If done right, AI could expand access to financial planning. - Getty Images/iStockphoto
A year ago, if you had asked me whether you should use artificial intelligence to help with your retirement planning, I would have said, absolutely not.
Artificial intelligence officially went mainstream in 2023 — a year that marked its leap from cubicles of tech geeks into the daily lives of average Americans. Since then, AI’s reach has grown rapidly. Today, average Americans are turning to AI for all sorts of everyday tasks and diversions: planning vacations, creating recipes out of ingredients in the fridge and pantry, writing college essays, producing videos, and creating action figures.
And it’s not just personal use. AI is increasingly embedded in the workplace, where workers — or at least those not yet displaced by AI — are using it for data analysis, content creation, vibe coding, customer service, and digital marketing
So if you asked me today whether it’s a good idea to use AI for retirement planning, my answer would be yes — provided you can become proficient in what’s known as prompt engineering. Prompt engineering is the practice of crafting clear, specific instructions or questions to get more accurate and relevant responses from AI tools like ChatGPT.
I say this for several reasons.
I’ve been experimenting quite a bit with AI and getting feedback from subject matter experts who are thoroughly impressed with the responses I’m getting back from AI systems like Gemini Deep Research.
For instance, I asked Gemini Deep Research to develop a retirement spending plan based on research by Michael Hurd and Susann Rohwedder, both of Rand, which showed that spending declines on a real basis 1–2% over the course of retirement, as well as research by David Blanchett, then with Morningstar, which showed expenditures decrease in real terms for retirees throughout retirement and then increase toward the end.
Afterward, I shared AI’s report with Hurd and Blanchett.
“This is terrific,” said Hurd.
And Blanchett said: “Looking at the analysis, it looks pretty accurate and straightforward… High level. I’d say I’m excited about the potential role of AI when it comes to helping people get more personalized advice/guidance around things like retirement versus what you might see reading an article online or something.”
Another reason to pay attention to AI in retirement planning comes from the work of Andrew Lo, a professor at MIT, and Jillian Ross, a doctoral student who collaborates with him. Together, they are exploring whether large language models like ChatGPT can provide reliable financial advice to everyday investors.
Early results suggest the potential is there.
In their paper Can ChatGPT Plan Your Retirement?, Lo and Ross argue that AI-powered financial advisers, often referred to as robo advisers, could represent the future of financial planning. However, they note that these tools still require significant improvement before they can truly replace human advisers. One key area of development is the ability for AI to exhibit more humanlike traits, such as empathy and an understanding of each person’s emotional relationship with money.
Lo and Ross also emphasized the importance of embedding ethical standards into AI systems. These standards should reflect a fiduciary duty, meaning the AI must always act in the best interest of the client, rather than in the interest of its creators or affiliated companies.
Democratizing financial planning
If these improvements can be achieved, the authors believe AI financial advisers could dramatically expand access to quality financial planning. By combining deep analytical capabilities with a more personalized, human approach, AI could help make professional-level advice available to people who might not otherwise afford it.
Looking ahead, Lo and Ross envision a transformation of retail investment where everyone with investible wealth can make optimal investment decisions aligned with their life goals – essentially a “full democratization of finance.”
And reason number three has to do with my discussions with James Mallory, a professor at the National Technical Institute for the Deaf, one of the colleges of Rochester Institute of Technology.
Mallory, who has degrees in computer science and electrical engineering technologies, demonstrated for me how someone could use AI to handle complex financial calculations that previously required extensive spreadsheet work.
“With AI, you could run a simple table or do a complex Monte Carlo analysis,” said Mallory. “You could have AI calculate RMDs based on your unique situation and show the actual mathematical formulas for doing so.”
During our Zoom call, Mallory shared how someone could upload financial documents in the form of PDFs, Excel spreadsheets, or even screen captures or images from reputable brokerage firms or financial institutions, then ask AI to analyze them and generate retirement scenarios.
