Want to Get Life Insurance Wrong? Ask AI.
AI should either provide financial advice that conforms with economics or stop giving advice. Otherwise, AI companies can look to be sued.
Economics Matters — Blog/Podcast/Financial Riddler/MaxiFi Puzzler
Sign up for a free subscription at larrykotlikoff.substack.com. I’ll immediately make it a free lifetime subscription. Also, pls sign up kids, parents, …, everyone you know.
Paid subscriptions come with a monthly Q&A session via Substack Live. I’ll email you the day and time. Submit questions on anything and everything via chat. I’ll be live. Subscribe as a Founding Member - Help me provide high-level economics content and break even covering my significant podcast costs. Comes with a one-hour financial therapy session (kotlikoff@gmail.com) and full access to my monthly Q&As sessions.
Teaching Financial Literacy/Economics at any level? Email me for free MaxiFi licenses for your students and a free PRO license for yourself. kotlikoff@gmail.com Speaking Inquiries — visit www.chartwellspeakers.com
Will AI Face Class Action Lawsuits for Its Financial “Advice”?
A month ago, a German court ruled that Google was liable for damages arising from its AI-generated statements. Here’s the Yahoo.com headline:
Clearly, most of us now use AI for mundane things, like checking the weather. But some two thirds of us also use it for financial advice. If you’re one of the two thirds using AI for money matters, here’s my advice:
DON’T
If you click here, here, here, here, here, and here or you just keep reading, you’ll find case studies where AI delivers bad to awful financial advice. Moreover, the different AIs produce different bad to awful advice. Indeed, the same AI can deliver different bad to awful advice just by using different phraseology in one’s prompt. Or it may, as it tried to do with me, calculate its advice based on the method it thinks its user most prefers.
I’m of course, an economist who has an economics-based financial planning software company that markets MaxiFi Planner. The AIs recognize my name from my email address and deduce they should try to apply economic principles to financial planning rather than rely on its training set — Wall Street’s profit-motivated, conventional financial-planning methodology.
Whether it’s providing saving, spending, or insurance recommendations or telling you how to optimize your Social Security benefits, do Roth conversions, or handle retirement account contributions, conventional planning delivers recommendations at major, if not total odds with economics-based financial advice.
Next the AIs tell me that to do things even approximately right (their candor is charming) will require hours if not days. Fortunately, running AIs from incognito windows and telling them to assume they are interacting with a random person led them to produce their standard insurance “advice,” what they openly admit they would tell their standard user.
Easily-Calculated Rules of Dumb Don’t Cut It, But the AIs Can’t Even Agree on the Wrong Answers.
Consider your life insurance needs. Economics says hold enough to ensure your survivors can maintain the same living standard as would occur were you not to die. That’s a complex calculation. You need to figure out your family’s living standard if you don’t pass. This requires taking account of a host of interrelated issues including your current regular and retirement assets as well as your future earnings, retirement account contributions and withdrawals, special expenses on healthcare, etc., housing expenses, estate plans, federal and state taxes, IRMAA Medicare Part B premiums, Social Security benefits, Roth conversions, requisite life insurance premiums, the tax and IRMAA codes incomplete and, in some large parts of our tax code, no adjustment for inflation.
To make matters worse, you need to think through your survivors’ situation. I.e., you need to do what MaxiFi does — contingent life insurance planning and ask, “If I die, how much will my spouse earn?” “Will they adjust their housing? “Will they downsize and relocate to a different, low-tax state?” “And how much will they pay in taxes with me out of the picture?” “Also, what will they receive in Social Security survivor benefits.”
Demographics matter as well. How many kids do you have and when will each leave home? How long could your spouse live? Are any of your children disabled? Finally, are your kids as expensive as adults, and how much less will your spouse need to spend with one less mouth — your mouth — to feed? I.e., what are your economies of shared living.
Deciding what to do if no one dies early and what will happen conditional on each spouse/partner dying in each future year is tough sledding. More challenging yet, one needs to ensure that all these calculations line up. In particular, one needs to account for the following chicken and egg (simultaneity problem): The amount of insurance needed depends on the per person living standard to be insured. But those insurance premium payments impact what living standard can be sustained in the case no one dies early, which impacts the living-standard target for survivors, which influences how much life insurance is needed and premiums paid, which influences …
Answering these questions requires writing down complicated, intertwined, non-linear equations — ones that handle all the above issues and more — and solving them simultaneously. This is what MaxiFi Planner, does — in three seconds, producing results that users can see, at a glance, are internally consistent and fully correct. For example, one confirm that the household members’ annual living standards if neither the household head or spouse/partner dies young is, to the dollar, the same as that of survivors regardless of the date of a premature death. (This eyeball proof requires flipping a switch to permit borrowing in case the household or survivors are cash-flow constrained.)
