Build a chatbot for the work around advice, not advice itself: meeting preparation, notes and follow-up drafts, routine client questions, onboarding and internal knowledge search. Answer client questions from approved content with citations, pull account data from the source system rather than the model, and route anything resembling a recommendation to a licensed advisor. Recordkeeping, supervision and disclosure obligations generally apply to these conversations, so involve compliance before design starts. This is not legal or compliance advice.
Where does an AI chatbot fit in a financial advisory practice?
Advisors spend a large share of their week on work that is not advice: preparing for meetings, answering routine client questions, collecting documents, writing up notes and keeping the CRM current. A chatbot or agent can take much of that load. It should not give personalised financial advice, and a well-built one is designed so that it cannot.
That boundary is the core design decision. Anything that amounts to a recommendation for a specific client belongs to a licensed person, and the firm is responsible for what its tools say on its behalf.
What can a chatbot do for financial advisors?
| Use case | What it does | Who it serves | Starting autonomy |
|---|---|---|---|
| Client service questions | Answers 'where is my statement', 'how do I update my address', 'what are your fees' from approved content | Clients | Answers from approved documents; escalates anything else |
| Meeting preparation | Summarises recent activity, open tasks and past notes into a one-page brief | Advisors | Read-only |
| Meeting notes and follow-ups | Turns a transcript into notes, action items and a draft follow-up email | Advisors | Drafts for the advisor to edit |
| Onboarding and document collection | Explains what is needed, collects forms and chases missing items | Clients and operations | Runs the checklist; flags exceptions |
| Internal knowledge | Answers questions about firm procedures, product sheets and compliance policy, with sources | Advisors and staff | Read-only, cites sources |
| CRM hygiene | Updates contact records and tasks from emails and notes | Operations | Proposes changes; a person approves |
Meeting preparation and follow-ups are usually the best first projects. They save advisor time, stay inside the firm, and never reach a client without a person reading them.
Where is the line between information and advice?
| Acceptable | Should go to a licensed advisor |
|---|---|
| Explaining what a fee schedule says | Whether a client should change their allocation |
| Describing how an account type works in general terms | Which account type suits this client |
| Reporting a balance or transaction from the system | Whether to sell, buy or rebalance |
| Listing what documents a transfer needs | Predicting returns or promising outcomes |
| Summarising a client's own notes for the advisor | Anything referring to a client's goals as a basis for a recommendation |
Enforce the line in two ways: instructions in the prompt, and a check on the output that catches recommendation-like language and replaces it with a hand-off to an advisor. The hand-off should carry the conversation, as in how to escalate an AI agent to a human.
What compliance obligations does a financial advisory chatbot raise?
Rules vary by country and by how the firm is registered, so take advice from your compliance officer or counsel. This is not legal or compliance advice. The themes that come up repeatedly are:
- Recordkeeping: regulators generally expect firms to retain business communications. Chatbot conversations and generated client communications usually need to be captured and searchable.
- Supervision: the firm remains responsible for what its tools say. That means sampled review, escalation rules and an owner.
- Disclosure: clients should know they are talking to an automated assistant and how to reach a person.
- Marketing and performance claims: output that could be read as promotional should come from approved wording, not be generated freely.
- Privacy and data protection: client financial data is sensitive. See keeping sensitive data out of training sets and how agents handle regulated industries.
- Third-party and vendor oversight: where client data goes to a model provider, the firm needs to know the terms and be able to evidence due diligence.
How should the chatbot be built?
Answer from approved content
Client-facing answers should come from a controlled library of the firm's own documents - fee schedules, service guides, FAQs - using retrieval and citing the source, as in how a RAG chatbot cites its sources. When the library does not cover a question, the right answer is that an advisor will follow up.
Account data through read-only, authenticated calls
Balances and transactions come from the custodian or portfolio system through validated read-only calls for the signed-in client, never from the model's memory. Numbers are quoted from data, not generated.
Log everything
Store every conversation with the documents used and any hand-off, in the firm's own records system, with retention set by compliance.
How should an advisory firm start?
- 01
Involve compliance first
Agree the information-versus-advice line, recordkeeping and disclosure wording before design starts.
- 02
Choose one internal use case
Meeting preparation or follow-up drafts. Measure advisor hours spent today.
- 03
Collect real examples
Past meeting notes, typical client questions and the right answers, with client details removed.
- 04
Build the narrow version and test it
Score it on the examples, including questions designed to draw out advice it should refuse.
- 05
Pilot with a few advisors
Capture their edits, and review a sample of outputs with compliance.
- 06
Consider a client-facing assistant later
Only after the internal version is stable, with approved content, disclosure and escalation in place.
Before any client-facing launch, work through the AI agent go-live checklist.
What mistakes do advisory firms make with chatbots?
| Mistake | What happens | Instead |
|---|---|---|
| Starting with a client-facing bot | Unsupervised statements reach clients | Start with internal tools an advisor reviews |
| Relying on the prompt to avoid advice | A rephrased question gets a recommendation | Add an output check and a hand-off |
| Not capturing conversations | A recordkeeping gap | Retain transcripts in the firm's records system |
| Letting the model quote numbers | Wrong balances or fees | Look values up from the source system |
| Skipping compliance review | Rework or a regulatory problem later | Involve compliance from the start |
How can financial advisors use AI chatbots in 2026?
Use them for the work around advice, not advice itself: meeting preparation, notes and follow-up drafts, document collection, routine client questions and internal knowledge search. Answer from approved content with sources, take account data from the source system, keep a person responsible for anything resembling a recommendation, retain every conversation, and involve compliance before you design anything. Start internal, then consider a client-facing assistant.
If you are considering one, our AI agent development team can scope it with your compliance requirements in mind.
