Strategic Acquisition Brief · Confidential
Google Finance left beta on June 25, 2026 after a ten-month rebuild — Gemini-powered research, Deep Search, scheduled agentic briefings, an Android app, live in more than 100 countries. And Google’s own help page says exactly what it is: it does not provide personalized financial, investment, tax, or legal advice; AI can make mistakes; verify independently. That disclosure is correct, and it is also a precise description of a ceiling. MaxiFi is the engine that removes it: for a household’s facts and assumptions it solves, not guesses, the lifetime plan, every dollar of taxes and benefits computed under current law. Deterministic, reproducible, auditable.
The new Google Finance is not a market-data page with an AI feature bolted on; it is a Gemini interface with market data attached. It shipped after a ten-and-a-half-month public beta, reaching more than 100 countries in April and Europe in May, with Deep Search, live earnings-call transcripts, prediction-market probabilities and an agentic system that delivers recurring briefings on a schedule the user sets.
Alongside it sits the standard language: no personalized financial, investment, tax or legal advice; AI can make mistakes; verify independently, and consult a licensed professional before making decisions.
A user who has received an automated briefing every morning for eight months has stopped reading it skeptically. That is the gap regulators tend to examine after something goes wrong rather than before — and the exposure scales with exactly the distribution that makes the product valuable.
The constraint is not Gemini’s reasoning. It is that a probabilistic model has no correct reference point, so no error in its output is decidable. Nothing in the model layer fixes that, however good the model becomes.
MaxiFi does not compete with Gemini and does not need a surface of its own. It is the computation service the assistant calls when a question has a dollar answer and a forty-year consequence. The user never sees it. What changes is what Google is able to say about the number it returns.
Search, Google Finance, the Gemini app, the Android app. Same interface, same conversational surface, same product velocity.
Gemini keeps doing what it does uniquely well: understanding the question, gathering the facts, explaining the result in plain language.
The rules, the solver, the audit trail. Same inputs, same answer, every time, traceable to the law tables in force on the day the plan was produced.
A class of money question moves from “informational purposes only” to an answer that can be reproduced and defended.
This is the architecture Google already applies wherever a wrong answer is unacceptable. Nobody ships a language model’s guess at a flight time, a unit conversion or a share price — those resolve against systems built to be right, and the model narrates the result.
Personal financial planning is the largest remaining category where that discipline has not been applied, and it is the one where being confidently wrong is most expensive to the household.
MaxiFi is the financial-planning platform of Economic Security Planning, Inc., built over more than three decades by Professor Laurence Kotlikoff of Boston University. It uses consumption smoothing and dynamic programming to compute the single, mathematically optimal lifetime plan — solving simultaneously across Social Security strategy, federal and state taxes, Roth-conversion sequencing, withdrawal order, life-insurance need, estate planning, and upside investing.
Goals-based tools and rule-of-thumb calculators answer “What is the chance you hit your number?” MaxiFi answers “What is the optimal path, and how much can I spend today without jeopardizing tomorrow?” It is not a better simulator. It is a different class of engine.
Prof. Laurence Kotlikoff — William Fairfield Warren Professor at Boston University; Harvard Ph.D.; former Senior Economist on the President’s Council of Economic Advisers; named by The Economist among the 25 most influential economists. He intends to stay on with the acquirer in whatever capacity best serves the product — architect, spokesperson, advisor.
MaxiFi’s economics build on Nobel-laureate work, and Nobel laureate Robert Merton teaches with MaxiFi at MIT Sloan as an “outstanding science-based lifecycle and retirement management platform.” Featured in Bankrate’s “Best financial planning software of 2025” roundup, cited as best for near- and long-term tax planning and the decumulation phase.
The moat is the rulebase as much as the solver: thirty years of encoded, continuously maintained federal and state tax, Social Security and benefit rules, carried under a regression suite re-run against every law change, plus patent-winning optimization algorithms built from economic theory rather than scraped text. The maintained surface is concrete: federal, Social Security, Medicare Part B and 42 state income tax codes, updated by the engineering team as provisions are released, on an annual law-update cycle. Stated plainly, because it will be checked: the solver is the replicable half — the mathematics is published, much of it by Kotlikoff himself. The rulebase is not, because encoding thirty years of law correctly is the decade.