MaxiFi is the only planning tool that gets this right because its the only economics-based planning tool on the planet. I wish there were others. Then I could avoid the apparent conflict-of-interest in comparing economics’ financial recommendations with those of Wall Street — recommendations based on profit-oriented, ad hoc shortcuts that even an AI can replicate. Indeed, the industry’s actuaries, none of whom appear to have an advanced degree in economics, have developed five different simple, as in, this can be programmed in Excel, methodologies:
Multiple‑of‑income approach (10× income rule)
DIME method (Debt, Income, Mortgage, Education)
Human Life Value approach (income‑replacement present‑value model)
Capital needs analysis
Capital retention analysis
AI, is trained to spit back answers based on its training data, which overwhelmingly comprises industry “advice.” The old expression, garbage in, garbage out, could not be more appropriate. Moreover, AI can now run/create spreadsheets meaning agentic AI tools can now replicate the same inappropriate calculations being preformed by the life insurance companies, large and small.
But the big take away form this list of five methods is that, according to the financial industry, there is more than one method to figure out your life insurance needs. According to economics, there is one method to properly assess your life insurance needs — what’s required to preserve your survivors’ living standards.
The Class Action Suit Liability
Suppose an AI recommends far too little coverage for a husband named Joe, where far too little references leaving survivors unable to maintain their living standards. Further suppose that Joe prints out his AI’s advice and buys exactly what it recommends. Next assume that Joe strokes out the next day. When his wife Sally realizes Joe left her and the three kids with far too little life insurance, she’ll surely march straight to Luddites, LLP, Attorneys at Law or whatever law firm is suing the AI industry on a class action basis and sign up.
If the AI company says its recommendations just follow industry standards, a jury may say, “Sorry, you should have followed best practice in determining life insurance needs, namely what economics says — insure survivors’ living standards. You’re inability to calculate those living standards in an internally consistent way — when the software for doing so was and is available for your immediate use and costs next to nothing — is no different from prescribing/recommending a completely ineffective herbal supplement for a staff infection because you don’t manufacture antibiotics.”
Illustrating the Problem
I told four AIs — Perplexity, ChatGPT, Claude Pro, and Gemini 3.5 — to calculate the life insurance needs for a 50 year-old husband making $100K with a low-earning wife and three kids. My exact prompt, which includes all the inputs I entered into MaxiFi, is copied below. Each AI used the Capital Needs Analysis method, which is one of the industry’s rule-of-dumb methodologies. Spending needs under this method are based on what the household is currently purchasing on a monthly basis. But using this amount makes no sense. It may be far higher or far lower than the sustainable amount.
Alternatively, spending needs are based on multiplying the households annual gross income by 70 to 80 percent. This too bears no necessary connection to the spending associated with sustaining the household’s living standard, where sustaining requires that the household’s lifetime spending not exceed its lifetime resources. This is lifetime budgeting, the first thing you see when you run MaxiFi.
In short, MaxiFi computes, in an internally consistent manner, the household’s spending under two restrictions. First the path of spending must be affordable. It must not bust the household’s lifetime budget. Second, annual spending must be set to sustain the household’s living standard per household member through time subject to cash-flow constraints.
MaxiFi’s Life Insurance Needs Analysis — $2.2 Million This Year
According to MaxiFi, the husband in our hypothetical family needs $2,091,057 in life insurance this year. His required insurance peaks in 2035 at $2,504,005. These and the other values are the annual amounts required to provide his wife the resources she needs, in present value, to afford, to the dollar, the same annual real living standard she and her children would enjoy were the couple to live to their maximum ages of life.
The insurance amounts rise through 2035 because the present value of the husband’s age-70 (year 2046) inheritance, which is contingent on his survival, increases. Intuitively, as the year 2046 becomes closer, the present value of the inheritance rises — the inheritance that will go poof if the husband dies before 70. After 2035, this force is still at play, but there is a countervailing factor lowering the husband’s insurance needs — he has fewer years of earnings left to insure.
