlighted was the prevalence of data silos. Financial planning data is kept in one silo, separate from CRM data, and separate from portfolio management data, which makes it difficult for information to be aggregated across the firm. As a result, most data is analyzed inside its silo, and larger trends are missed. Connecting data across the entire firm creates a better foundation for features like next best actions, Iskowitz emphasized.
The Power of AI
The sudden arrival of ChatGPT onto the technology landscape marked the rise of a new type of artificial intelligence, known as generative AI. Iskowitz explained that up until ChatGPT, the public had mostly engaged with conversational AI, which often takes the form of chatbots and simulates human conversation in a very narrow focus.
ChatGPT is the first widespread example of generative AI, which creates new content based on the data it was trained with — in this case the Common Crawl database of billions of web pages as well as multiple other sources, like Wikipedia.
This wealth of information and creative capacity positions generative AI models to take over much of the mundane work that data analysts and knowledge workers perform. Analyzing files to create charts and compiling information are tasks that skilled workers can remove from their workload to focus on higher level activities. Iskowitz encourages his employees to try ChatGPT and other generative tools for any online tasks and has seen efficiency rise among his team
Multiple conversations are happening every week at Ernst & Young about the possibilities and effects of AI, Kumar observed. Though the technology has a high barrier to entry and isn’t necessarily seeing immediate adoption, it is increasingly the central point of focus for conversations around future growth. At Orion’s Ascent Conference this year, 33% of those surveyed listed AI as the most disruptive technology trend facing the financial services industry.
“Anything you learn past the age of 18 has the ability to be replaced by AI,” Napoli declared. If you learned something from reading books and doing research, then AI can learn it, too. When looking down the line to determine how AI will impact wealth management, Napoli was hopeful and insisted that the technology will open up new possibilities for what advisors are capable of and how they can spend their time.
He cited a McKinsey survey showing that 80% of advisors’ time is spent on non-revenue generative functions, and then predicted that this statistic will eventually reverse itself so that they are spending 80% of their time forming relationships with clients and doing what they’re best at with the rest of the functions managed by AI. “I don’t need to know how to code anymore,” Napoli explicated, “I just need to be able to tell the machine what I need in the most descriptive syntax possible and then copy and paste the answer.”
The one area he predicted would remain out of reach for the current era of technological advancement is portfolio management, which he believes to be still much too complex for the current AI models to understand. (See We’re Running the AI Leg of a Digital Marathon in Wealth Management)
Assessing The Risks of AI
While AI has many promising aspects, “most of our clients are focused on high value, low risk solutions,” Kumar explained, which leaves many use cases out of reach. For example, AI making recommendations that can impact investment plans would certainly be valuable, but it requires customers’ personal information to be entered into the AI system. This creates a high-risk factor as that information could potentially be shared with other users.
One of the places that AI can be safely and effectively used is enhancing knowledge repositories. Rather than advisors picking up the phone and calling the home office for information, many of the most common questions can be answered by AI-powered systems. If progress is made slowly and carefully, Kumar foresees many sectors including market research and portfolio curation being fully taken over by AI in a few years.
As a counterpoint, Napoli contested that it isn’t necessarily that the technology will become safer but that users’ tolerance for risk will increase over time as the tools are integrated into their daily lives. He described how Uber used to only have black cars, until passengers became comfortable enough with the system to get in any car that showed up with matching information and UberX became the default option.
Another possible future was proposed by Iskowitz, who made a prediction that most large companies would soon all have their own private language learning models. This would eliminate the risk of putting sensitive information into a public model and could even work without access to the internet. Perhaps one day everyone will have their own language model on their phones instead of just chatbot assistants, he postulated.
This type of technology is perfect for retrieving information and could greatly improve efficiency for consultants and other knowledge workers by making archival knowledge scalable, Napoli emphasized. However, he also observed that it would be nearly impossible to run an AI system fully in-house using cloud resources with current technology, as the process is very computationally intensive, and you’d certainly run out of server space.
Current Use Cases for Safe AI Implementation
While some AI use cases remain out of reach, many companies are finding novel ways to incorporate machine learning technology into their user and advisor experiences. Pershing has implemented an intelligence search system based in machine learning, Nair described, which is hyper personalized to keep track of users and feed them relevant information. They’re also working to build out a system to generate next best actions and AI assistants that augment all aspects of an advisor’s workflow.
Iskowitz brought forward three examples of software vendors that have built successful products leveraging AI technology: ForwardLane, Pulse360, and Intergen Data. ForwardLane takes clients data and runs it through AI models to produce customized ‘signals’ telling advisors what to look out for. Pulse360 saves advisors over 10 hours a week via meeting automation, and also launched a natural language model specifically to generate written content for advisors. Intergen Data gathers information from a wide variety of public and private sources to generate predictions around the life and health events that are most likely to impact an individual that an advisor would need to account for.
Implementing AI in Your Business
When it comes to the practicalities of implementing AI-based solutions, Kumar recommended a top-down approach starting with strong buy-in from leadership and establishing a clear direction with end-to-end use cases. “This isn’t just a tech issue,” he reinforced, “you need to have the right technology infrastructure to handle the volumes of data and analytics involved.”
Beyond buy-in from leadership, Iskowitz stressed the importance of strategy cohesion across all parts of a company. A simple way to clarify technology decisions is to run an annual technology review meeting where stakeholders across the organization review the tech stack to assess what is working and what isn’t. Now that AI technologies are more common, it’s natural that they enter those conversations.
Another important practicality Iskowitz highlighted was the cost of AI solutions, not just to build and implement them but to access all the GPUs and data sets required for daily use. He recalled a recent client issue with a single employee running a query that had such a wide scope it cost the company tens of thousands of dollars. Setting limits for what each individual can run on your cloud system will reduce the risk of any one employee running up unexpected costs.
Conclusion
There is no doubt that the AI revolution has already arrived, but wealth management firms need to take careful steps to ensure that they have the infrastructure in place to support the technology and protect their clients. Comprehensive data structures, vetted security measures, and clear leadership will set your firm up to take advantage of all AI has to offer without undertaking undue risks.
