Industries / Finance and business services / Finance and banking / Wealth management
Financial planning
Financial planning is helping a person or household set money goals and plan their saving, retirement income, insurance and the rest of their finances over a lifetime.
7
Milestones achieved, from 1969 to 2026.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1969
A profession is named
On 12 December 1969 Loren Dunton gathered 13 financial-services leaders in Chicago to sketch out a new profession. The College for Financial Planning, founded in 1972 by its own account, awarded the first Certified Financial Planner (CFP) marks in 1973, and the CFP mark is now the field's best-known title. The size of that first class is disputed: 42 people by the Financial Planning Standards Board's (FPSB) count, 35 by Lewis Walker's account in the Journal of Financial Planning. By FPSB's count, 236,300 people held the mark at the end of 2025 across 29 territories, led by the United States (107,529), China (32,599) and Japan (27,786).
financialplanningassociation.orgfinancialplanningassociation.orgcffp.edufpsb.orgfpsb.org
Credit
Loren Dunton; the College for Financial Planning; FPSB, formed in 2004
1994
The 4% rule
William Bengen, a planner, tested retirement withdrawals against US stock and bond returns since 1926 and published the results in the October 1994 Journal of Financial Planning. A client who took 4% of the starting portfolio in the first year, and raised that sum with inflation each year after, never ran out of money in less than 33 years in any past period, while starting at 4.25% could empty a portfolio in 28 years. The "safe withdrawal rate", the highest starting withdrawal that lasted through history, became a basic planning measure, and Morningstar's yearly retirement research still uses Bengen's set-up of fixed withdrawals that rise with inflation.
Credit
William P. Bengen
2004
Save more tomorrow
At a manufacturer near Chicago, a consultant's software told most workers to raise their saving to the maximum at once, and few would, as Richard Thaler told the US Congress on 10 March 2004. Thaler and Shlomo Benartzi instead had workers sign up in advance to send part of each future pay raise into their retirement plan; 78% of the workers offered the plan joined, and their average saving rate rose from 3.5% to 13.6% of pay over 40 months, by the authors' account in the Journal of Political Economy. Automatic yearly increases are now in US law: the SECURE 2.0 Act requires 401(k) and 403(b) workplace plans set up after 29 December 2022 to enroll workers automatically from 2025 and to raise the default saving rate by 1 percentage point a year until it reaches at least 10%, with some plans, including those of small and new businesses, exempt.
LessonA saving plan people keep to does more than a better plan they turn down.
Credit
Richard Thaler; Shlomo Benartzi
2012
Saving by default
The UK began phasing in automatic enrolment in October 2012: employers put eligible workers into a workplace pension, and workers may opt out. The minimum contribution is 8% of qualifying earnings, at least 3% of it from the employer. By the Department for Work and Pensions' count, the share of eligible employees saving into a workplace pension rose from 55% (10.7 million) in 2012 to about 90% (22.6 million) in 2025. The default matters because few people see a planner: in 2024 only 8.6% of UK adults, 4.6 million people, had received regulated financial advice in the previous 12 months, by the Financial Conduct Authority's (FCA) count.
LessonA default does the saving part of a plan for people who never see a planner.
Credit
the UK government; the Department for Work and Pensions, for the figures
2018
Fiduciary at all times
CFP Board, which sets the rules for the CFP mark in the US, was set up in July 1985 as the International Board of Standards and Practices for Certified Financial Planners. In March 2018 it approved a new code requiring CFP professionals to act as fiduciaries, meaning they must put the client's interests first, "at all times when providing financial advice", not only during a full financial-planning engagement as before. The code covered about 80,000 US CFPs, by InvestmentNews's count, and took effect on 1 October 2019; enforcement was pushed back to 30 June 2020 after requests from the Financial Planning Association and others.
investmentnews.comwealthmanagement.comfinancialplanningassociation.org
Credit
CFP Board
2021
The safe rate moves
Morningstar estimates a "safe" starting withdrawal rate each year with Monte Carlo simulation, which runs 1,000 random market paths for each mix of stocks and bonds; a rate counts as safe if at least 900 of the paths still have money left after 30 years. Its base-case figure was 3.3% in 2021, 3.8% in 2022, 4.0% in 2023, 3.7% in 2024 and 3.9% in 2025. Morningstar puts the 2025 rise down to a change in how it forecasts returns, and says its old method would have given 3.6%; in the same tests, flexible spending rules allowed starting withdrawals of up to 5.7%.
LessonA single "safe" number is only as steady as the return forecasts and method behind it.
Credit
Morningstar (Amy C. Arnott, Christine Benz, Jason Kephart, Tao Guo)
2026
AI plans, measured
Researchers at MIT and Stanford had 1,000 people write requests for money advice, put them to ChatGPT 5.2 and Gemini 3 Flash, and applied the answers to simulated lives from age 22 to 90, comparing the results with standard economic models of saving and spending over a lifetime. The AI advice moved simulated savers closer to those models on saving, spending and investing, but it adjusted spending poorly after income shocks and recommended too little gradual drawdown in retirement; prompts from users with low financial literacy led to nearly $50,000 (4.11%) less wealth at age 60 than prompts from high-literacy users. The study is a working paper, and MIT Sloan's release says more than half of adults in the US and UK have asked an AI for financial advice.
LessonThe same model gives better or worse plans depending on how the question is asked.
