Is my bike covered if I leave it outside a shop?
Customers ask support whether they are covered before they do something, or after something happens. Is my phone covered abroad? Does my policy cover my business equipment at home? Am I covered if my lodger causes damage? The question arrives in the app chat.
The agent finds the customer in the policy system, checks which product and wording version they have, opens the wording PDF, searches for the right section, checks the schedule for limits and excesses, and looks for endorsements. Then they try to phrase an answer that is accurate without promising too much. Some agents escalate everything; others answer from memory of a similar case.
Why cover questions take so long
The answer depends on documents specific to that customer, and the agent has to assemble them every time.
- Wordings exist in several versions, and the customer's version depends on their policy dates.
- Limits and excesses are on the schedule, not the wording.
- Endorsements change the standard cover for some customers.
- Wording language is hard to translate into a plain answer.
- There is no guidance on when to escalate and when to answer.
The risk in quick answers and the cost of slow ones
Slow answers frustrate customers at a moment when they are deciding whether to trust you. Quick answers from memory risk saying someone is covered when they are not, which comes back at claim time as a complaint. Escalating everything overloads underwriters and claims handlers. New agents take a long time to handle cover questions confidently.
Whether something is covered is decided under the policy by your claims team. Support can explain what the wording says; they should not decide claims in advance.
A policy-aware assistant for agents
What we build helps agents find and quote the customer's own documents.
- A panel in your helpdesk identifies the customer and loads their policy schedule, wording version and endorsements from your policy system and document store.
- The agent's question, or the customer's message, is matched to the relevant sections of that customer's documents, using search and an AI model such as Anthropic Claude.
- The panel shows the relevant clauses, limits and excess, with links to the exact place in the document.
- A draft reply explains what the wording says in plain language and quotes it, using phrasing your team approves, such as explaining what the policy says rather than confirming a claim will be paid.
- Questions your rules mark as needing an expert, or where the match is weak, are sent to an underwriter or claims handler with the context attached.
- Agents edit and send. Nothing goes to the customer without an agent.
- Common questions and gaps are reported, which can inform clearer wording or help articles.
| Question type | Assistant does | Route |
|---|---|---|
| Where cover applies, such as abroad | Finds territorial limits clause and schedule | Agent answers with the quoted clause |
| Whether an item is included | Finds item definitions and limits | Agent answers, or escalate if unclear |
| Something already happened | Finds relevant section | Guide the customer to make a claim, no prediction |
| Unusual situation | Weak or no match | Underwriter or claims handler |
What changes for agents and customers
Agents answer from the customer's actual documents, quoted, rather than from memory. Answers are more consistent. Hard questions reach the right expert with context. Customers get a clear answer that points to their own policy, and when something has already happened, they are guided to claim rather than given a guess.
Is this your support team's experience?
- Agents open wording PDFs to answer cover questions.
- Answers differ between agents.
- Cover questions are often escalated to underwriters.
- Customers have been told they were covered when they were not.
- New agents avoid cover questions.