Financial Product Guardrails¶
Status note¶
This standard is marked status: proposed because it addresses a domain (personal and
business finance tools) that carries real customer risk and likely warrants formal
review — probably legal — before any finance-related product ships. It should not be
elevated to a settled/approved status by an AI agent. Treat it as the working
requirement set until a human owner formally reviews it.
Why this standard exists¶
This venture's backlog includes multiple personal-finance and small-business-finance products (retirement planning, debt payoff, cash flow forecasting, pricing and profitability, and others). These products can influence real decisions about a customer's money. This standard sets the minimum bar for any product in that category, in addition to the general calculation and formula standard.
Requirements¶
- Clearly state assumptions. Every financial calculation must disclose the assumptions behind it (interest/growth rates, inflation, time horizon, contribution amounts, tax treatment, etc.) in a visible, editable configuration area — never buried inside a formula.
- Auditable formulas. All calculation-and-formula-standard requirements apply in full; financial calculations in particular must avoid opaque nested formulas given the stakes of getting them wrong. See calculation-and-formula-standard.md.
- Distinguish educational tools from personalized advice. Every financial product must clearly state that it is an educational/planning tool and not personalized financial, tax, or legal advice. This distinction should appear in the product itself (e.g., on the Start Here tab), not only in external marketing material.
- Avoid promising investment performance. No product may state or imply a guaranteed or expected rate of return, or otherwise promise investment outcomes. Where a rate of return is used as an input, it must be clearly labeled as a customer-adjustable assumption, not a promise.
- Avoid presenting tax/legal conclusions without review. Products must not present definitive tax or legal conclusions (e.g., "this is the correct way to file X"). Where tax or legal concepts are referenced, they should be framed as general information, with a recommendation to consult a qualified professional for the customer's specific situation.
- Include sensitivity analysis where useful. For products where the result is materially sensitive to an uncertain assumption (e.g., a retirement planner's assumed rate of return), the product should include a way to see how the result changes under different assumptions (a sensitivity table, scenario comparison, or similar), rather than presenting a single point estimate as if it were certain.
- Explain nominal vs. inflation-adjusted values where relevant. Any product projecting values over a meaningful time horizon must be explicit about whether figures are nominal or inflation-adjusted, and should ideally offer both views where practical.
- Document rates, dates, and compounding assumptions. Any interest, growth, or discount rate used must state its compounding frequency and the time basis it's applied over; any date-based calculation must state its reference date logic (e.g., how partial years are handled).
- Test edge cases. Financial calculations must be tested against edge cases: zero values, negative balances where applicable, very short or very long time horizons, and boundary conditions (e.g., retirement date in the past, payoff amount exceeding balance).
- Avoid real customer financial data in examples. Sample data must be synthetic. See ../ai/guardrails.md.
- Include appropriate customer-facing disclaimers before launch. Every finance-related product must include a disclaimer disclosing that it is not professional financial, tax, or legal advice, before it is offered for sale. The exact disclaimer language should be reviewed (ideally by qualified counsel) before first launch of any finance-related product; the requirement to include some appropriate disclaimer is not optional in the interim.
Relationship to AI companion content¶
If a finance-related product includes AI-assisted companion content (e.g., an AI-generated explanation or prompt pack), that content is additionally subject to ai-companion-content-standard.md, which prohibits presenting AI output as guaranteed-accurate advice in high-risk domains including finance.