A financial advisor charges about $6,815 a year to build the plan in this article. The average fee sitting on top of that is 0.96% of everything you have invested, every year, whether the market went up or not. On a $311,580 portfolio compounding at 7%, that fee alone takes $198,869 off your side of the table over twenty years.
I ran the same job on Ling-3.0-flash-Fin, the finance model Ant Group released free through its InclusionAI lab. 43 months of one household's raw records: bank exports, investment statements, a mortgage, two credit cards, and 18 reporting periods with holes in them.
It came back with the balance sheet to the cent. $709,279.03 in assets, $241,040.92 in debt, $468,238.11 left over.
Then it answered the question people actually pay for. When can this family stop working?
It gave two dates. 37 months, and 57 months.
Same family, same day, same arithmetic. The 20 months in between are the house they live in, and I have never seen a plan that puts both numbers on the same page.
The whole prompt that did it is further down, ready to paste into a free endpoint tonight.
By the end of this you will know:
- the one line that splits your money into what you can spend and what you cannot
- how it backed out real spending when nobody had logged a single receipt
- the two numbers that decide whether a household bends or breaks
- the four reporting periods it refused to average, and why that matters more than the answer
- the exact prompt, copy and paste
- what it will not do, and where a human still has to sign
The household is synthetic, built for the test. The arithmetic is not.
The 20 months live in the house
This family spends about $32,574 a year. Under the 4% rule that means they need roughly $814,352 before the money can carry them.
Count everything they own and they are 57.5% of the way there. About 37 months left.
Count only what they can actually sell and draw from, the $50,085.88 in cash and the $261,495.09 in investments, and they are 38.3% of the way there. About 57 months left.
Same family. Same day. Same math. The gap is the roof over their heads.
A house you live in shows up as wealth on every net-worth tracker ever built, and it pays you nothing. To get money out of it you sell it, downsize, or borrow against it, and you still need somewhere to sleep. Home equity belongs in a retirement number only when the plan actually includes selling.
Neither view is wrong. Showing one of them alone is.
What it started with
Household finances almost never arrive as a clean ledger.
Daily bookkeeping falls apart within weeks, and a single unrecorded month makes an expense log useless. Account balances say nothing on their own, because a family can hold cash while carrying a mortgage, a car loan, and credit-card debt at the same time. FIRE stays abstract, since the 4% rule only becomes a real target once you pin down spending, investable capital, contributions, and returns.
The test case was a two-income US household: 18 irregular reporting periods spread across 43 calendar months, January 2023 through July 2026, with bank and investment balances, salary income, some large expenses, internal transfers, a residence, a mortgage, and credit cards.
That is the shape of real records. Not a spreadsheet, a pile.
Solving for the spending nobody wrote down
When daily spending is missing, you can back into it:
Implied routine outflow = total income minus known large expenses minus change in net worth
That number is not an audited expense figure, and the model said so. Market moves, property revaluation, debt changes, internal transfers, and unclassified items all have to be pulled out first. What the method really does is show you which pieces are still missing, without anyone typing in a grocery receipt.
Over the timeline the picture moved hard.
Gross assets rose from $502,950.00 to $709,279.03
Liabilities fell from $283,300.00 to $241,040.92
Net worth went from $219,650.00 to $468,238.11, up $248,588.11, or 113.17%
The vehicle loan went from $12,400.00 to zero
The mortgage came down from $268,500.00 to $239,260.00
Ending position: $50,085.88 in cash, $261,495.09 in taxable and retirement investments, a home valued at $397,698.06, and $1,780.92 sitting on credit cards.
It reconciles exactly:
$709,279.03 minus $241,040.92 = $468,238.11
That single line is what separates gross property value from real equity, and investable money from money nailed to the ground.
The two questions that matter
Out of the whole balance sheet, two questions carry the weight.
How long could the family live on liquid reserves alone. And how much of their wealth is actually available to invest.
The answer to the second one is $311,580.97, cash plus portfolio. Total debt sits at roughly 34.0% of gross assets. Both numbers tell you more than any bank balance, because they measure whether the household bends or breaks when something goes wrong.
Assumptions, out in the open
The FIRE scenario ran on:
monthly spending of about $2,714.51
monthly contributions of about $5,681.40
annual spending of about $32,574
a 4% target of about $814,352
a 7% default annual return
Four of the reporting periods were thin, so the cash-flow figures came from the 34 better-documented months and went in as scenario inputs, not as audited 43-month averages. The model flagged that itself rather than smoothing it over.
