Genuine versus Artificial Intelligence

AI Will Eat Conventional Financial Planning's Lunch, but Not MaxiFi's. When it Comes to Economics-Based Planning, AI is Hopeless and Truly Dangerous to your Financial Health.

Source: Fiona Wright

Economics Matters — Blog/Podcast/Financial Riddler/MaxiFi Puzzler

Sign up for a free subscription at larrykotlikoff.substack.com. I’ll immediately make it a lifetime subscription. But please support Economics Matters. Just unsubscribe and resubscribe here on a paid basis.

Paid subscriptions come with a one-hour, free financial consultation. Email me to find a time at kotlikoff@gmail.com. You can unsubscribe and resubscribe for free. Also, please sign up your kids, friends, colleagues, students, … !

Teaching Financial Literacy in High School or College? Contact me at kotlikoff@gmail.com. I’ll arrange free MaxiFi licenses for your students and a PRO license for you.

AI Can Do Conventional Planning in a Nanosecond? How Come?

I asked Perplexity AI and ChatGPT if conventional financial planning was using the same methodology it used 50 years ago. Both said yes and both graciously offered to build me a conventional plan. They said a growing number of households are using them, not financial planners, to do conventional planning. Why? Because they are quicker, as accurate, available 24-7, and free.

Why can LLMs replicate conventional financial planning? The reason is not that LLMs are so brilliant. The reason is that conventional planning (CP) is so simple. It’s simple because CP isn’t actually financial planning, at least not what economists recognize as such. CP was designed by the financial industry to gather AUM (assets under management) on which the industry charges fees. Thus, CP comprises a sales pitch, one with just enough financial lingo and mathematical veneer to convince households to hand over their money to be “managed?”

How CP Works

You meet with your planner. Their first question:

“How much do you want to spend in retirement?”

Your answer:

“A trillion dollars a day.”

They say,

“Sorry. Not even Elon can afford that. Let’s use the industry’s standard .80 income-replacement ratio to set your retirement-spending target. You ok with spending 80 percent of your pre-retirement income year in and year out in retirement?”

You say,

“Absolutely! Given how little I’ve saved, how little I save, and how conservatively I invest, I never thought I could spend, actually cruise, that much. ”

They say,

“Well, let’s make sure you can make your target.”

This is classic bait and switch. You’re baited with a target that’s generally miles more than you can safely afford. And then you’re switched — you’re told it’s your target.

“I’ll run you through my Monte Carlo simulator 1000 times. Each time, I’ll randomly draw annual investment returns and accumulate up and decumulate down your assets taking into account three things — your current assets, your pre-retirement saving, and your post-retirement targeted spending. I’ll then calculate the share of these simulations that succeed.”

Where’s the mathematical veneer? It’s in these words: income-replacement rate, Monte Carlo simulations, random draws, probability, and succeed.

“Oh, gee. Sorry. Your plan only has a 35 percent chance of success. The reason’s clear. You’re investing far too conservatively. But here’s the good news. I ran the simulator based on my managing your money. It shows a safe 85 percent success probability. How did I rescue your plan? I invested you in my preferred stocks. Stocks are, as you know, safe long term. So, not to worry. Check out cruises and leave the rest to me.”

More financial malfeasance. An 85 percent chance of success is an unmentioned 15 percent chance of failure, i.e., a 15 percent chance of ending up with zero assets at some point in retirement, including the first day you retire. Probabilities are averages. Take the number of times the roulette wheel lands on red 19 divided by the number of spins. That’s the probability of landing on red 19. It’s an average. But none of us bet based on averages. Elon and his ilk aside, none of us would take a 50-50 bet on winning or losing even $1K.

We are all risk averse for good reason. The pain from losing is far greater than the pleasure from winning. If we cared just about average outcomes, we’d never buy insurance. We’d say, “Sure our house can burn down. But not on average.” Finally, stocks aren’t safe long term. Yes, the longer you hold stocks, the higher the chance of a big final pot. But longer holding periods gives stocks more time to proverbially hit the fan — go to zero. If stocks were safe long-term, no one would buy long-term inflation-indexed bonds at the far lower safe rate they yield.

As for the “fact” that stocks have beat TIPs over all 30-year holding periods since 1935, that’s just more evidence that ‘figures lie and liars figure’. That statement doesn’t reference observations of 30 independent returns because the same annual returns are being used repeatedly. E.g., the cumulative stock returns between 1930 and 1960 and between 1931 and 1961 use 28 of the same annual returns — those between 1931 and 1960. Here’s the truth. We only have 3 independent obervations of 30-year cumulative stock market returns to consider over the past century. That’s 3 data points. A fourth 30-year cumulative stock return is the return on the Nikei between 1989 and 2019. In 2019, it was 55 percent lower in real terms than it was 1989.

