You can find the original article here on WealthManagement. Subscribe to our free daily WealthManagement newsletter.
There is no escaping the fact that the use of artificial intelligence is increasing in all areas of the consultant’s workflow. This includes comprehensive areas such as investment management and portfolio analysis. Consultants are reportedly using tools like ChatGPT, Perplexity and Gemini in these spaces, which is raising alarms among some industry experts.
DeepWest, an AI-powered investment platform, recently conducted a six-week study, testing general-purpose AI models against themselves in 10 real investment workflows. General purpose models fail 85% of these tasks, producing incorrect calculations, incorrect data or no results. In contrast, DeepWest claims that its AI platform, specifically designed for financial workflows, has successfully performed tasks, matching “ground truth” or verified calculations.
DeepWest’s motivation in conducting the study was to show that its system was indeed different from the major language models. The company has market data on Snowflake, a cloud-based data platform, and has written more than 15 investment agents with access to more than 100 investment tools. LLMs are only used to summarize data and the results they feed into, not to do math.
You can think of DeepWest as a 24/7 agent tool for advisors to build portfolios, conduct individual stock research, select portfolios and create investment recommendations, said Toby Wade, CEO of DeepWest.
“Within 10 minutes, it can leverage Deep West’s agent investment framework to generate a complete proposal report for its advisors to win over those clients.” “But they can go from that to building a portfolio using DeepWest’s investment brain, where they don’t have to pay a model portfolio, model marketplace portfolio fee.”
Wade said there are significant efficiency gains from using AI platforms for investment tasks, but there are problems with advisers using general-purpose tools for those tasks.
However, not everyone is convinced. Mohan Naidoo, CEO of Alftaina, an AI-powered direct indexing platform, claims that the DeepWest study was not an apples-to-apples comparison, as the company compared its own, specially trained model to general LLMs. Naidu said that you would expect the custom model to perform better than the generic model.
Naidu is doing a lot with AI, but he has yet to let it make investment decisions. The problem, he says, is that it is “probable”.
“It’s not deterministic, which means it’s going to give you the best possible answer every time, and you’re never guaranteed to get the same answer, given the same information,” he said. “In investment decisions, in my book, it’s a big no-no. You can’t make a decision-making process that’s probabilistic. It has to be deterministic, meaning that if you have the same set of data, you should expect to get the same output. And AI isn’t like that.”
AI can help a portfolio manager or advisor reach a decision or provide information to help them, but it shouldn’t make the decision for them, Naidu said. With other use cases for AI, like meeting notes and CRM, it’s okay to have some non-zero tolerance. “There is zero room for error when it comes to investment decisions.”
That said, there are some investment workflows where AI is useful.
“The job that it’s really good at is explaining things, and what’s happening in the account or in the portfolio or different things an advisor should try or an advisor should try, or to shift the focus or attention of the advisor, the portfolio manager, to areas that deviate from the norm,” Naidu said.
Some advisors and portfolio managers spend a lot of time on these tasks.
“AI really helps the advisor and the user get to these things faster because it can really look at a lot of data and be able to process it and say, ‘Hey, out of the thousands of accounts you have, these 20 look unusual,'” he added.
Naidoo said AI could eventually get to the point where it does investment management, but asks, “Is that what we want?” One of the biggest benefits of AI is getting a different perspective.
“If everybody has the same opinion, you’re not going to get the return on investment there,” he said. “You’re just following everybody. Add little value. It feels like at this point AI can do pretty much anything you want it to do, given enough time and resources, but I don’t know if we want it to do with investment decisions.”
Ugur Hamaloglu, who leads EY America’s wealth and asset management consulting practice, said that analytical tools supported by generative AI that are adopted by the market can better serve advisers and make their lives easier.
But he acknowledged that AI today is not ready to directly serve up immediate investment decisions. Investment management is risky and still requires governance and human oversight.
One of the problems, he said, is that even with the right data, these models are still biased.
For AI systems that are trained on this topic and equipped with the right data, they can be useful for performing simulation analysis and stress testing.
“I don’t see a lot of risk because it’s insight for the advisor to take action, or there’s insight for the advisor to educate their clients because the job of a financial advisor is part knowledge, but part like teacher, mentor,” Hamaologlu said. “But making buy, hold, sell decisions, especially promising some financial outcome. … So trusting a computer system, no matter how good it is, without human oversight for these subjects, I don’t think is appropriate now.”
“Five years from now, we might look back and say, ‘Actually, they’re doing pretty well. Right now, we don’t have enough empirical data to say ‘they’re doing well’ or not.”
Hamaloglu said there are other fertile areas where asset managers can apply AI, such as customer acquisition. The risk is not as high in investment management, and this can speed up the advisor’s approach.
