What problem does it solve?
Clients think in goals: retire at 60, pay for a child's university, buy a second home. Turning those goals into a plan means collecting a lot of information, running projections across several scenarios and then explaining uncertainty in a way a client understands. If a firm measures how much of that effort is data gathering and document assembly, a full written plan can look costly to provide for smaller client relationships.
Language models are good at the conversation and the explanation, but they are not a reliable source of exact arithmetic and they cannot see everything in a complex personal situation. The design question is how to use them for what they are good at while the numbers come from an engine the firm can audit.
How does it work?
- Gather goals and facts. A conversation, with the client or the advisor, captures goals, timelines, income, assets, liabilities and constraints, and flags what is missing.
- Run the engine. A planning engine the firm can audit (rule based cash flow projections, or Monte Carlo scenario models with recorded assumptions and a fixed seed) calculates the plan with documented assumptions for returns, inflation and taxes.
- Explore what ifs. The client or advisor asks "what if I retire two years later" and the assistant reruns the engine and compares the scenarios.
- Explain in plain language. The model narrates the results, the trade offs and the uncertainty, quoting numbers only from the engine output.
- Advisor validates. The advisor reviews the plan, adjusts it for what the model cannot see, and signs it off before it is presented as advice.
- Audience
- Employee facing
- Autonomy
- Copilot
- Adoption
- Emerging
- Channels
- Internal tools, Web chat, Mobile app
What is it worth?
Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.
No public deployment has disclosed a measurable outcome yet.
Value drivers: Employee productivity, Inclusion and access, Customer experience, Revenue growth.
Indicative value
A wealth manager with 200 financial planners
USD 800,000 to USD 6 million
Value of planner time released per year per year
How this is calculated
Formula: planners * plansPerYear * hoursSaved * hourlyCost. The low scenario uses every low input, the high scenario every high input.
| Input | Low | High | Basis |
|---|---|---|---|
| Financial planners planners, planners | 200 | 200 | The reference firm. |
| Plans or plan reviews per planner per year plansPerYear, plans per planner per year | 50 | 100 | Editorial assumption, replace with your own planning volumes. |
| Hours saved per plan on data gathering, scenarios and write up hoursSaved, hours per plan | 1 | 2 | Editorial assumption, replace with your own. No public source on this page states a time saving for AI assisted plans. |
| Fully loaded planner cost per hour hourlyCost, USD per hour | 80 | 150 | Editorial assumption, replace with your own fully loaded cost. |
What it leaves out: Leaves out the larger but less certain effect of serving clients who did not get a written plan before, the cost of the planning engine and the review time that remains.
Who already uses it?
2 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.
Vanguard
United States · Wealth and asset management · 2020
Vanguard Digital Advisor is an all digital advice service that gathers a client's goals, time horizon and risk tolerance, uses an algorithm to build a portfolio for them, and keeps it on track with ongoing monitoring. It rebalances when a portfolio drifts more than 5% from the recommended allocation and adjusts holdings when a client adds goals. Vanguard discloses limits of the automated assessment, for example that it does not assess the suitability of selling existing holdings, and it labels its projections and goal forecasts as hypothetical and educational, not guarantees. The service was live by 2020: a Vanguard page archived in December 2020 refers to Digital Advisor clients who enrolled before 1 October 2020.
No outcome disclosed.
CIMB Niaga
Indonesia · Banking · 2026
CIMB Niaga, one of Indonesia's largest banks, built purpose built AI agents with its AI Center of Excellence and Artefact on Google Cloud. The agents help bank staff offer tailored advice and proactive guidance matched to a customer's financial goals and life stage. No outcome figures were published.
No outcome disclosed.
How do you implement it?
A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.
Data you need
- A validated planning engine with documented capital market and tax assumptions
- Client fact find data from CRM and account aggregation
- Approved explanations of key concepts (risk, sequence of returns, inflation)
- Disclosures and plan templates per market
Systems to integrate
- Financial planning engine through an API
- CRM and account aggregation
- Document generation for the plan report
- Advisor desktop or client portal
Complexity: Medium
If the firm already runs a planning engine, the work is connecting the conversation to it, keeping every number from the engine, and designing the advisor review so plans remain advice the firm stands behind.
- 1
Keep the maths in the engine
Connect the assistant to the existing planning engine through an API and forbid the model from calculating projections itself. Every number in the narrative must come from an engine call.