That is the part worth stealing from this whole exercise. A paid plan hands you one number and buries the assumptions in an appendix. This one refused to average four months it did not trust, and told me which four.
Then it built the dashboard
The output came out as an interactive HTML dashboard: a net-worth timeline, asset and liability views, a cash-flow chart, a risk panel, explanations in both technical and plain language, and sliders for contributions, expected returns, and retirement spending.
The sliders are the part that matters. Drag the return assumption from 7% to 5% and the retirement date moves in front of you. A static report hides its assumptions inside a PDF. This one puts them under your finger.
The prompt, ready to paste
This is the whole thing. Open Ling-3.0-flash-Fin on OpenRouter, export your accounts however they come out, drop them in, and paste this underneath.
You are my household financial analyst. Here are my records, exactly as they came out of the bank and the brokerage. The periods are irregular and some of them are incomplete.
Step one, build the balance sheet.
For every reporting period, give me gross assets, liabilities and net worth. Separate cash, investable assets and the residence. Show the reconciliation line: assets minus liabilities equals net worth. Report to the cent.
Step two, solve for the spending nobody wrote down.
I have no expense log. Back into it:
implied routine outflow = total income minus known large expenses minus change in net worth
Strip out market moves, property revaluation, debt changes, internal transfers and anything unclassified before you do. Then tell me what the number still does not include.
Step three, answer the only question that matters. When can I stop working?
Use the 4% rule. Target equals annual spending divided by 0.04.
Give me the answer twice. Once counting everything I own. Once counting only what I can sell and draw from. State both dates. Do not average them and do not pick one for me.
Step four, put your assumptions on the table.
List monthly spending, monthly contributions, annual spending, the 4% target and the return you used. If any reporting period is too thin to trust, name it, leave it out, and tell me you left it out. Do not smooth over a gap.
Step five, build me an interactive HTML dashboard.
Net worth timeline, assets and liabilities, cash flow, a risk panel, and sliders for contributions, expected return and retirement spending, so the date moves when I move an assumption.
Rules for you. Never hand me a number you cannot trace back to a record I gave you. Where you had to infer, label it as inference. You are organising and checking, not advising.
What the $6,815 is actually buying
The retainer is the visible price. The fee on top of it is the one that compounds.
The average advisory fee in 2026 is 0.96% of assets under management, charged every year regardless of what the market did. On this household's $311,580.97 of investable capital, at their own 7% assumption, that fee takes $198,869 over twenty years and $561,901 over thirty. Those two figures are derived from the numbers above, not quoted from anyone.
None of that means an advisor is worth nothing. It means the organising, the arithmetic and the assumption-checking, which is most of what the first meeting is, now costs zero and takes one pass.
The same thing works for a small business
With decent records, the identical workflow organizes inventory, sales, expenses, and bank data into multi-period cash-flow and profit views. It tracks receivables, fixed costs, and how long the runway actually is. It flags cash pressure before it turns into a problem, and it explains movements in gross margin, depreciation, taxes, and working capital in language a founder can act on.
All of it still depends on the quality and the accounting treatment of what you feed it.
If you want to run it
The model is Ling-3.0-flash-Fin, built by Ant Group's InclusionAI lab on top of Ling-3.0-flash.
124 billion parameters with 5.1 billion active, a 262,144 token context window, and up to 32,768 tokens of output. The context matters here: 43 months of statements go in without splitting them.
The free endpoint is inclusionai/ling-3.0-flash-fin:free at openrouter.ai.
Zero per token, and rate limited: 20 requests a minute, 50 a day on a fresh account, 1,000 a day once you have ever put $10 of credits on it. For one household's records you will not come close to the ceiling.
What it will not do
A finance-tuned model makes structured analysis faster and cheaper. It holds no credentials and it cannot make a decision for you.
Changing your allocation, selling a home, refinancing, buying insurance, retiring early, all of that runs through risk tolerance, job security, taxes, family, and priorities that no model can see. Returns and inflation are uncertain. The 4% rule is a planning heuristic, not a law.
So use it the way it works best. Let the model organize the records, check the arithmetic, expose the assumptions, and generate the questions. Let a qualified human validate the data and make the call.
The prompt is above. Your own statements are already sitting in a folder somewhere. That is the entire setup.
