Insurance

On Blits.ai this is an AI agent on the voice channel (real time streaming speech recognition and synthesis over telephony, with DTMF input) and on web chat and WhatsApp, with the insurer's own app connected through the API channel. A flow holds the intake schema per claim type, using ask question, slot filling and receive attachment blocks for the facts, photos and documents. Custom functions call the policy system to match the policy and the claims system to open the claim and return the claim number.

Guardrails block statements about cover, PII masking with configurable patterns keeps identifiers such as policy numbers, contact and payment details out of model prompts and logs, and human handover escalates to a claims handler through automatic rules, with live takeover also on voice calls. Test suites run sets of multi turn intake conversations for every claim type on each change, monitors run scheduled checks against the agent, and analytics show interactions and satisfaction per bot. The platform is model agnostic. Run Blits.ai in our cloud in the EU, Qatar, Singapore, the UAE, the US or Australia, in your private cloud, or fully on premises to meet data residency requirements.

On Blits.ai this is an agentic workflow triggered through the API when a claim is opened. An AI agent reads the claim and its documents through the knowledge base document ingestion (PDF, images and Outlook email files), writes a summary with structured output, and calls custom functions that fetch cover and limits from the policy system, pull fraud and severity scores, and write the routing decision back to the claims system.

Human in the loop approval holds any payment above the configured threshold for a handler, and the tool execution policy limits which systems the agent may change. Every run has a full audit trail, test suites grade summaries and routing against labeled historical claims on each change, and monitors run scheduled checks against the agent and alert by email or webhook when an expectation fails. The platform is model agnostic. Run Blits.ai in our cloud in the EU, Qatar, Singapore, the UAE, the US or Australia, in your private cloud, or fully on premises to meet data residency requirements.

Blits.ai is not a fraud scoring engine; insurers use their own models or a specialized vendor and connect them through custom functions. Blits.ai adds the investigation workflow around the score: an agentic workflow that, when an alert fires, gathers the claim file, documents and earlier claims through custom functions and SQL knowledge bases, and has an AI agent write a structured alert summary with the reasons and open questions for the investigator.

Human in the loop approval steps put holds, referrals and letters to the customer in front of a person before they happen, the audit trail records every step of each run, and a tool execution policy limits which functions the agent may call. Test suites check summaries against known cases. Run Blits.ai in our cloud in the EU, Qatar, Singapore, the UAE, the US or Australia, in your private cloud, or fully on premises where claims data must stay local.

On Blits.ai this is an AI agent with a knowledge base of policy wordings and endorsements, searched with hybrid retrieval, plus custom functions that call the policy administration system's REST APIs to fetch the customer's own schedule and wording version and to make mid term changes. Regulated journeys, such as changing payment details, run as a flow with an authentication step, and payment links handle any premium due.

The same agent serves web chat, WhatsApp and voice, and the insurer's own mobile app through the REST or WebSocket API channel, with streaming speech on the phone and multi language support. Guardrails check inputs and outputs, PII masking and card number tokenization happen at the gateway, and human handover passes the summary to the contact center, including live takeover. Test suites replay real coverage questions on every wording release, analytics show interactions, satisfaction and top intents, and conversation logs show each handover in full.

On Blits.ai this runs as an agentic workflow, started through the REST API from the broker portal or mailbox integration, or from a flow on the inbound email channel with a trigger workflow block. The platform's document ingestion reads PDF, DOCX, XLSX and Outlook .msg files, and an agent with structured output extracts the fields into the insurer's schema. Custom functions call the clearance service, the appetite rules and approved data providers over REST, and a SQL knowledge base gives the agent read access to account and loss history.

Declines and referrals go through human in the loop approval above a configurable threshold, so an underwriter confirms before anything reaches the broker. PII masking at the gateway, with custom masking patterns per bot, masks identifiers such as email addresses, phone numbers and account or policy numbers before they reach a model. Every run has a full audit trail and trace, and test suites replay a labeled set of past submissions before each change goes live to track field accuracy. The platform is model agnostic, so extraction and ranking can use different models. Run Blits.ai in our cloud in the EU, Qatar, Singapore, the UAE, the US or Australia, in your private cloud, or fully on premises.

On Blits.ai the sales process is a flow for the regulated steps (demands and needs, disclosures, confirmation) with an AI agent block for the open conversation, and a knowledge base of product documents for coverage questions. Custom functions call the rating engine and policy system over REST, so the price always comes from the insurer's own rules, and payment links through Stripe, Mollie or Adyen take payment with card data tokenized at the payment provider.

The same sales process runs on web chat, WhatsApp and voice, and inside the insurer's own app through the API channel, with rich cards for options and product recommendations and multi language support. Output guardrails with admin authored policies catch price or cover statements the agent should not make, human handover routes advice requests to licensed staff, and test suites replay misselling scenarios on every change. Analytics and flow statistics show how each step performs, and human in the loop approval can be required for agent actions above a set threshold.