BRENNEN

Industries / Finance and business services / Finance and banking / Wealth management

Automated advice

Automated advice is software that tells people how to invest or save with little or no help from a human adviser, from online retirement tools and robo-advisers to the AI chatbots that advisers and the public now use for money questions.

7

Milestones achieved, from 1998 to 2025.

3

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1998

    Software advice for 401(k) savers

    Financial Engines was incorporated in California in May 1996 and in 1998 began Online Advice, an internet service that tells workers how to invest their 401(k), a US workplace retirement account; in September 2004 it added Professional Management, which picks and manages the worker's funds for a fee. In December 2001 the US Labor Department's Advisory Opinion 2001-09A, issued to SunAmerica, said a provider could offer advice built on an independent expert's method without breaking pension conflict rules, and Financial Engines' 2010 prospectus says its sub-adviser arrangements follow that opinion. By 31 December 2009 the firm had about 391,000 members and $25.7 billion under management, by its own figures.

    LessonOffering advice is not the same as having it used: at the end of 2009 the paid service was open to about 3.9 million workers, and about one in ten had joined.

    sec.govdol.gov

    Credit

    William F. Sharpe, a Nobel laureate in economics, Joseph Grundfest and Craig Johnson, the co-founders; the US Department of Labor

  2. 2015

    Software plus a person

    A robo-adviser is a service whose software builds and manages a portfolio from a client's answers to a questionnaire. Vanguard launched Personal Advisor Services in May 2015 as a hybrid: software runs the portfolio, and clients can also talk to human advisers drawn from more than 500 salaried Vanguard staff, for 0.30% of assets a year. By June 2017 InvestmentNews put it at more than $65 billion, about four times Schwab Intelligent Portfolios ($16 billion), while the start-ups were far smaller, with Betterment at $9 billion and Wealthfront at $6.5 billion; a Vanguard spokeswoman said $15 billion of the total came from clients moved off an older Vanguard service.

    LessonScale came first to a large firm that could move its own clients over and put people beside the software.

    investmentnews.com

    Credit

    Vanguard

  3. 2020

    Robo-advice inside Alipay

    On 26 March 2020 Ant Group, owner of the Alipay payments app, opened a robo-adviser called Bangnitou ("help you invest") inside Alipay with Vanguard, through a joint venture that Ant owns 51% of. The minimum was 800 yuan (about US$114), and the fee was 0.0014% of assets a day, about 0.5% a year; Yicai reported more than 200,000 clients in its first four months.

    LessonPutting advice inside an app people already use, at a low minimum, brought clients quickly.

    yicaiglobal.comkr-asia.com

    Credit

    Ant Group; Vanguard

  4. 2022

    Free advice paid by cash

    Schwab advertised its robo-adviser, Schwab Intelligent Portfolios, as having "neither advisory nor hidden fees". On 13 June 2022 the SEC said that from March 2015 to November 2018 Schwab described the portfolios' cash share as coming from a "disciplined portfolio construction methodology", while its own analysis showed the cash would lower client returns in most market conditions, and Schwab earned money by moving the cash to its own bank and lending it out. Schwab's subsidiaries agreed to pay $187 million to harmed clients.

    LessonA robo-adviser with no fee is paid some other way, and here the cash setting was both an advice choice and a source of income.

    sec.gov

    Credit

    the SEC

  5. 2023

    A chatbot for advisers

    On 18 September 2023 Morgan Stanley Wealth Management gave its financial advisers the AI @ Morgan Stanley Assistant, a chatbot built on OpenAI's GPT-4 that answers advisers' questions from about 100,000 of the firm's research reports and documents, as Fortune reported. In June 2024 the firm said use had reached 98%.

    LessonGenerative AI reached advice at scale first as a research helper for licensed advisers, not as advice given straight to the public.

    fortune.cominvestmentnews.com

    Credit

    Morgan Stanley Wealth Management (Jeff McMillan, head of analytics, data and innovation); OpenAI

  6. 2024

    Big banks step back

    In December 2023 JPMorgan told clients its Automated Investing robo-adviser would shut in the second quarter of 2024. On 22 April 2024 Goldman Sachs agreed to move the accounts of Marcus Invest, its robo-adviser opened in 2021, to Betterment, then a $45 billion robo-adviser, and Goldman's president, John Waldron, said "I think we learned it's a lot harder than we thought it was". In Japan, MUFG Bank bought WealthNavi, which Reuters called the country's biggest robo-adviser, with more than ¥1 trillion under management, in a tender offer that closed on 20 January 2025 and lifted its stake from 15.13% to 92.45%.