The system could help evaluate a range of planning decisions, including:
– Optimal Social Security claiming strategies – Roth IRA conversion timelines – Tax-efficient withdrawal strategies – Healthcare cost projections
Mallory demonstrated for me how to develop a plan using AI that could achieve a sample client’s retirement goals. During our meeting we ran a typical client scenario to:
– Not run out of money before age 95 – Minimize required minimum distributions and maximize Roth conversions – Manage Medicare income-related monthly adjustment amount surcharges
Into AI, we plugged in facts and circumstances: account balances, income sources and amounts, planned withdrawals, and assumptions for portfolio returns, inflation, and taxes.
And then we made the request. We asked AI to generate a year-by-year table from current age to 95 showing: – IRA withdrawals – Roth withdrawals – Social Security income – Taxable account withdrawals – Remaining balances in each account – Estimated federal and state taxes paid annually – RMD projections once applicable – Highlight if/when funds may be depleted – A plain-language summary – Output in a downloadable Excel file and a Word summary – Use clear, short column headings – Follow plain text formatting (no bold, no hyphens, use commas instead of dashes)
It could even run a Monte Carlo simulation at any confidence you wanted for success probability.
Suffice it to say, AI generated exactly what we wanted.
AI, said Mallory, is like having a hammer. “I can build a school or a church with it, or I could hit somebody over the head with it,” he said. “The hammer’s not bad or good, it’s just a tool, and it depends on what you are going to do with it. AI is the same way, and I’ve been playing around with it. With AI, we could run our own Monte Carlo analysis. We could figure RMDs, put in different numbers, dump tables, spreadsheets, etc.”
Mallory is completely trusting of the computed results in ways I never was a year ago, when AI would typically get much wrong with respect to retirement and financial planning. He doesn’t feel the need to talk to a subject matter expert for most of these kinds of routine things as long as the user has a basic knowledge to create and verify the input prompting questions and the outputs are at least “in the ballpark” with what is expected to verify its accuracy. He doesn’t need a sounding board for this, although he strongly suggested running the ideas by an investment professional before any big decisions are made.
“So you’re 99% trusting of it, if not more?” I asked Mallory. “In financial planning, yes,” Mallory said.
And why is that the case?
“For a couple reasons,” he said. “One, I have found that none of the financial experts from different reputable firms agree on exactly how things should be allocated. One expert would say there is too much cash invested, where another may say it’s fine for derisking. Some recommend an S&P 500 SPX rated portfolio with a percentage in bonds and cash, others believe in an equal weight S&P 500 fund with no cash and a percentage in bonds. For example, some say 80/20, some say 60/40, some say 60/30/10. To a lay person, that conflicting advice can be confusing.”
“And two,” he said, “AI can tap into the entire world’s body of financial knowledge and expertise to answer virtually any question someone might have.”
In that sense, Mallory sees it as more reliable than any single human adviser or firm, since it can compare and synthesize a range of expert perspectives and recommendations.
Mallory describes himself as a “common user in the financial sector,” but emphasized that getting useful financial guidance from AI requires effective “prompt engineering,” which means knowing how to ask the right questions in the right way. This, he believed, is where the value of a financial expert would be beneficial, designing the prompts.
“It’s all about prompt engineering,” the professor said. “For example, you might tell the AI, ‘I want you to act as an expert in this area. Think through what an expert would do. What guidance does the government provide? What are the required minimum distribution rules at age 72 or 73? What are the formulas involved?’”
Use selectively and with caution
Does all this mean that there’s no use for financial advisers in the future?
Not necessarily. Despite his enthusiasm for AI tools, Mallory acknowledged its limitations and emphasized that he still values expert guidance for the larger picture.
“It’s not a replacement for an adviser clearly, but it takes a lot of the heavy lifting and tedious financial calculations away,” he said.
Mallory offers one important caution around privacy and security. He strongly advises against uploading any documents that contain account numbers, Social Security numbers, or other sensitive personal information. For birthdays, he recommends using only the month and year, and substituting fake names whenever possible. Even better, you can simply tell the AI the person’s age along with a placeholder name.
While large language models, or what many refer to as LLMs, may claim they don’t store your data, the information still passes through the cloud at least once, which raises concerns. Many paid AI platforms promise not to use your data to train their models, but free versions may use your inputs to improve their systems, something users should be aware of.