A deterministic engine is the rare asset that gets more valuable the wider it is deployed, because the marginal cost of a computed answer is near zero and the value of correctness compounds with the number of households relying on it. MaxiFi already runs at consumer scale on ordinary inputs — ages, balances, wages, state of residence — which is precisely what an assistant can gather in conversation.
For a company that builds frontier models, the natural question is why any of this would be bought rather than built. The answer is the distinction between the two halves of what is on offer.
The solver is the replicable half. The mathematics of lifecycle consumption smoothing is published, much of it by Kotlikoff himself, and the patent has expired. A capable team could write one.
The rulebase is not. Thirty years of encoded federal tax law, Social Security provisions, Medicare rules and 42 state income tax codes — versioned, continuously maintained, and carried under a regression suite re-run against every legislative change. Not because the rules are secret, but because encoding them correctly and keeping them correct across three decades of legislative change is the decade.
Consider Intuit. Its enduring competitive advantage is not TurboTax’s interface or its AI features. Its moat is the tax-calculation engine. Large language models can generate plausible explanations, but they cannot reliably compute taxes, optimize outcomes, or produce audit-ready answers. Intuit can confidently deploy AI because every conversational interaction ultimately resolves against a deterministic rules engine designed to produce correct and defensible results.
The same principle applies to retirement and financial planning. Advisors and consumers will interact through increasingly sophisticated AI interfaces, but the value will reside in the analytical infrastructure beneath them. The AI asks the questions. The rules engine produces the correctly computed answer.
A study published in July 2026 prompted seven AI systems — ChatGPT, Claude, Copilot, DeepSeek, Gemini, Meta AI and Perplexity — with questions on emergency savings, asset allocation and retirement withdrawals, and found advice that could be inaccurate or demographically biased, varying widely by program. Stanford’s AI Index has separately identified financial context synthesis as among the hardest benchmarks for modern models.
That variance across engines on identical questions is the whole argument: the correctness cannot come from the model layer. Improving the model narrows the spread; it does not create a correct reference point.
FINRA’s 2026 Annual Regulatory Oversight Report identifies, as explicit risks of agentic AI: auditability and transparency — multi-step agent reasoning can make outcomes difficult to trace or explain; domain knowledge — general-purpose agents may lack what complex, industry-specific tasks require; and autonomy — agents acting without human validation. FINRA and the U.S. Treasury have since published an AI Lexicon and a Financial Services AI Risk Management Framework.
Google is not a broker-dealer and the disclaimer is doing real work. But the substance of what a user acts on is what ultimately gets examined, and an agentic briefing arriving every morning for a year is a different object from a search result. The exposure grows with adoption, which is to say it grows with success.
A correct-by-construction engine produces an answer that can be reconstructed and defended under the law in force on the day it was given. And because the engine is deterministic, the assurance can be underwritten — a bounded accuracy guarantee no probabilistic system can offer, because the warranted event cannot even be defined without a correct reference point.
It also starts from the defensible number: the most a household can safely spend with what it has, sustainable by construction — not an aspirational target.
The gap between a confident answer and a correct one is no longer a matter of opinion. It has been measured by independent researchers, published in a peer-reviewed journal, and reported by CNBC, Newsweek, Money and Quartz.
The Journal of Financial Planning (June 2026) put identical, detailed household scenarios to seven widely used AI tools — ChatGPT, Claude, Gemini, Copilot, DeepSeek, Meta AI and Perplexity — and asked two questions: do they give consistent recommendations to the same prompt, and are those recommendations consistent regardless of the user’s gender and ethnicity?
On the first, no. For one identical family, emergency-fund recommendations ranged from $19,500 to $37,500 — a statistically significant spread. Portfolio allocations differed significantly in equities, cash and alternative assets.
Nicolini, Cude & Chatterjee · Journal of Financial Planning 39(6) →
Holding every financial fact constant and changing only the described race or gender of the household head, some tools returned identical recommendations and others did not. One assigned a 75 percent bond allocation to an African American–led household while giving otherwise identical White-led households materially higher equity.