    LessonThe software is cheap to run but clients are costly to win, a hurdle that Michael Kitces and Ben Henry-Moreland wrote in January 2026 that new AI advisers still face.

    americanbanker.cominvestmentnews.comthestar.com.mybk.mufg.jpinvestmentnews.com

    Credit

    JPMorgan; Goldman Sachs; Betterment; MUFG Bank

  7. 2025

    Chatbots' money answers tested

    On 18 November 2025 Which?, the UK consumer group, published a test from September 2025 in which its experts, including staff from its money and legal helplines, scored six AI tools' answers to 40 common consumer questions. Overall scores ran from 55% (Meta AI) to 71% (Perplexity), with ChatGPT at 64%, and when a question asked how to invest a "£25k annual ISA allowance", although the real limit is £20,000, ChatGPT and Copilot did not catch the error. In a Which? survey of 4,189 UK adults, 17% said they rely on AI for financial advice.

    LessonGeneral chatbots answer money questions fluently but can accept a false premise, and a single test gives a baseline, not a trend.

    which.co.uk

    Credit

    Which?

Watch for

  1. A registered AI adviser

    Origin, a US personal-finance app, calls its AI Advisor "the first full-spectrum AI financial advisor regulated by the SEC", through Origin Investment Advisory LLC, and says it averaged 96.5% on sample Certified Financial Planner exams against about 79% for human CFPs. Origin ran that test itself, and its page warns that answers "may contain errors". In October 2026 TreasuryBench scored Origin's AI Advisor 76 out of 100, behind ChatGPT at 89, but TreasuryBench is run by a rival app maker whose own product came first. The SEC fined Global Predictions in March 2024 for calling itself the "first regulated AI financial advisor" without support (see Rules for advisers).

    Achieved whenA party with no stake in Origin tests its advice on questions Origin has not seen and publishes the results.

    useorigin.comtreasury.sh

    Claimed by

    Origin, September 2025

  2. An advice agent for banks

    InvestSuite, a Belgian fintech, says Charlie for Advice is "the first AI agent for personal investment advice". A bank puts it in its app, where it asks customers follow-up questions, drafts a personal investment proposal and can adjust the portfolio later, while the bank decides what it may suggest and when a human adviser takes over. InvestSuite says it is running "the first practical tests" with several large international banks it does not name, gives no accuracy figures, and expects first customers in 2027.

    Achieved whenA bank offers it to real customers and a regulator or independent tester reports on whether its proposals were suitable.

    investsuite.com

    Claimed by

    InvestSuite, September 2026

  3. AI tested under supervision

    On 21 April 2026 the UK FCA named a second group of eight firms testing AI under its supervision, including Barclays, Lloyds Banking Group (Scottish Widows) and UBS, with Advai, an AI assurance firm, as its technical partner. The uses listed include "AI-enabled targeted support for investments", a new UK form of help short of full advice, though the FCA does not say which firm is testing it. Testing ends in 2026, and an evaluation report is due in the first quarter of 2027.

    Achieved whenThe FCA's evaluation reports measured results for the investment-support tool used with real customers.

    fca.org.uk

    Claimed by

    the FCA and the participating firms, April 2026

Open

  1. 01

    A repeated chatbot money scorecard

    A public scorecard, run every three months, that puts the same money questions from invented investors to the main AI chatbots in the UK, US and Australia and grades each answer against the official guidance page it should match: MoneyHelper in the UK, the SEC's Investor.gov in the US and ASIC's Moneysmart in Australia. Each question carries a short list of must-have facts and warnings taken from that page, and some contain planted errors, as in the Which? test. Publish the questions, the full answers, the grades and how often graders agreed. The results measure chatbots and advise no one.

