Most advisors agree that AI technology can provide efficiencies, whether it be through streamlining workflow or communicating more effectively with clients. However, the jury is out on whether these budding but costly investments are worth it.
“This technology is really expensive. I mean, hundreds of thousands of dollars a year for some of these [AI] vendors,” said James Bogart, CEO and president of Bogart Wealth based in McLean, Virginia, which works with AI vendors as well as building some advanced capabilities in-house. “So in order for me to make that level of change, there has to be a significant uptick in value creation, efficiency.”
A recent report by Deloitte projected enterprise software companies will approach a $10 billion run rate by the end of 2024, but that pales in comparison to the projected $1.6 trillion in global enterprise IT spending this year.
This comes at a time when advisors already have trepidation around the explosion of AI, especially after the U.S. Securities and Exchange Commission (SEC) cited two advisory firms on March 18 for allegedly making boastful claims that they were using AI-based investment decisions when they were not.
READ MORE: Beware, boasters: SEC challenges firms’ extravagant AI claims
“We’ve seen time and again that when new technologies come along, they can create buzz from investors as well as false claims by those purporting to use those new technologies,” SEC Chair Gary Gensler said in a news release about the scams. “Investment advisers should not mislead the public by saying they are using an AI model when they are not. Such AI washing hurts investors.”
The SEC also extended that warning to firms partnering with AI vendors.
Ways to measure AI technologies are difficult, but options exist
Since many AI vendors specific to wealth management are private companies and often new, it is harder to know their underlying profitability because they don’t report financials like public companies.
The accuracy of the AI in the tools being sold is also difficult to determine. That’s partly because the AI programs are still learning and are prone to so-called hallucinations or false outputs, as seen with the widely used AI-language model, ChatGPT.
READ MORE: Advisors know ChatGPT, but that doesn’t mean they trust it
One metric that’s mostly used by the AI developers and testers, called the F1 score, measures the precision and recall score of machine learning models. It essentially calculates the percentage of accurate responses against the volume of information received.
“So, if you’re looking to receive a response on 100 data points, and 94% came back perfectly accurate, then your F1 score is effectively going to be a 94%,” said Danny Lohrfink, co-founder and chief product officer at Wealth.com, a tech-based platform for estate planning for financial advisors based in Phoenix, Ariz. “So it’s a combination of volume and accuracy.”
Lohrfink said when Wealth.com first started developing AI technology within its trusts business, it had a F1 score of 71%.
“Now, we’re up around 90% accurate on everything,” he said, adding that it took about 18 months of teaching the AI model. “We’ve been able to train and condition the model and then collect more data and get better and better. So the score keeps going up.”
Advisors want AI for greater efficiencies, but at what cost?
When it comes to AI providers, the pricing structure is wide-ranging, with some charging advisors through a license or subscription-based model, or fees based on number of users or assets under management.
And the pricing structure is also evolving as some companies, like SalesForce, are moving away from a licensing structure into a cost-per-data usage in its cloud services.
“As we’re heading into this AI world, we’re starting to move more toward looking at allowing customers to have a pricing model where they look at tiers of usage and look at what they might consume from data cloud,” said Michelle Feinstein, general manager and vice president of global financial services at Salesforce, a cloud-based customer relationship management software provider in San Francisco.
Both Feinstein and Bogart agreed that if an AI provider can prove how their model can bring efficiencies in workflows and customer experience, it can be worth the price.
For example, Back Diamond Wealth Platform “has been a fantastic partnership. It’s also, hands down, my most expensive piece of technology,” Bogart said. “But the fact of the matter is that they provide a tremendous amount of value.”
When it comes to shopping for AI vendors as a firm with nearly $3 billion in assets under management, Bogart said his No. 1 question is: “Can you handle a firm my size?”
Regardless of whether a vendor builds the AI technology or the advisor in-house, it still costs. And the bigger the firm, or amount of data input required to train the AI, the more costly it becomes to develop these platforms.
“It’s like building a house, right? I don’t want to have a proposal and a budget [set] and then all of a sudden, because you didn’t know what you were doing, we’re two to three times that,” he said. “Because I’ve had that experience happen.”