What does Warren Buffett think of the investment philosophy of your family office? Unless you have the Oracle of Omaha on speed-dial, you probably don’t know.
In the age of artificial intelligence, however, that could be changing.
New AI functionality embedded in North Carolina-based Eton Solutions’ AtlasFive, a family office-focused enterprise resource planning (ERP) platform, now allows clients to use plain language not only to extract data from their investment and asset portfolios, but also to offer new investment insights drawing from outside information sources.
The recently launched EtonGPT is now available to more than 750 clients around the world, including Canadian family offices that use AtlasFive.
“The philosophy behind AtlasFive was to create an enterprise resource planning product using an integrated database model,” says Muralidhran Nadarajah, chief information officer, Eton Solutions. “For example, if a transaction comes in from the bank, it’s recorded in the investment ledger, but also creates journal entries in the accounting ledger to make audits and reconcilement of books easier. Based on the OpenAI GPT-4 engine, EtonGPT provides conversational AI and cognitive processing capability on top of that.”
Nadarajah notes that 80 per cent of a family office’s data is locked up in documents. That can make data analysis a challenging—and time-consuming—task. But EtonGPT promises to allow clients to securely interrogate their own private database to improve operational efficiency, parsing data buried in a wide range of document types, from trust and estate plans to contracts and partnership agreements.
For example, “some of our family office clients want to invest in commercial or residential rental properties,” Nadarajah says. “EtonGPT can go through every contract because there might be a poison pill somewhere in there. It can find tenancy agreements that are longer than expected, grandfathered agreements, rent increase protections, automatic renewals and termination clauses that are onerous. You can even analyze letters and email complaints to the landlord to identify high-maintenance tenants, turning a job that would have taken you 40 to 80 hours into an hour.”
In addition to eliminating time spent on repetitive tasks, EtonGPT can also scrape the Internet and other data sources on which it has been trained to extract insights from outside the client’s database and combine that information to offer business insights.
“If you’ve got Tesla stock, you can ask EtonGPT to compare the company with BYD Auto,” Nadarajah says. “But you can also ask for the latest news about Tesla, and what the analysts are saying about that stock.”
The company’s AI can also generate answers to more esoteric questions. Is an investment an ethical one, according to family office parameters that might include potential effects on climate change or ESG adherence?
Nadarajah says that asking novel questions of EtonGPT has also generated surprising—and potentially useful—results.
“We asked it to imagine a love triangle, with the balance sheet and the income statement both fighting for the attention of cash flow,” he says. “The balance sheet made some very nasty remarks about the income statement to win the affections of cash flow. By following the AI-generated discussion in this novel context, you can develop new insights.”
As intriguing as such functionality may be, AI is not about to replace humans in the family office. Oversight remains critical, and AI systems are far from ready to operate autonomously in key family office functions, says Dan Conner, a CPA and general partner at Ascend Venture Capital, an early-stage VC firm in St. Louis.
Having reviewed thousands of tech ventures and invested in multiple high-growth AI companies, he believes that the technology currently remains best suited to assisting family offices with investment analysis, portfolio allocation optimization, risk management and alternative data analysis.
“In some cases, the complexity of AI engines has already exceeded the capacity of their developers to grasp fully the reasoning and layers of abstraction behind computational model outputs,” Conner says. “Because AI capabilities show no signs of slowing down, this means it will be increasingly crucial for humans to be in the loop to ensure we arrive at the outcomes we want.”
In addition to securing sensitive data to prevent breaches and privacy infringements, he also stresses that AI models should screen out low-quality data inputs that can lead to erroneous decisions or financial losses.
Conner also points out that AI algorithms are subject to their own biases, potentially amplifying prejudices in data sources or swaying decisions inappropriately or unethically.
“This can be tempered by ensuring the data used to train the AI engine is representative of the ethical considerations unique to each family office and that the humans in the loop provide a diversity of perspectives to boost awareness of potentially problematic outcomes,” he says. “Regularly conducting model audits and analyzing outcomes are best practices to identify missteps and shortfalls before they perpetuate.”
Please visit here to see information about our standards of journalistic excellence.