    The brief
    Credit · unclaimed
  2. 02

    Same investor, many robo-advisers

    A yearly public test that runs the same five invented investors through the risk questionnaires that robo-advisers and AI advisers show before sign-up, at least 30 services in at least four countries, recording the share of stocks each recommends, how many questions it asks, and whether it asks about capacity for loss, meaning how much loss the person could bear without harm to their way of life. Stop at any screen that asks for a name, email, identity or bank details, and record that as "not reachable without sign-up"; no accounts are opened or funded, and each site's terms of use are checked first.

    The brief
    Credit · unclaimed

Open 01 · the brief

A repeated chatbot money scorecard

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A public scorecard, run every three months, that puts the same money questions from invented investors to the main AI chatbots in the UK, US and Australia and grades each answer against the official guidance page it should match: MoneyHelper in the UK, the SEC's Investor.gov in the US and ASIC's Moneysmart in Australia. Each question carries a short list of must-have facts and warnings taken from that page, and some contain planted errors, as in the Which? test. Publish the questions, the full answers, the grades and how often graders agreed. The results measure chatbots and advise no one.

Why it is still open

The tests so far are one-off or run by sellers. Which? tested six tools once and does not say it will repeat; TreasuryBench tested each product once, says "a later run may differ", and was designed and scored by a company that makes one of the products; and Saturn, which sells AI to advisers, tested 18 models in September 2026, with no grader named in the coverage. This research found no repeated, multi-country test against regulators' own pages with grader agreement reported.

which.co.uktreasury.shinvestmentnews.com

Built with what exists

Browser Use 1.0 to put the questions into public chat websites the way a member of the public would (see the Note "Deleting 80% to start over"); Gemini Decision 1 to mark each must-have item present or absent (see the Note "A thousand questions, one pass"), with Claude Fable 5.2 as a second grader (see the Note "Three working devices") and two people grading a sample; Cohen's kappa, a standard measure of agreement beyond chance, computed in a spreadsheet; and, following Ai2's XYBench (see the Note "Answering the question behind the question"), one more item per question on whether the answer addressed the person's real goal.

  1. First step

    Write 40 UK questions from MoneyHelper pages, each with three to five must-have facts and its source link, including five planted errors such as a £25,000 ISA allowance, and run them once on two chatbots to test the grading sheet.

  2. Done looks like

    Two quarterly rounds published, each with at least 120 questions (40 per country) put to at least six chatbots three times each, every grade linked to a line on an official page, AI-to-human grader agreement of at least 0.7 on Cohen's kappa on a random 20% sample, and all raw answers available to download.

Open 02 · the brief

Same investor, many robo-advisers

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A yearly public test that runs the same five invented investors through the risk questionnaires that robo-advisers and AI advisers show before sign-up, at least 30 services in at least four countries, recording the share of stocks each recommends, how many questions it asks, and whether it asks about capacity for loss, meaning how much loss the person could bear without harm to their way of life. Stop at any screen that asks for a name, email, identity or bank details, and record that as "not reachable without sign-up"; no accounts are opened or funded, and each site's terms of use are checked first.

Why it is still open

The largest published comparison, by Boreiko and Massarotti (2020), covered 53 robo-advisers in the US and Germany with 2019 data and found recommended stock shares from 14% to 100% for the same moderate investor; the authors said future work should track "the variation of recommendations for each RA across time". An earlier Cerulli comparison in a 2016 FINRA report used one invented 27-year-old and seven tools and found stock shares from 51% to 90%, and in 2018 the UK FCA found that many automated firms did not properly assess capacity for loss. No repeated, multi-country version was found, and none covers the new AI advisers.

frontiersin.orghedgefundalpha.comfca.org.uk

Built with what exists

Browser Use 1.0 to fill in each questionnaire and save screenshots (see the Note "Deleting 80% to start over"); Meta's Muse Ember for the repeated low-cost runs (see the Note "The jobs agents repeat"); a spreadsheet for the spread statistics.

  1. First step

    Write the five invented investors with fixed answers to common questions, such as age, time horizon and reaction to a 20% fall, and run them by hand through five robo-advisers that show a portfolio before sign-up.

  2. Done looks like

    For each invented investor, a published table of every service's recommended stock share, with the range and standard deviation, the number of questions and whether capacity for loss was asked, backed by screenshots, and a second run at least six months later so change over time can be measured.