- 2
Start with advisors, not clients
Use the assistant to prepare fact finds, scenarios and plan drafts for advisors first. Client self service comes later, once explanations and refusals are proven.
- 3
Write the explanation library
Agree plain language explanations of assumptions and uncertainty with compliance, so the model narrates within approved wording.
- 4
Define what the assistant must hand over
Complex situations (business owners, cross border tax, estate structures, vulnerable clients) go to the advisor with a summary rather than an automated plan.
- 5
Test scenario consistency
Build test cases where the right answer is known and check that the assistant calls the engine correctly, reports the numbers exactly and discloses the assumptions.
Guardrails
- Numbers only from the planning engine, never generated by the language model
- Every assumption disclosed and reproducible in the plan output
- Projections labelled as illustrations, not promises
- Handover of complex or vulnerable client situations to an advisor
- Advisor sign off before a plan is presented as advice
KPIs to instrument
- Time from first conversation to signed off plan
- Share of plans where the advisor changed numbers or assumptions, and why
- Number of clients with a current written plan
- Engine call errors and narrative number mismatches found in tests
- Client satisfaction with plan explanations
Human in the loop
The advisor reviews and signs off every plan, can override assumptions with a recorded reason, and owns the advice. The planning engine's assumptions are approved and reviewed periodically by an investment committee or equivalent.
Common failure modes
- Model does the maths
- The narrative contains a number that no engine call produced. Enforce engine only numbers and check the output automatically.
- False precision
- A single projected value is presented as a promise. Show ranges and scenarios with the assumptions.
- Complex cases forced through
- A business owner or cross border case gets a generic plan. Detect complexity early and hand over.
- Guidance crossing into advice
- A client facing version starts recommending products. Keep product recommendations in the advised process with suitability checks.
What are the risks and rules?
EU AI Act
Depends on design
Planning support for advisors is not listed in Annex III. A client facing version must disclose that the client is talking to AI (Article 50). It becomes high risk if it is used to assess the creditworthiness of individuals (Annex III point 5(b)) or for risk assessment and pricing of life or health insurance for individuals (Annex III point 5(c)).
Rules that apply
Guidance
- Harnessing AI in the Financial Planning Profession (CFP Board, North America). Scenario based report (October 2025) on how AI may change financial planning by 2030 and what planners should do now.
- ESMA public statement on the use of AI in the provision of retail investment services (European Securities and Markets Authority, Europe). Applies MiFID II conduct and organisational duties when AI supports investment advice, including transparency to clients about its use.
- PS22/9: A new Consumer Duty (Financial Conduct Authority, Europe). The policy statement sets rules for four outcomes under the Duty, including a consumer understanding outcome, which applies directly to how projections and uncertainty are explained.
Controls to put in place
- Planning engine and its assumptions inventoried, validated and version controlled
- Automated check that narrative numbers match engine output
- Standard disclosures on every projection
- Advisor sign off recorded for each plan
- Periodic review of explanation wording by compliance
Frequently asked questions
- Can a language model do financial projections?
- It should not. Use it to gather information and explain results, and let a planning engine the firm can audit do the calculations, so every number is reproducible and auditable. Vanguard's Digital Advisor, for example, uses an algorithm to build and rebalance portfolios for a client's goals, and labels its projections and goal forecasts as hypothetical and not guarantees.
- Does AI replace the financial planner?
- Not in the design on this page: the assistant prepares and explains, and the planner validates and signs off. Fully automated services exist, such as Vanguard's Digital Advisor; Vanguard's Personal Advisor offers ongoing financial planning and access to an advisor for those investing $50,000 or more. CIMB Niaga's agents are described as helping bank staff give tailored advice and proactive guidance.
- Is a client facing planning assistant regulated advice?
- It depends on what it does. Explaining concepts and running projections is usually guidance. Recommending specific products to a client is investment advice under MiFID II and needs the full suitability process. ESMA's 2024 statement says MiFID II conduct duties still apply when firms use AI, and the firm stays responsible for the outcome.
How to cite this page
Blits.ai AI Use Case Library, "AI assistant for goal based financial planning", last verified 27 September 2026, https://www.blits.ai/ai-use-cases/goal-based-financial-planning-assistant. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 27 September 2026: First published