This guide provides example prompts for each business area. Connect your AI agent/copilot/assistant and select your BriteCore role before using them. See [Getting started](./getting-started.md) if the connector is not connected yet.
Replace values in angle brackets with identifiers or names from your tenant. The connector uses your BriteCore permissions, so two users may receive different results for the same prompt. Product setup also varies by tenant.
For any change, state exactly what you want and ask the assistant to show the planned action before it proceeds. The examples below do not change saved data.
Policies and coverage
Example 1: review product setup by effective date
> Show the five most recent configured effective dates. For the most recent date that contains product setup, list the lines of business and enabled states. Keep the IDs in the response. Do not change anything.
The connector uses `list_effective_dates`, `get_all_lines`, and `get_all_states`. Asking for the most recent date with setup matters because a future effective date may exist before anyone adds products to it.
Example 2: find policies for a contact
> Look up contact `<contact-id>`. If the contact exists, list up to five associated policies and summarize their status and effective dates. If the contact does not exist, stop. Do not guess another ID.
The connector checks the contact with `retrieve_contact_info` before it calls `get_policies`. This avoids confusing an invalid contact with a valid contact that has no policies.
Claims
Example 1: review the peril catalog
> What claim perils are available in this tenant? Group them by the kind of loss and call out any entries that have no name. Read only.
This calls `get_all_perils`. The answer reflects the tenant's current claim configuration, not a generic insurance list.
Example 2: prepare a claim review
> Resolve claim number `<claim-number>`. If it exists, summarize the claim and list its exposures with their current accounting position. If it does not exist, stop. Do not create a loss or change reserves.
The connector resolves the number with `resolve_claim_number`, then uses `get_claim_snapshot`, `list_exposures`, and `get_exposure_accounting_data`. The prompt keeps reserve changes out of the
review.
Underwriting rules
Example 1: summarize a product's rules
> For policy type `<policy-type-id>`, list the underwriting rules and summarize them by rule type and enabled status. Include the rule IDs. Read only.
This uses `list_rules`. A rule ID lets you request the full rule with `get_rule` in a follow-up without asking the assistant to infer which rule you meant.
Example 2: validate a proposed rule without saving it
> For policy type `<policy-type-id>`, inspect the live rule configuration and
> comparable rules. Draft and validate a `<referral-or-decline>` rule for `<business-condition>`, but do not create or update it. If a field, trigger value, or effect ID cannot be derived from the live configuration, stop and tell me exactly what is missing.
The connector reads `get_rules_configuration` and may inspect existing rules before it calls `validate_rule`. Validation does not save a rule. Creating or updating one is a separate request that should use the validated rule details and require confirmation.
Reporting and analytics
Example 1: find an existing report
> List the SQL Editor report categories and the first five reports I can access. Include whether each report is a draft and its output format. Do not run a report.
This uses `list_report_categories` and `list_reports`. It reads the report catalog without starting a report job.
Example 2: choose a documented view
> Find the reporting views relevant to `<business-topic>`, then show the documentation for the single best view at `<claim-or-policy>` grain. Explain which related views I would need for financial detail. Do not execute a query.
The connector uses `list_views` and `get_view_documentation`. Review the view grain and columns before asking it to build or validate SQL.
Documents and forms
Example 1: audit forms on a product
> List the forms attached to policy type `<policy-type-id>` and its items. Group them by form code and identify duplicate file names. Read only.
This calls `get_policy_type_forms`. An empty result can be valid, so the assistant should distinguish an empty product from a truncated response.
Example 2: get a temporary form download link
> For policy type `<policy-type-id>`, find the form named `<form-name>` and generate a download URL. Report when the URL expires, but do not download the file. If the form is not attached, stop and do not guess a file ID.
The connector first calls `get_policy_type_forms`, then passes the returned file ID to `get_form_download_url`. The URL is temporary and should not be copied into long-lived notes or tickets.
Quoting
Example 1: find products open to quoting
> List the products available for a new quote. Include the line of business, state, product type, and policy type ID. Do not create a quote.
This calls `get_available_policy_types`. The tool returns the tenant's products that are open to quote creation. It does not accept an effective-date or state filter, so check the returned fields instead of assuming the list was filtered.
Example 2: prepare a stateless premium preview
> Prepare a stateless premium preview for policy type `<policy-type-id>` on effective date `<effective-date-id>`. Inspect its quoting questions and resolve the coverage names I provide. List every applicant, risk, coverage, and underwriting answer still needed. Show the planned request and ask me to confirm before calling `create_and_rate_quote`. Never change `stateless` to `false`.
The connector can use `get_quoting_questions` and `resolve_coverage_names` to build the request from live product setup. A confirmed call with`stateless: true` calculates a preview without saving a quote. Omitted rating inputs may fall back to configured defaults, so review the planned request before execution.
If the AI assistant needs more information
Supply the missing tenant identifier or business decision, then continue in the same conversation. Do not ask the assistant to invent IDs, coverage values, rule effects, or customer data.
If a tool returns `FORBIDDEN`, your selected BriteCore role lacks the required permission. See [Authentication](./authentication.md) before reconnecting with a different authorized role. For tool details and write warnings, see [Capabilities](./capabilities.md).
