Policy administration for the agentic era.

Viola is an AI-native, API-first Policy Administration System, or PAS. RegenAI InsurTech builds it. A team of specialised AI agents runs the whole policy lifecycle: underwriting, policy administration, claims, payments, and distribution. Humans stay in the loop on the decisions that carry risk. A PAS is the system of record for an insurance book: it issues quotes, binds policies, stores them, handles endorsements and renewals, and holds the current state of every in-force contract. Viola runs that as seven systems on one engine, covering 30 named capabilities: Multi-Agent Intelligence, the Rating Engine, Document Intelligence, Payments and Billing, Real-Time Operations, Viola Concierge, and Security and Governance. Personal auto is live today and commercial lines are next. Products, rates, and rules are configured rather than coded, so changing the operation does not mean changing the software. API-first matters for the same reason from the outside: every capability is exposed through clean APIs and real-time channels, so systems you keep can read and write the same record rather than syncing against it. That is what makes a partial deployment possible instead of a full replacement.

MGAs, program administrators, carriers, and digital insurers. The common thread is wanting to automate the policy lifecycle without waiting on legacy vendors. MGAs use it to scale into new programs and states. Carriers use it to modernise policy administration without rip-and-replace risk. Program administrators run high-volume auto and personal-lines books on one source of truth. What those three have in common is ambition constrained by tooling. An MGA wants a new program live this quarter, not next year. A carrier wants to retire a legacy core without betting the book on a migration. A program administrator wants one record every desk reads from. Viola is designed for all three, and the same platform serves them because products, rates, and rules are configuration rather than a bespoke build.

AI-native. The core was built AI-first from the first line of code. Multi-agent intelligence orchestrates the whole policy lifecycle. It is not AI bolted onto a legacy core. The agents handle data ingestion, rating, document analysis, and claims decisions in parallel. That was the founding bet: generative and agentic AI changed what was buildable, and the obvious move was to wrap AI around an existing core. RegenAI InsurTech built the core itself AI-first instead. The practical difference is where intelligence sits. In an AI-wrapped system the assistant sits beside the workflow and hands work back to a person. In Viola the agents are the workflow. Human-in-the-loop gates sit on the decisions that carry risk, so nothing consequential happens without a person’s approval.

Under one week. A Georgia expansion takes five working days. Rate tables are imported on day 1. State factors and rules are configured on day 2. Actuarial review happens on day 3. The rate version is approved and scheduled on day 4. The first Georgia policy binds on day 5. Rating stays versioned, effective-dated, and approval-gated, so every change carries an audit trail. What makes that possible is treating a new state as configuration rather than an IT project. Base rates, factors, and coverage rules are set per state inside the platform, and actuaries import rate tables in bulk from the spreadsheets they already use. The platform enforces compliance guardrails throughout, so speed does not come at the cost of the approval trail.

Four stages, running continuously. Capture takes in and validates data from applications, emails, police reports, rate tables, and third-party feeds. Orchestrate runs underwriting, policy administration, and claims end to end, with no swivel-chair handoffs. Resolve decides pricing, routing, reserves, and settlement in real time, with people approving where judgment matters. Evolve feeds every transaction back into the platform. Across the book that covers three stages of the contract itself. Underwrite takes a submission from data capture to a bindable, accurately priced quote in minutes. Service handles endorsements, renewals, cancellations, documents, billing, and payments in one place, with every change auditable end to end. Settle reads incoming claims the moment they arrive, extracting documents, estimating severity, and routing to the right adjuster automatically.

A chat assistant built into every portal. Customers can quote, buy, and service a policy entirely in conversation. Admins ask about the book in plain English and get real numbers back: policies bound, open claims, active premium. It is role-aware, account-scoped, and asks for confirmation before it acts. The distinction from a support chatbot matters. A support chatbot answers questions about the product. The Concierge does the work: a policyholder moves from quote to bound policy without leaving the conversation, and an admin gets real figures rather than a link to a dashboard. Being account-scoped means it can only see and act on the records the person asking is entitled to, and the confirmation step means it never commits a change on an inferred intent.

The work stops queuing behind people. Specialised agents run in parallel: one reads the documents, one assesses the risk, one moves the workflow, one watches outcomes. A submission is captured, rated, and priced while it would otherwise be sitting in an inbox. Quotes, claims, assignments, and approvals stream to every screen live, so no one waits for a nightly batch or a refreshed queue. Assignments, approvals, and status changes raise instant notifications, and operational dashboards show live performance rather than yesterday’s. The agents also retrieve against your own policy and claims knowledge by meaning rather than exact keyword, so a question finds the right prior file even when the wording differs. Every transaction feeds back in, so retrieval keeps improving as the book grows.

No. It removes the work around their work. Automation handles intake, extraction, routing, and the routine cases that need no judgment. Underwriters and adjusters get the files that actually need a person, with the facts already assembled. Human-in-the-loop gates sit on the decisions that carry risk, so pricing approvals, reserves, and settlements stay with people. On a claim that means OCR reads the police report, AI extracts parties and damages, triage scores severity, coverage match and completeness on arrival, and rules auto-assign the file with a reserve recommendation attached. Duplicate detection, phase tracking, and acknowledgements run unattended. The adjuster still decides the reserve and the settlement, but arrives at the decision with the file already built rather than spending the first hours assembling it.

One platform holds the whole lifecycle, so there is nothing to re-key into. Document AI ingests applications, police reports, estimates, photos, and emails, then extracts the facts and validates them against the file. Underwriting, policy administration, claims, and billing read the same record. Work moves between desks without a swivel-chair handoff or a re-entry step. The document side covers PDF, DOCX, images, spreadsheets, and email, and the platform generates policy documents, acknowledgements, and filings from the same record rather than from a separate document system. Billing is inside the bind and renewal flows rather than beside them, so a payment does not need reconciling back to the policy. The architecture is API-first, so systems you keep integrate against the same record too. The re-keying problem is worth naming precisely, because it is rarely a typing problem. Every re-entry is a chance for two systems to disagree, and reconciling that disagreement later costs more than the original entry did. Holding one record removes the disagreement rather than detecting it.