AI Assistant for Life Insurance
From «Call Me» to an instant quote. Between the customer's interest and the first concrete answer sat a phone call and office hours — not any more.
Summary
An insurer operating in Portugal had the sector's usual path on its website: anyone interested in life cover found, essentially, a «Call Me» button. They left a number and waited. BigLearn replaced that step with an AI assistant that talks, gathers what the quote needs and returns a result in the same interaction.
Client anonymised. This describes a delivered project; we do not publish conversion or revenue figures here, for the same reason we do not publish them elsewhere — we cannot show you where they came from. What is described is what changed in the process.
The cost was not the call, it was the gap before it
Someone shopping for life cover usually wants three things quickly: what it might cost, what the options are, and whether it fits their situation. The old path was website → callback request → wait → phone call → information gathering → quote. Several steps stood between interest and the first concrete answer.
In a digital channel, that gap is the problem. Insurance research often happens at 11pm, at the weekend or on a public holiday — precisely when nobody is on the phone. Making someone wait until the next working day introduces friction at the moment of highest intent, which is also the moment a competitor is one search away.
A conversation can ask what a form cannot
The path became website → AI assistant → conversation → details gathered → quote, all in one interaction. The assistant explains the process, answers the first questions and asks for information as it goes: age, sum assured, term, whether it is tied to a mortgage.
The advantage over a form is not cosmetic, it is one of order. A form has to show every field up front, including the ones that do not apply to this customer. A conversation asks only what it needs at that point, and what it asks next depends on what it just heard. The customer also does not need to know the insurer's internal vocabulary to answer.
A chatbot answers. This one executes.
This is the difference between bolting a chatbot onto a website and applying AI to the sales process. A chatbot replies «yes, we offer life insurance» and stops. This assistant understands, asks, gathers, structures, triggers the quoting process, presents the result and routes it onward.
What the customer experiences as a conversation becomes, on the inside, organised information: age, sum assured, term, type of need, mortgage, contact details, the answers the calculation requires. The AI stops being a question-and-answer interface and becomes a step in the process.
Business knowledge stays separate from the model
The assistant does not answer from the language model's general knowledge. The architecture chains customer → web chat → commercial rules and instructions → life insurance knowledge base → language model → quoting engine → result or handover.
That separation is the technical decision that matters most over time. It lets the commercial rules change without touching the model, and the model change without rewriting the business knowledge — which, in a regulated sector, is the difference between an update and a project.
Human contact does not disappear, it starts further along
When a salesperson is needed, the customer has already cleared the main questions, explained what they are looking for, supplied the initial details, seen a first quote and shown concrete intent. The sales conversation no longer starts from nothing.
The callback button is still there for anyone who would rather speak to a person from the start. What changed was not removing it — it was that it stopped being the only way forward.
What this case shows about applying AI to a process that already exists
The aim was not to add AI to a website because the technology is fashionable. It was to name one concrete point of friction — the customer wants to move now and the process asks them to wait — and change it.
It is the same reasoning we apply in other sectors: in the event configurator for a hotel group, an email enquiry gave way to a journey that returns a proposal. In every case, what gets shorter is the distance between the intent of the person searching and the first useful answer.
Frequently asked questions
Can an AI assistant produce a life insurance quote?
It can run the conversation that gathers what the quote needs — age, sum assured, term, whether it is tied to a mortgage — and hand those to the quoting engine the insurer already uses. The calculation stays with the insurer's engine; the assistant handles the gathering and the presentation of the result.
How is this different from a chatbot?
A chatbot answers a question and stops. This assistant takes several steps before returning a result: it understands, asks for what is missing, turns the answers into structured data, triggers the quoting process and returns or routes the outcome. It takes part in the sales process rather than commenting on it.
Does the customer stop speaking to a person?
No. Human contact still happens, and it happens at a more useful point: by the time it does, the customer has cleared the first questions, explained what they are looking for, supplied the initial details and seen a first quote. The sales conversation no longer starts from nothing.
Why does out-of-hours availability matter here?
Because insurance research often happens in the evening, at weekends or on public holidays — precisely when no sales team is available. Making someone wait until the next working day introduces friction at the moment of highest intent.
How does a conversation become usable data?
Each answer is converted into structured fields — age, sum assured, term, type of need, mortgage, contact details. To the customer it is a conversation; to the internal systems it is an organised record ready to feed the next step.
Does the assistant answer from the model's general knowledge?
No. It uses a knowledge base and commercial rules defined by the insurer, kept separate from the language model. That separation lets the rules change without touching the model, and the model change without rewriting the business knowledge.
Does the callback button stop making sense?
No. It still serves people who would rather speak to someone from the start. It simply stops being the only way forward — which was the problem, not the button itself.
Who did this work
BigLearn is a Portuguese artificial intelligence consultancy, founded in 2017 and based in Lisbon. We work with organisations that already have working processes and want to know where AI improves them — insurance included.
This case came out of our AI consulting for companies and web, SEO and digital transformation work. Every project starts with a proof of concept with a written success criterion agreed before we begin — and if it does not pass, we say so.
We work in hospitality and tourism, local government and public administration, insurance, healthcare, manufacturing and professional services. The other case studies are published under the same rule: client anonymised, verifiable figures, and the nature of the document stated up front.