Credit
Taha Choukhmane (MIT Sloan), Matthew Akuzawa and Weidong Lin (MIT), Tim de Silva (Stanford)
Watch for
An agent inside planning software
RightCapital, a planning-software company, added an AI agent called Iris that checks client profiles for missing or inconsistent data and reviews cash-flow assumptions. Given a target "probability of success", the share of simulated market paths in which the money lasts, it proposes three ways to reach it, such as a later retirement or lower spending. RightCapital calls Iris the first AI agent of its kind built for financial planning and says every figure comes from its own calculation engine; InvestmentNews reported no outside test and said the claim "is likely to be debated".
Achieved whenAn outside reviewer runs Iris on plans with planted errors and publishes how many it catches, and how often its proposed changes hold up when rerun.
Claimed by
RightCapital, June 2026
Ten-hour plans in minutes
Conquest Planning, a planning-software firm in Winnipeg, Canada, says its new AI tools cut the time to build a plan "from an average of 10 hours to minutes", while the advice logic stays in "auditable, deterministic rules rather than general-purpose generative models". The tools were due out in April 2026. For scale, InvestmentNews cites Kitces research that a plan takes 15 to 25 hours of team time.
Achieved whenAn independent time study across several firms finds a large, lasting drop in plan-building hours with no rise in errors.
Claimed by
Conquest Planning (Ken Lotocki, chief product officer), March 2026
Open
01
An open household answer key
A public set of invented households, stored as plain data files, each listing ages, earnings history, savings, spending and planned retirement date, with worked reference answers to fixed planning questions: the yearly US Social Security benefit at claiming ages 62, 67 and 70; how many years a balance lasts at a given withdrawal and return; and the gap between expected retirement income and spending. A scoring script runs any AI model or planning calculator on the households and reports its error rate by question type. The results measure errors and recommend nothing to anyone.
The briefCredit · unclaimed02
A dated pension-rules test
A question set on public retirement rules in about ten countries, such as the US, UK, Japan, Australia, India and Germany, asking things like the state pension age for a given birth year or the earliest age to claim. Each answer is tied to an official government page and the date the rule took effect. Every quarter the set is run against AI models, counting wrong answers and, separately, "stale" answers that were right under a rule since replaced.
The briefCredit · unclaimed
Open 01 · the brief
An open household answer key
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public set of invented households, stored as plain data files, each listing ages, earnings history, savings, spending and planned retirement date, with worked reference answers to fixed planning questions: the yearly US Social Security benefit at claiming ages 62, 67 and 70; how many years a balance lasts at a given withdrawal and return; and the gap between expected retirement income and spending. A scoring script runs any AI model or planning calculator on the households and reports its error rate by question type. The results measure errors and recommend nothing to anyone.
Why it is still open
Public tests of AI on personal money use quiz questions or are run by sellers, and none lets an outsider rerun checked household numbers. Hean, Saha and Saha scored AI models on 554 financial-literacy quiz questions, which they answered about 68% correctly on average. Saturn, which sells AI tools to advice firms, tested 18 models on 121 money questions in September 2026 and "designed the methodology, applied the scoring and paid for the work", and no outside rerun has been reported. One advice firm, Mission Wealth, found that a client's AI tool said the client could retire while its own advisers concluded "the client could not retire".
arxiv.orginvestmentnews.comresultsense.cominvestmentnews.com
Built with what exists
Open Social Security, an open-source claiming calculator under the MIT license, as one reference; Python for scoring; models to test such as Claude Fable 5.2 (see the Note "Three working devices"), Inkling 2 Small, a model that checks its own work (see the Note "Checked without being asked"), and Mistral Large 4 run on a firm's own machines (see the Note "A trillion open parameters"). Frontier models now do multi-step arithmetic well enough that their errors are worth measuring at scale.
First step
Write 20 invented households, work out their benefits at 62, 67 and 70 by hand, check them against Open Social Security, and publish the files with any mismatches.
Done looks like
200 invented US households with at least three reference numbers each, every number worked out separately by two people with disagreements logged and settled; published error rates for at least three AI models; and an outside person who reruns the script and matches the scores within 1 percentage point.
Open 02 · the brief
A dated pension-rules test
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A question set on public retirement rules in about ten countries, such as the US, UK, Japan, Australia, India and Germany, asking things like the state pension age for a given birth year or the earliest age to claim. Each answer is tied to an official government page and the date the rule took effect. Every quarter the set is run against AI models, counting wrong answers and, separately, "stale" answers that were right under a rule since replaced.
Why it is still open
Hean, Saha and Saha warn that because models learn from past data "their advice can be outdated", but they tested general quiz questions, not rules by country and date. The rules keep moving: the OECD's Pensions at a Glance 2025 finds that under current law the average normal pension age for men rises from 64.7 for those retiring in 2024 to 66.4 for those starting work in 2024, as reported by European Pensions. This research found no public test of how often AI answers lag behind pension rule changes.
Built with what exists
Browser Use 1.0, a web agent, to collect, save and date official pension pages (see the Note "Deleting 80% to start over"); Gemini Decision 1, which returns probabilities for fixed-answer questions, to score answers cheaply each quarter (see the Note "A thousand questions, one pass"); the OECD report as a cross-check.
First step
Pick five countries, write 30 questions from their official pension pages, and record each page's link and the date it was read.
Done looks like
300 questions across ten countries, each with an official link and an effective date; results for at least three models with the shares of wrong and stale answers per country; and a second full run three months later with the change reported.