The retirement question is the sharpest case. Nearly every recommendation was the traditional 4 percent rate — and the only variation that appeared came from changing the household’s described race or gender.
For a regulated institution deploying guidance at scale, that is differential output from a process that cannot be traced. A deterministic engine is examinable by construction: every input that affects the answer is explicit, so when a variable moves the output you can see which one, and by how much. That makes fairness testable rather than asserted.
The authors measured consistency and fairness, and call for future work across larger sets of financial scenarios. Whether a recommendation is the economically optimal one for a particular household was outside their design.
That is the question a computed engine exists to answer. Change a fact that matters economically and the answer moves — by an amount you can inspect and reproduce. The paper publishes its prompts in full, so anyone can run the same household and compare.
A concrete, checkable example: AI engines trained before the One Big Beautiful Bill Act (enacted July 2025) told users the federal estate-tax exemption would “sunset” on January 1, 2026 — reverting to roughly half its level. In fact, the Act permanently raised the exemption to $15 million per person starting in 2026.
A model repeating pre-2025 training data would confidently tell a household to rush an irrevocable estate move it no longer needs — a costly, hard-to-reverse error delivered with total confidence. A computed engine, fed current law, does not carry stale assumptions forward as fact.
Asked to allocate a portfolio, Gemini alone declined — advising the user to consult a licensed professional rather than assigning percentages. That is the right instinct, and it is a credit to how the model has been built.
It is also an exact statement of the ceiling. The instinct to withhold is correct precisely because there is nothing underneath capable of producing a defensible number. Supply that layer and the instinct is no longer necessary: the assistant can answer, because the answer is computed rather than generated.
On the withdrawal-rate question Gemini did answer — returning the traditional rate, and varying it by a percentage point across demographic profiles whose finances were identical.
Larry’s Economics Matters Substack — 137,000+ subscribers — has run a six-post sequence testing named frontier engines against MaxiFi on dollar-specific household problems. Where the academic study measured variance, these show the dollar cost of it — dated, checkable, and using the same prompts a household would type into an assistant.
Acquiring MaxiFi acquires the megaphone these pieces ship from — pointed, with credibility no one in the category can match, at exactly the questions Google Finance now fields at global scale.
Durable value accrues to whoever owns the deterministic engine under the trusted interface — not to the interface, and not to the model. Google has built the interface and the model. The computation layer for personal finance is unowned, and there is one of it.
Personal finance is among the highest-intent categories in Search and the one where users most explicitly want an answer rather than a list. Being the only assistant that computes rather than estimates is a durable reason to ask Google the money question — and it is a claim OpenAI, Anthropic and Perplexity cannot truthfully make.
Every assistant claims helpfulness on money questions; none can substantiate accuracy. MaxiFi’s determinism makes a computational error objectively decidable, which is what turns a marketing claim into a warrantable one. That asymmetry is not copyable by a better model.
A correct-by-construction engine addresses the largest overhang on delivering money guidance at planetary scale, precisely as regulatory attention to AI advice rises. And there is exactly one MaxiFi — unowned, it remains available to every rival through the same API anyone can rent.
Academically canonical IP with clean provenance, built by a named economist at Boston University over three decades. There is one of it, it is not a category with alternatives, and it will be owned by someone.
The rebuild answered how well an assistant can research markets. It cannot answer whether a particular household should convert to a Roth this year, claim Social Security at 67 or 70, or draw from the taxable account first — because those have correct answers, and a model has no way to know it found one. There is an engine that does.
MaxiFi is being offered through a focused strategic process — the engine, its IP, and thirty years of R&D. The preference is an acquisition; that is where the strategic value sits. Continuity de-risks it: Larry Kotlikoff intends to stay on with the acquirer in whatever capacity best serves the product — architect, spokesperson, advisor. The next step is a 30-minute live demonstration: MaxiFi solves a real household’s plan while the leading models are asked to match it. The gap is the thesis. Evidence deepens with commitment — nothing is deployed, nothing left behind, and the full case is provable in an acquirer’s first quarter of ownership.