ProductUse casesPricingBlogContact
Dashboard Sign in Start free
Artificial Intelligence in Business

Voicebot in Banking and Insurance: What It May Answer

Voicebot in Banking and Insurance: What It May Answer

A voicebot in banking and insurance can safely answer general questions about products, fees, documents and procedures, and it can route callers to the right team. It must not give personal investment, credit or insurance advice, and it should not touch account data unless the customer’s identity has been checked properly. This guide sets out what such a bot may say, how identity checks work and where the line of regulated advice runs.

A customer asks a bank’s assistant late in the evening: “Which savings account should I open, and should I move my pension money into funds?” The first half is a product question with a factual answer. The second half is investment advice, which in the EU only a licensed adviser may give, after assessing the person. A well-built bot answers the first half and politely declines the second.

What a voicebot in banking may answer

The safe zone is everything that is the same for every customer and already published by the institution. Answers come from approved content: product sheets, fee tables, terms and FAQ pages. Typical questions:

  • Product facts: what an account or policy includes, what documents are needed to open it, how to apply.
  • Fees and terms: read from the current published fee table, with a link to it, never from memory.
  • Procedures: how to report a lost card, how to file an insurance claim, what to send after a car accident, where to find a form.
  • Service information: branch hours, contact channels, how to book a meeting with an adviser.
  • Explaining terms: what a deductible is, what “variable rate” means, how a grace period works in general.

Everything outside the approved sources should be out of scope. The article on keeping an AI bot inside its scope explains how to set this up so the bot declines rather than improvises.

Identity checks: when the bot needs to know who is speaking

As soon as an answer depends on the customer’s own account, policy or claim, the bot needs a verified identity. A general question (“How do I change my address?”) needs none. A personal one (“Has my claim been paid?”) does.

In phone banking and insurance lines, identity checks are part of the core system: the caller is verified through the institution’s own authentication, and EU payment services rules require strong customer authentication when a customer accesses a payment account online or starts an electronic payment. A voicebot connected to such a line works inside that process; it does not invent its own checks.

A few rules apply to any bot in finance, whatever the channel:

  • Never ask for a PIN, full card number, password or one-time code in a conversation. Say so in the bot’s answers, because fraudsters imitate bank assistants.
  • Do not reveal whether a person is a customer before verification.
  • For anything account-specific, send the customer to the logged-in app or online banking, where authentication already happens.
  • Keep personal data out of the bot’s knowledge base. The guide on what never belongs in bot sources lists the usual mistakes.

A related risk is fake bots. Customers should know where your real assistant lives and what it will never ask for; the piece on telling a real bot from a fake gives wording you can reuse.

Regulated advice the bot must not give

The hardest boundary is advice. In the EU, investment advice is a regulated service under MiFID II, Directive 2014/65/EU: a personal recommendation must be based on an assessment of the client’s knowledge, financial situation and objectives. A bot that says “this fund is right for you” crosses that line.

Insurance works in a similar way. The Insurance Distribution Directive (EU) 2016/97 requires distributors to identify the customer’s demands and needs before a contract, and a personal recommendation counts as advice. Credit decisions follow their own creditworthiness rules.

In practice, the bot should decline and redirect in these cases:

  • “Which fund, share or product should I choose?”
  • “Is this policy enough for my family?”
  • “Will I get the loan?” or “How much can I borrow?” beyond a published calculator.
  • “Will my claim be accepted?”
  • Tax consequences of a specific decision.

A good decline is short and useful: the bot says it cannot recommend a product for a personal situation, explains who can, and offers the way to book that person. The same pattern is described for legal services in the article on enquiries without legal advice.

Telling customers they talk to a bot

Under the EU AI Act, people must be told they are interacting with an AI system unless it is obvious, and this transparency duty applies from 2 August 2026, as set out in Regulation (EU) 2024/1689. A voicebot should say it at the start of the conversation, in one plain sentence.

Website voice assistant versus phone voicebot

Many banking voicebots run on phone lines connected to core systems. That is a large integration project with authentication, call routing and recording. A simpler first step is a voice and chat assistant on the website that answers the general questions above from published content and sends personal matters to the app or an adviser.

Botino works in that second way: the voice assistant runs in the browser on your website, answers only from the content you approve and replies only in the languages you configure. It does not run phone lines, and it does not access accounts or policies.

Checklist before launch

  1. List the questions the bot may answer, and the sources for each.
  2. Write fixed declines for advice, credit decisions and claim outcomes.
  3. Decide which requests go to the app, which to an adviser, which to the claims team.
  4. Add a security line: what the bot will never ask for.
  5. Check the AI disclosure at the start of every conversation.
  6. Have compliance review the answers before launch and after every product change.

FAQ

Can a voicebot explain an investment product?

It can describe published features, costs and documents of a product. It must not say whether the product suits a particular person, because that is a personal recommendation.

Can a website bot check a claim status?

Only if it is connected to a system that verifies the customer first. A bot that answers from public content should send the customer to the logged-in portal or the claims team.

Who should approve the bot’s answers?

The product owner for accuracy and the compliance team for regulated wording. Repeat the review whenever fees, terms or products change.

Start your voicebot in banking or insurance with the published questions only, write the declines for advice before anything else, and add personal services only when proper identity checks are in place.