How Iris works

Iris is a language model, not an advisor. Iris holds no insurance licence, signs nothing on your behalf, and makes no decision that binds an insurer. When you need a licensed person, we hand over to one: see the last section.

We keep Iris on a tight leash and test it continuously, but a language model can still misread a question or paraphrase a clause loosely. The document that governs is always your policy wording and contract, never an answer in a chat window. For anything that matters (a limit, an exclusion, a waiting period), check it against your own document, or ask us.

The data Iris can read today covers travel and health insurance only. For vehicle or property, Iris has nothing to read and will say so rather than guess.

For a question about a specific insurer or plan, the system searches Insurnao’s verified store: policy wordings, benefit tables, eligibility rules and exclusions published by the insurers themselves, extracted by us and checked back against the original brochures. Iris answers benefits, limits and exclusions ONLY from those passages. For general questions (insurance concepts, Vietnamese insurance law, how buying on Insurnao works, or fixing a problem on the platform), Iris explains from general knowledge and always says that details vary by insurer and plan.

A hard rule in the system: Iris may not invent a phone number, email, link, form, opening hour, procedure or processing time. If something is not in the verified data, the correct answer is “I don’t have that”, and that is what Iris is built to say.

Many original wordings are written in USD. Every figure Iris gives you is converted and shown in VND: you should never have to do the exchange-rate maths yourself.

Underwriting belongs to the insurer. If you ask “will I definitely be accepted?”, Iris will not answer yes or no: that answer is not ours to give.

Any figure Iris gives is indicative, for comparison. The binding premium is set by the insurer against your actual application.

Iris lays out the verified facts and puts plans side by side so you can compare. Iris does not rank them as “better for you”: judging personal suitability is for a licensed adviser, or for you.

If you ask whether some condition, medication or history needs declaring, the answer is always yes. Under-declaring is the most common reason a claim is refused later. Iris is not permitted to suggest leaving anything out, even if you ask it to.

Those questions fall outside what Insurnao does, and Iris will say so rather than answer for the sake of answering.

Your message is checked first on our servers in Vietnam. A GENERAL question with no personal information (for example “what is a maternity waiting period?”) is sent as written to a language model hosted abroad to draft the reply, through providers configured not to store it or train on it.

If a message contains personal information (a name, phone number, email, ID number, date of birth, or your or a relative’s health), those words are NOT sent. The system sends only a short fixed topic label (for example “waiting periods”), or Iris answers entirely on our side.

Iris is built not to ask for your name, date of birth, phone number, ID or passport number, or medical records in the chat. When those are genuinely needed for a quote or an application, we move you to the appropriate secure form.

The full data picture: Privacy & Data.

Saying “I want to talk to a person” is enough. Iris also offers to hand over when it can see you have asked more than once without getting an answer. That offer appears only during office hours: 9:00–18:00, Monday to Friday, Vietnam time; outside them we tell you the hours rather than make a promise we cannot keep.

We don't run a call centre for quotes or advice. Reach us on the Support page, at hello@insurnao.ai, or on the Insurnao Zalo OA; we reply within 1 to 3 business days. If Iris ever gives you a phone number, that is a bug, so please tell us.

Send us the question and the answer you got through the Support page. We read every one of these: if it is a data error we fix the data; if it is a behaviour error we add a rule and a test so it cannot come back.