Can the agent actually act in Guidewire / Duck Creek, or only read?
Both. Native read/write into Guidewire ClaimCenter, PolicyCenter and BillingCenter; Duck Creek Claims / Policy / Billing; Majesco; Sapiens. The agent operates within scoped tools — it can open a claim, seed a reserve against your reserving-philosophy table, post a payment, write a coverage decision the adjuster has authorized, and update the file disposition. It cannot, for example, override a coverage decision or bind beyond filed authority. Scope changes flow through your existing IT and your claims-ops change-management process.
How does the agent stay clean of the Unfair Claims Settlement Practices Act?
By construction. The agent will not misrepresent policy terms, will not use coercive language on quantum, will not threaten denial, and will not interpret an exclusion. Any quasi-judicial conversation — coverage disputes, denial language, settlement quantum — routes to a licensed adjuster. Your UCSPA-aligned scripts are the boundary, signed off by your General Counsel before any state goes live, and we walk your market-conduct readiness team through the call sample on request.
What happens to a call recording under GLBA, state insurance data-security laws, and DPDP?
Recordings and transcripts live under your KMS, in your residency, with PII redacted before any model call. Retention is configured per the record-management policy your General Counsel has approved, plus the state-by-state and India-specific (DPDP Act 2023) requirements. Right-to-access and right-to-erasure are honored through the same workflow your CRM and policy-admin platform already supports. We do not stand up a parallel data store.
How are reserves seeded? My chief actuary will need to defend the number.
Reserves are seeded against your reserving-philosophy table — loss type × coverage × jurisdiction × prior-claim signal — not by the model. The agent records the seed source, the seed value, and the rule applied. Your chief actuary can defend the seed against the same documentation in a market-conduct exam, in a Schedule P review, or with the reinsurer. We do not let the model invent a number.
What about fraud? Are you doing detection or just intake?
Repliant computes a composite background signal from ISO ClaimSearch, LexisNexis, velocity-of-claim, geo-consistency, voice-stress and your own SIU rules — and surfaces it to your SIU through the same case-management surface they already use. The agent does not accuse the policyholder, does not deny, and does not imply suspicion on the call. We are not your fraud detection vendor; we are the layer that captures cleanly and flags honestly.
TCPA — can it call my renewal book without putting me on the front page of a class action?
Yes. Outbound is restricted to numbers with a documented consent record tied to your in-force or pre-bind book; quiet hours, STOP / HELP keywords, frequency limits, Reg DNC, and state-by-state restrictions are enforced at the channel layer. The agent does not initiate to a number without consent on file, and the consent record is exportable for plaintiff-defense. Renewals on producer-of-record accounts honor commission attribution end-to-end.
Can my independent producers see what the agent did on their book?
Yes — the producer's existing agent portal shows the same authoritative claim, renewal and service status as the policyholder's WhatsApp and the adjuster's ClaimCenter view. We do not stand up a parallel producer experience. Commission attribution is honored; no policy is silently rewritten under a different code. Independent producers are typically the first stakeholder we walk through during UAT, because they are the loudest if anything is off.
What does the audit trail look like? My market-conduct readiness team will want a sample.
Every session emits a signed, immutable record: channel, identification path, tools called with arguments, policies evaluated, the prompt-pay timestamp, the reserve seed source, the model trace, and the final disposition (with the ClaimCenter / Duck Creek write confirmation). Records are exportable to your existing GRC (ServiceNow IRM, MetricStream, RSA Archer) on whatever cadence your team configures. We have walked carriers' market-conduct readiness teams through full call samples inside fifteen minutes.