Of course, none of this goes through your head. You’re already on the cruise, floating in the pool, four martinis to the wind. Hence,

You say,

“Sign me up.”

As described here, since CP doesn’t actually do financial planning, it doesn’t correctly answer any financial questions — be they how much to save, how to invest, what job to take, where to live, whether to rent or buy, how much life insurance to hold, when to retire, when to take Social Security, how much and when to Roth convert — you name it.

The Irony of Quick and Dirty

An LLM can estimate your pre-retirement income, multiply it by .80, and run Monte Carlo simulations in a nanosecond. It can also robo-invest your money based on achieving a 85 percent success probability. Thus, AI is about to bite Wall Street in the derriere — not just in providing what Wall Street calls financial advice, but in doing up to 40 percent of all adviser tasks.

Don’t take this from me. I just asked Perplexity AI, “Can you replace my financial adviser?” It answered, “In many areas I can effectively stand in for a traditional adviser on analysis, modeling, and what‑if strategy design.” I then asked “How many financial advisers will likely be replaced by AI?” Its answer: “A plausible medium term picture is up to 30 percent fewer advisers.”

Can AI Do Economics-Based Financial Planning?

Perplexity’s answer:

“Yes. AI can implement economics‑based (life‑cycle, consumption‑smoothing, utility‑maximizing) financial planning in principle and, in some niches, already does parts of it in practice.”

I’m not sure who trained LLMs to exaggerate its capacities. But when it comes to economics-based financial planning (EBP), LLMs are badly hallucinating as I’ll demonstrate below using my company’s MaxiFi Planner.

MaxiFi is the only software that does EBP. Yes, some CP tools, like Boldin, are now using EBP lingo, including “consumption smoothing,” as part of their pitch to sell their software and manage your money. But talking the talk and walking the walk are two very different things.

AI Can’t Do EBP. It Can’t Solve Extremely Difficult, Unfamiliar Math Problems

Economics-based planning references simultaneously solving the basic, highly detailed equations of household finance. These equations were developed by a Who’s Who of economists and one polyglot mathematician over the past century. The list includes Irving Fisher, John von Neumann (the polyglot), Oscar Morgenstern, Harry Markowitz, James Tobin, William Sharpe, Menahem Yaari, Paul Samuelson, and Robert Merton.

These economists tackled different problems. Fisher focused on consumption smoothing — saving enough to maintain your living standard in retirement. von Neuman and Morgenstern developed economics’ expected utility approach to uncertainty. Yaari clarified the economics of longevity risk, including the roles of life and annuity insurance. Markowitz, Tobin, and Sharpe considered static (households live for one period) optimal portfolio choice. And Samuelson and Merton examined optimal dynamic portfolio and saving decisions where households are making interconnected decisions over their lifetimes.

MaxiFi solves the equations of household finance simultaneously, which means all its answers are internally consistent. But it does so taking into account something the fathers of finance left out — our crazy complex fiscal system, with 500 plus federal and state tax and benefit programs, plus household-specific cash-flow constraints.

These factors make the elegant, if highly complex, equations of household finance extremely ugly, introducing, in mathematical terms, non differentiabilities, non convexities, and discontinuities. MaxiFi jointly solves these ugly equations in seconds. Its solutions are exact, unique, and immediately verifiable from its reports.

Yet the tool is highly user friendly and its findings make immediate sense. This, at least, is Bankrate’s view, which ranked MaxiFi the best financial planning tool of 2025. So, above the hood, we’re talking about just stepping on a gas peddle and steering. Below the hood, we’re talking about 33-years of actual intelligence — human intelligence — slowly but surely solving a problem that no one believed could remotely be solved.

Financial Planning Is Not a Guessing Game

MaxiFi computes its answers to the dollar using an array of pathbreaking algorithms to which LLMs don’t have access. Nor can they run MaxiFi with large data to approximate MaxiFi’s answers. Instead of admitting they can’t do EBP, they provide quick, piecemeal answers based on the wrong algorithms with no concern for internal consistency — answers that are, routinely, wildly off the mark.

In short, using LLMs to do EBP is like using a hammer to open a beer bottle. They are the wrong tools for the problem. Indeed, the best way for an AI company to do EBP would be to solicit data in its front end and use MaxiFi as its backend. Why guess at the wrong answer when you can instantly provide the right answer?