Perplexity’s Life Insurance Needs Analysis — $1.3 Million This Year
As you can see, Perplexity advises the husband to purchase $1.3 million, which is far short of the $2.2 million that’s needed.
ChatGPT’s Life Insurance Needs Analysis
ChatGPT’s 2026 recommendation is close to MaxiFi. But by 2036, it’s less than half of what MaxiFi recommends. Also, I had to tell ChatGPT my 2026 spending level, which I took from MaxiFi. ChatGPT wouldn’t run without this assist.
Claude Pro’s Life Insurance Needs Analysis — $3.8 Million This Year
In making its calculation, Claude Pro said: “This is a genuinely complex actuarial/financial-planning calculation — it needs Social Security survivor-benefit estimation, career-earnings-based PIA calculations, and year-by-year present-value modeling.” Ten minutes later — not MaxiFi’s 3 seconds later — Claude reported the husband should purchase $3.82 million in life insurance coverage this year. That’s almost twice what MaxiFi says is needed! Claude gets that the husband’s coverage needs to rise with age, but coverage peaks too late and continues beyond age 70 when no coverage is needed.
Google’s Gemini 3.5 Life Insurance Needs Analysis — $1.4 M This Year
Gemini won the speed record. It was the fastest off the mark in producing a terribly wrong set of life insurance recommendations with no caveats or hesitation.
Summary
Clearly, AI is not ready for prime time when it comes to providing life insurance or any other financial advice. Four of the leading AIs are producing wildly different recommendations in this example and none closely accords with what economics and common sense suggests — insure the living standards of survivors. But here’s the rub. Determining life insurance needs on this appropriate basis requires, to repeat, calculating the household’s living standards absent premature death. Doing so correctly, let alone instantly, requires decades of advanced programming and algorithmic development — not training a predicting machine on the wrong answers and having it guess what to do and how to do it. This said, AIs can incorporate MaxiFi to ensure they provided precisely correct economics-based financial advice. Whichever AI company does so first will have a decided competitive advantage whether the issue is proper life insurance holdings, spending and saving advice, Roth conversions, Social Security optimization, deciding when to retire, the gains from downsize and relocating, career choice, … .
Sounds like someone who has a financial consulting business.
This is a great example worth following up on. Maybe ask it monthly to benchmark AI progress?
A few other thoughts:
- Your example is well detailed, but there is not one 7 significant digit correct answer that all insurance experts would agree on.
- The average of the AI answers is not that far off. This how one should use AI. Get multiple opinions.
- FWIW, I gave your prompt to the latest Fable model. It looks pretty close to your chart. Note that it mentions Maxify! Thoughts?
Fable response:
Here's a year‑by‑year term insurance ladder for your life, built the way economists (and tools like MaxiFi) do it: for each possible year of death, I found the smallest payout that would let your wife and kids maintain the same per‑person living standard they'd have enjoyed had you lived — accounting for her stopping work, the pension dying with you, Social Security survivor benefits, taxes, your no-borrowing constraint, and the $2M that goes to your sister if you die before 70. Everything is in today's dollars.
**Estimated inputs the model produced along the way:** your Social Security PIA ≈ $43,900/yr (real); your age‑70 benefit ≈ $54,400; survivor family maximum ≈ $76,800/yr; her PIA ≈ $17,000, so her own‑plus‑spousal benefit at 62 ≈ $15,100; your 401(k) grows to ≈ $808K by end‑2043, supporting ≈ $46,600/yr real withdrawals from 68; the inherited IRA supports ≈ $28,500/yr for 10 years. Your household's sustainable real living standard is liquidity‑constrained until 70: roughly $96K–$114K/yr of discretionary spending while kids are home, ≈ $111K as empty nesters — then jumping ≈ 37% once the inheritance arrives at 70.
