How LLM 'superbots' are transforming wealth management

Thanks in large part to the record growth and awareness of OpenAI's ChatGPT, curiosity is growing about the transformational potential of large language models, or LLMs. Here, we consider the possible role of the emerging technology for wealth management firms. 

LLMs allow for the integration of artificial intelligence with human intelligence in order to augment human abilities, from generating content and answering questions quickly to improving decision-making and problem-solving abilities. 

Next-gen bots powered by LLMs — aka "superbots" (in addition to ChatGPT, the open source LLM Bloom and NVIDIA NeMo) — offer the promise of moving beyond single-turn conversations, scripted interactions and rules-based, hard-coded logic. Intelligent virtual assistants can provide conversational language and multi-turn conversations by employing machine learning to continually learn and improve, recognize and analyze sentiment. 

Sentiment analysis is the process of recognizing emotions expressed in text and can analyze customer reviews or feedback to determine the overall views about a product or brand. Advisors can then calibrate responses to clients by asking the bot to fine-tune the response. This process can accommodate both structured and unstructured data to deliver fully contextual multi-turn conversations. 

In the context of investment decision making, natural language processing (NLP) — one of the multiple AI technologies represented by LLM — plays a significant role. NLP powers the "bot" in LLMs by translating data into humanlike output. NLP can currently decipher, merge and connect data — images, speeches and regulatory filings, for example — to create context and content. 

Wealth managers — assuming the appropriate regulatory and compliance oversight is in place — may choose to further embrace and develop use cases such as reviews of market trends and future market outlooks, client portfolio updates or specific investment opportunities. This work is currently not performed by bots but by market analysts or chief investment officers; the content is then reviewed by compliance and must conform to regulatory guardrails. 

Superbots can transcribe earnings calls and create financial reports, benefitting advisors and firm employees across the front, middle and back offices. They can create ideas for customer engagement, jump-start the content creation process, create and edit social media posts and marketing content, aggregate data to create client profiles, streamline the onboarding process and improve understanding of a user's business. Apps using LLM can function as a virtual assistant in order to provide technical support and troubleshooting.  

Curate and clarify
A chat app Morgan Stanley is planning to deploy this year for its network of financial analysts is one such example. Currently being tested with 300 advisors and client service associates, the app leverages OpenAI's GPT-4 — the world's largest and deepest LLM — to curate data from more than 100,000 internal documents. 

This use of generative AI scales Morgan Stanley's intellectual capital in a privately controlled ecosystem. Transparency is a priority. For example, if an LLM-powered virtual assistant suggests an investment to a client, an audit trail (including asset allocation rationale and regulatory analysis) can help clarify why the recommendation is in the client's best interest. Morgan Stanley's advisors and client service associates will be provided with the rationale for the answers they're given, along with an audit trail of documentation used to create the response, to help demonstrate the value of these LLM-generated recommendations.

Artificial intelligence

Wealthtech experts say advisors who don't embrace AI now are 'crazy'

July 18, 2023 7:00 PM

LLMs can also be used to provide sentiment analysis and synopses of news by  recognizing emotions expressed in text and can comprehend the tone of a statement, as opposed to merely recognizing whether particular words within a group of text have a negative or positive connotation. For example, answering the investor question, "Tell me about this stock relative to a market downturn?" is an effective use for LLMs.

Risk, rewards and costs
As third-party tech providers, particularly ones specializing in cloud and conversational AI, improve LLMs for wealth management applications, wealth managers will need to keep abreast of regulatory developments and discuss regulatory expectations with their tech providers. 

In the near term, LLMs will continue to deliver greater performance and text generation accuracy along with improved ability to mimic human text and speech. However, the industry has yet to overcome some major pitfalls associated with the current LLMs, such as inaccuracy, lack of consistent responses, outdated data and the integration of regulations. The future of LLMs depends on model improvements, fine-tuning by AI vendors, transparency, regulatory clarity and the elimination of harmful elements. 

The White House's Blueprint for an AI Bill of Rights — "a set of five principles and associated practices to help guide the design, use and deployment of automated systems to protect the rights of the American public in the age of artificial intelligence" — may provide critical guidance in the rapidly growing AI market.

As LLMs move from the present phase of early adoption and R&D to full maturity by around 2030, costs will come down, helping drive the commercialization of competing models. Specific early adoption use cases in financial services include Stripe's use of LLMs to improve customer support and the vetting of users, FMG's AI-powered content personalization engine and Orion's use of ChatGPT to compare portfolios, refine marketing content and respond to RFPs. 

Bank technology

Can AI do the work of research analysts?

June 29, 2023 1:25 PM

Wealth management firms that delay evaluating and adopting LLMs will give an advantage to early adopters that may prove difficult to close. Organizations that continue to rely on error-prone, tedious manual processes, i.e., front-office facing responses to common queries, data collection processes linked to onboarding and compliance/regulatory checks, may encounter long-term challenges to efficiency, performance and profitability.   

Integration unknowns
Wealth management firms must consider the impact of AI applications on employees. AI and LLM will change the way organizations operate. It is difficult to project if this will result in downsizing, as it may require new resources and skills, some of which may be fungible. It may also increase efficiency and productivity in ways that allow employees to simply do more. We are just beginning to think about how this transformation will happen. 

As for customers, those who are accustomed to interacting with LLM chatbots that provide multi-turn chat capabilities won't find the slow, stock answers of rule-based bots satisfying, potentially harming the overall customer experience. 

For those wealth management firms that choose to move forward with LLMs, augmented intelligence tools must be thoughtfully integrated into the workflow and easy to use — or risk being new, but underutilized, tools.