AI Is Miles Off the Mark — an Illustration

Meet John and Jane — a middle aged, upper-income, New Jersey couple with three children, ages 1, 3, and 6. The couple has a house with a mortgage, retirement accounts, 529 plans, special expenses, retirement plans, and the list goes on. Indeed, you can see all 28 pieces of the couple’s financial information that I entered into both MaxiFi and Claude. This may seem like a lot of inputs. But every household has a lot going on. This includes contingent plans — what will happen if one spouse dies. For example, were John to pass, Jane will stop working until the kids leave home.

To keep things simple, I assumed that the couple can earn a fixed real return on all its investments. John and Jane face considerable cash-flow issues. In the short term, they have three mouths to feed, three 529s to fund, a mortgage to pay off, alimony to cover, and retirement accounts to fund.

MaxiFi determines what the couple can spend each year to smooth its living standard without going into debt. Living standard references the couple’s discretionary spending per household member with the listed adjustments for the relative cost of children and economies of shared living. Discretionary spending excludes fixed spending on taxes, housing, 529 and retirement account contributions, life insurance premiums, federal and state taxes, IRMAA premiums, and the couple’s special expenses on alimony and out-of-pocket healthcare.

This consumption-smoothing problem is deterministic and unique. This means it only has one set of mutually consistent correct answers. You can see from MaxiFi’s reports that the program is finding those answers. An example is its lifetime balance sheet, which shows that the present value of John and Jame’s lifetime resources equals, to the dollar, the present value of their lifetime outlays — fixed and discretionary. Another is that the couple’s living standard rises as they become less and less cash-flow constrained, in their case, as the kids leave home, as they pay off their mortgage, and as they start receiving Social Security and retirement account withdrawals. More precisely, the couple’s regular assets always zero out before their living standard rises since bringing money into a period when your living standard is higher makes no sense.

So,

What is the couple’s discretionary spending this year?

MaxiFi’s correct answer: $43,347.

Claude’s incorrect answer: $53,546.

Claude’s too high by 24 percent. Worse, its advice will leave the couple unable to spend even $43,347 through 2030 when John is required to withdrawal the principal of his inherited IRA.

Here’s ChatGPT’s answer: “Their discretionary spending in 2026 is essentially ~$0, very tight, possibly slightly negative given the assumptions.” Here’s Perplexity AI’s answer: “Using the simplifying assumptions I coded, their 2026 discretionary spending comes out negative, about −134K for the household.”

—

What is the couple’s discretionary spending in today’s dollars in 2050?

MaxiFi’s correct answer: $102,562.

Claude’s incorrect answer: $137,491

Claude’s off by 34 percent.

How much life insurance should John and Jane hold this year to ensure survivors can enjoy, to the dollar, the same living standard?

MaxiFi’s correct answer: John needs $2.191 million, Jane needs $1.577 million.

Claude’s incorrect answer: John needs $1.410 million. Jane needs $1.355 million.

Claude’s too low by 55 percent on John’s insurance and 16 percent on Jane’s insurance.

How much does the present value of the couple’s federal plus state taxes change if John contributes to a Roth rather than a regular IRA?

MaxiFi’s correct answer: They fall by $9,463.

Claude’s incorrect answer: They fall by $111,000.

Claude’s off by 1,173 percent.

How much can John and Jane raise their lifetime discretionary spending if they both collect Social Security at 70?

MaxiFi’s correct answer: It rises by $361,180.

Claude’s incorrect answer — It rises by $156,481.

Claude’s too low by 43 percent.

I could go on. I could also compare MaxiFi with all other LLMs and find similar huge mistakes. And, I could compare MaxiFi’s investment advice, which is day and night different from AI’s because AI was trained to spit back CP investment “advice”.

Bottom Line

When it comes to financial planning, conventional planning is here to take your money and run. Likewise, AI is here to take conventional planners’ money and run. My advice to conventional planners— switch to MaxiFi. It’s what a century of economics recommends. It’s not built on conflicted, dangerous rules of dumb. It’s far cheaper. And it’s safe against AI. Most important, it fulfills your fiduciary obligation and will provide true job satisfaction in enriching, protecting, and improving your clients’ lives. At a minimum, use MaxiFi to check the plans you’re providing your clients. You’ll immediately see the enormous difference.

