AI Chatbots and Financial Advice: Useful Assistant, Risky Advisor
AI is great at compressing messy information into clean summaries. But as a source of financial advice, it can push clients toward risk they don't see coming.
Recent testing of popular LLMs showed a consistent pattern: portfolios skewed to higher risk, often overloaded with U.S. equities and influenced by whatever is trending. The answers sound confident and helpful. The underlying assumptions, not so much.
Where AI Helps (And Where It Doesn't)
A smart use case: using a chatbot to model mortgage repayments or compare term scenarios. One homebuyer fed in different rates and terms to estimate monthly costs and total interest, then manually checked the numbers. That's smart leverage.
The risky use case: asking for product picks, asset allocations, or tax positioning. Financial professionals report LLMs can misread local context, assume U.S.-centric growth rates, gloss over the "grey areas" in tax, and still present everything as if it's certain.
Why Chatbots Drift Into High Risk
- Hype bias: If AI and tech dominate headlines, expect AI to recommend AI-related equities.
- U.S. overweight: Many models tilt toward U.S. markets-sometimes aggressively.
- Overconfidence: The interface is friendly and decisive. That can lull users into trusting shaky assumptions.
- Black-box sourcing: Advice may echo non-expert content without clear attribution.
Regulatory Reality Check
LLMs are not licensed, not obligated to consider personal circumstances, and can invent details that sound plausible. For client-facing teams, that's a compliance trap.
For plain-English guidance on licensed advice and consumer protections, see ASIC's MoneySmart.
Practical Guardrails for Finance Teams
- Use for education, not advice: Summaries, concept explanations, draft client comms, comparisons to validate with primary sources.
- Lock in assumptions: When modeling, specify region, currency, tax status, inflation, fees, and rebalancing rules. Then verify with a calculator or spreadsheet.
- Demand provenance: Ask the model to list sources. Cross-check every claim against the original documents.
- No product picks from chat: Prohibit tickers, fund recommendations, and portfolio allocations without human review and documented methodology.
- Bias check: Cap country/sector weights and compare to a neutral benchmark. If the model goes 70-90% U.S. equities by default, flag it.
- Tax is nuanced: Route tax interpretations to qualified professionals. Use AI only to draft questions or summarize legislation for review.
- Keep audit trails: Log prompts and outputs. Add sign-off checkpoints before anything touches a client.
Safe "Jobs" for AI Inside a Finance Org
- Client explainers: Turn dense policy documents into plain-language summaries with links to the source.
- Scenario setup: Generate the structure of a model (inputs/outputs) for Excel or Python, not the final numbers.
- Policy Q&A: Internal FAQ drafts for product, fees, and processes-reviewed by compliance.
- Meeting prep: Condense research notes, earnings calls, or regulatory updates into brief outlines.
Red Flags to Watch
- Advice anchored to headlines or buzzwords
- Uncited claims and confident numeric outputs without method
- Cross-market assumptions (e.g., U.S. growth/fees/tax applied to Australia)
- Binary answers to grey-area questions (tax, disclosure, best interest)
A Simple Workflow That Works
- Ask: "Explain X in 150 words for a client with basic financial literacy. Include definitions. Cite sources."
- Verify: Check numbers and claims against primary documents, fund PDS, or regulator pages.
- Localize: Adjust for region, currency, tax, and fees. Note any assumptions in your client notes.
- Approve: Human review before client delivery. Archive the prompt-output pair.
Portfolio Construction: Keep Humans in the Loop
Use the model for idea generation and "what could go wrong?" lists. Keep allocations, risk budgets, and tax positioning under licensed, documented processes. If an LLM suggests a portfolio, treat it like an intern's draft: interesting, not authoritative.
Bottom Line
AI is a strong assistant for explanations and early-stage analysis. It's a weak substitute for regulated, context-aware advice. Keep your process accountable, your assumptions explicit, and your clients protected.
For a refresher on diversified vehicles, see MoneySmart on ETFs.
Further Learning for Finance Teams
Your membership also unlocks:
