Ask "am I covered in a case like this?" in the customer's own words, and get the policy section the answer rests on

Dozens of pages of policy per product, and a product map held in someone's head

At its core, life insurance consultation is about "the right answer, fast, with its evidence." The customer describes their situation and asks how the product works in that situation. The agent has to judge which condition in the policy that question falls under, and answer accordingly. The trouble is that the material this judgment leans on is scattered like this.

• Policy documents, one per product. Each runs to dozens of pages, and the wording and conditions differ subtly from product to product
• A varied product line. Which conditions attach to which product is written down separately in each document
• The customer information database (DB). This is where personalized recommendation is meant to plug in
• Questions that arrive mid-consultation. The customer describes their situation and asks how the product works in that situation
• The veteran agent's product map, held in their head. It's the instinct for hearing a customer's age, situation, and needs and narrowing the candidates — but it only travels with the person, it doesn't stay with the organization

So the burden on the floor comes down to four things. The lookup burden of having to dig through the policy yourself. The difficulty of pairing an answer with its evidence — having to point to "which policy, which part" alongside the answer. The limits of personalization, where choosing and comparing the right product for a customer's situation depends on experience. And the variance in consistency, where the depth and accuracy of the answer change depending on who fields the very same question.

Shuttling between memory and PDF

The way it's been done is simple but grueling. When a question comes in, the agent searches their memory, and if they aren't sure, they open the policy document. They search by keyword and pick out, from among similar clauses, the one line that exactly fits the customer's situation. All of this happens mid-call, with the customer waiting.

Handing over the evidence is yet another effort. To point precisely and say "this is in Article so-and-so of the policy," they have to find the answer and then re-verify its location. In a busy consultation, citing the evidence is easily skipped — and then both customer and agent move on carrying a small unease of "is this really right?"

Product recommendation and comparison were almost the territory of veterans. An experienced agent hears the customer's age, situation, and needs, then narrows the candidates from the product map in their head. But that instinct doesn't stay with the organization, and the depth of the answer splits from person to person.

Drop the policies in, turn them into a knowledge graph

The setup isn't something an agent repeats every time — it's a one-off job of putting the per-product policy documents into the workspace. The original policy files go in untouched.

Ask it in the customer's own words

"If I sign up now, am I covered in a case like this?"
→ Finds the relevant policy first and answers based on its content.
   The policy section the answer rests on comes attached, and a "Where this came from in your notes" card is drawn.
   Click a node on the card and that policy item opens in the graph view.

"Shortlist the products that fit this customer's situation"
→ Links to the customer information DB and narrows down product candidates that fit the situation.
   The final choice remains the agent's. It only shortlists the candidates that become a starting point.

"Lay out the differences in coverage and conditions among these candidates"
→ Organizes the differences in coverage and conditions among the recommended candidates so they can be weighed at a glance.

(When the customer shows interest)
→ Walks them through the enrollment process and, at the moment a human touch is needed, connects them to a sales representative.

The first question works because the policy documents were put in ahead of time. When a question comes in, the relevant policy is searched first, and the answer is based on its content. The aim is to let the agent reach a well-grounded answer without combing through the entire policy.

The point of the second and third questions is that they don't replace human judgment. The agent takes on only the repetitive, accuracy-critical work — policy lookup, organizing evidence, narrowing candidates. And at the moment the customer draws close to a real decision, the point where sales truly shines, it passes the baton to a person. The view is that what you choose not to automate matters as much as what you choose to automate.

While the agent digs through the policy and rounds up the evidence, the human focuses on the conversation only a human can have.

What changes

What we hold in hand right now isn't a proven result, but a direction worth validating. The next step is to put this flow onto a real consultation floor and watch whether the expectations turn into data.