My advice for DYIers? Join the tens of thousands of ordinary Americans shelling out $109 and using MaxiFi on their own. Bankrate’s best financial planning tool of 2025 wasn’t “best for advisers,” it was best for individual households. As for households using advisers, make sure they run you through MaxiFi to make sure you are getting economics-based advice, not a well honed sales pitch.

If the above sounds like a sales pitch, it is. But there is nothing wrong pitching financial health at a trivial cost. And it’s coming from someone you can trust. I’m not rich, but I’m not, as are so many professionals, focused on money. Boston University’s salary is just fine. This is why I’ve never taken a penny in salary or other income from my company in a third of a century. My goal has always been and will always be to make MaxiFi affordable for individual households as well as advisers who wish to deliver appropriate financial solutions, not put their clients in harm’s way.

As for AI doing economics based financial planning, forgetaboutit. AI’s here to brag and produce terribly wrong answers to terribly important financial questions — about your saving, your spending, your taxes, your education, your career, your job, your investments, your life insurance, your retirement accounts, your retirement age, your Social Security decisions, and all else. When it comes to real financial planning, AI is, as illustrated, genuinely stupid and genuinely dangerous.

John and Jane

1. They live in NJ

2. John is 46, Jane is 36. John was born on 1/1/1980, Jane on 1/1/1990.

3. Their 3 children: Jane Jr., Jane Jr. Squared, and Jane Jr. Cubed, ages 6, 3, and 1.

4. John is self employed. He earns $100K, which he assumes will stay fixed in today’s dollars. He’ll retire on 1/1/2043 when he turns 63.

5. John’s first job, at age 22, paid $40K. His salary grew at 4% each year through last year.

6. John will take Social Security retirement benefits at age 62 and file for child benefits for Jane Jr. Cubed.

7. If Jane dies, John will hire an au pair for $50K real per year until Cubed turns 16 on 1/1/41.

8. Jane earns $100K, which she expects will grow 2% faster than inflation.

9. Jane’s first job, at age 20, paid $60K. Her salary grew at 4.5% each year through last year.

10. Jane will retire when she turns 70 and take Social Security then.

11. If John dies, Jane will stop working until 2046 after which she’ll earn $50K real through retirement.

12. They own a $750K house with $10K in annual property taxes, $7.5K in annual insurance, and $20K in annual maintenance. They will leave their house to their children.

13. They have a 15-year, $500K mortgage on which they pay $3,698 per month.

14. John is making tax-excludable $35K nominal alimony payments for another 10 years.

15. John inherited, 5 years ago, a now $250K IRA from which he is just pulling interest income.

16. John has a $35K nominal pension that starts when he turns 65. There is a 50% survivor benefit for Jane.

17. John has a $300K IRA to which he plans to contribute $7.5K real each year.

18. John has a $25K HSA to which he’ll contribute $3K real each year through retirement. Starting in 2050, John will withdraw $10K real per year from his HSA to help cover Medicare Part B premiums and out-of-pocket expenses. He’ll do this until he exhausts his HSA account.

19. Jane has a $25K 401(k) to which both she and her company contribute 4% of pay.

20. They project inflation at 2.5%.

21. They invest all their assets safely (in TIPs) earning a projected 2% real return.

22. They plan to spend only 75% of their retirement accounts subject to RMDs.

23. Like their parents, Jane Jr., Jane Jr. Squared, and Jane Jr. Cubed were born on January 1st. Each will graduate high school at age 18.

24. Jane’s parents will gift her $250K real in 2065. If she’s dead, nothing will be gifted to John or the children. Jane’s parents despise John and his issue.

25. Jane Jr.’s, Jane Jr. Squared’s, and Jane Jr. Cubed’s 529 plans are funded $75K, $50K, and $15K, respectively. John and Jane will fund $50K in real college payments for each child from 18 through 22. This funding will occur via fixed real payments in each year before the year each child enters college.

26. John is planning to pay $10K real for healthcare at ages 63 and 64 after which he’ll go onto Medicare. Jane will start Medicare at 70. Each spouse expects to pay $7.5K real annually for medigap policies.

27. John and Jane wish to maintain their living standard through their maximum ages of life — both age 100 — to the degree possible without going into debt. Living standard is discretionary spending per effective household member. Effective household members equal the number of adults plus .7 times the number of children in the household (i.e., under age 18) raised to the power .642. The power adjusts for economies of shared living. This formula implies that two adults can live as cheaply as 1.6.

28. Discretionary spending comprises all outlays apart from federal and state taxes, Medicare Part B premiums, housing expenses, special expenses (alimony), the above-specified healthcare outlays, 529 contributions, and retirement account contributions.