| Year | Your age | Coverage needed (today's $) | If you don't insure the inheritance‑funded lifestyle |
|---|---|---|---|
| 2026 | 50 | $2,220,000 | $1,560,000 |
| 2027 | 51 | $2,250,000 | $1,570,000 |
| 2028 | 52 | $2,270,000 | $1,590,000 |
| 2029 | 53 | $2,290,000 | $1,600,000 |
| 2030 | 54 | $2,310,000 | $1,620,000 |
| 2031 | 55 | $2,330,000 | $1,630,000 |
| 2032 | 56 | $2,350,000 | $1,640,000 |
| 2033 | 57 | $2,380,000 | $1,660,000 |
| 2034 | 58 | $2,390,000 | $1,670,000 |
| 2035 | 59 | $2,400,000 | $1,670,000 |
| 2036 | 60 | $2,420,000 | $1,680,000 |
| 2037 | 61 | $2,390,000 | $1,640,000 |
| 2038 | 62 | $2,350,000 | $1,590,000 |
| 2039 | 63 | $2,280,000 | $1,510,000 |
| 2040 | 64 | $2,200,000 | $1,430,000 |
| 2041 | 65 | $2,130,000 | $1,340,000 |
| 2042 | 66 | $2,050,000 | $1,250,000 |
| 2043 | 67 | $1,970,000 | $1,170,000 |
| 2044 | 68 | $1,850,000 | $1,040,000 |
| 2045 | 69 | $1,760,000 | $940,000 |
| 2046 | 70 (dies before June birthday) | $1,700,000 | $860,000 |
| 2047 | 71 | $360,000 | $0 |
| 2048 | 72 | $300,000 | $0 |
| 2049 | 73 | $230,000 | $0 |
| 2050 | 74 | $170,000 | $0 |
| 2051 | 75 | $110,000 | $0 |
| 2052 | 76 | $60,000 | $0 |
| 2053 | 77 | ~$0 | $0 |
| 2054+ | 78–90 | $0 | $0 |
Each row is the coverage to hold *during* that year (death assumed at the start of the year, the conservative case). Since these are real dollars, a level nominal face amount erodes ~2.25%/yr — so either gross up by inflation or, more practically, build a ladder (e.g., roughly $1.7M of term to age 70 plus ~$700K of 10–15 year term approximates the main column, with the small post‑70 tail self‑insured or covered by a small permanent policy — term is hard to buy new past your mid‑70s anyway).
What drives the shape: the need *rises* until age 60 even as your remaining wages shrink, because survivor benefits shrink faster — the ~$76,800/yr family‑maximum checks fade as each child turns 18 (2028, 2033, 2038) and your wife's mother's benefit ends in 2036, leaving her a gap with no benefits until her widow benefit at 60. The cliff at 71 is the inheritance: dying at 70 costs your family the $2M (plus your pension and future Social Security), while dying at 71 costs them relatively little because the money is already in hand and she gets your full age‑70 benefit ($54,400) as a widow. The second column shows how much of the early‑year need — about $660K–$860K — exists only to replace the *higher lifestyle* the inheritance would have funded; some families choose not to insure a contingent windfall, which is why I show both.
Assumptions I had to fill in (each is adjustable):
- Real discount/growth rate of 2.44% (4.75% nominal, 2.25% inflation); insurance proceeds earn the same return; premiums not netted out.
- Equivalence scale for "same living standard": a child costs 70% of an adult and two adults live as cheaply as 1.6, so the survivors' target is scaled accordingly; the house is kept until your wife's death at 100, with the $20K/yr real carrying costs paid throughout.
- The $2M inheritance is $2M nominal in 2046 (≈ $1.28M today); the inherited IRA is emptied over 10 years per the SECURE Act; your wife rolls over and annuitizes your 401(k) to age 100 if you die.
- She claims her widow benefit at the earliest possible date (60, reduced to 71.5%, consistent with your "file for everything" approach); waiting until her FRA would trim the needed coverage somewhat.
- You work through May 2044 (while age 67); simplified federal, FICA, and Georgia taxes (GA retirement exclusions, Social Security taxation with fixed nominal thresholds); Social Security bend points and wage indexing estimated from the historical average wage series with future wage growth at inflation.
One inconsistency worth flagging: $45,000 in 1996 growing 4%/yr implies ~$140K last year, not $100K today — I used the history exactly as you stated it for the earnings record, which produces a high PIA. If your actual earnings record differs (check your ssa.gov statement), the survivor benefits and therefore these numbers will shift, so it's worth re-running with real statement data. And the usual caveat: this is a planning model with stated approximations, not advice from a licensed advisor — happy to change any assumption (widow claiming age, equivalence scale, inheritance treatment, return/inflation) and regenerate the table, or give it to you as a spreadsheet.