How CPIA replaced multi-day manual reporting cycles with a single-chat-turn AI policy platform
1 turn
Report generation for any carrier, any filter
Same-day
New carrier onboarding via pluggable architecture
1-click
Bound-vs-cash reconciliation, anomalies pre-flagged
~0 errors
Shipped downstream, caught before files leave
Industry
Specialty Insurance / Policy Lifecycle & Carrier Reporting
Organization
U.S. specialty-insurance services firm managing thousands of in-force policies, hundreds of new certificates monthly, and carrier reporting cycles for lender-driven property programs
Challenge
Brittle premium logic, hand-coded carrier exports, downstream error discovery, fragile ePay reconciliation, and multi-day manual reporting cycles per carrier
Solution
Rebuilt policy-lifecycle platform plus four Claude-powered agents that propose and pause for human review before acting
The Challenge
CPIA's platform had grown organically over years, and five compounding problems held the business back. Premium logic was spread across inconsistent code paths. Every carrier required a separately hand-coded data export, meaning a new carrier was weeks of developer time. Errors were not discovered until files reached the carrier. ePay failures could silently hold payments for days. And each reporting cycle consumed multiple person-days of filter, export, reformat, and re-save, multiplied across every carrier on the program.
Premium calculation logic distributed and brittle
Edge cases (multi-location quotes, mid-term endorsements, state-specific tax math) were handled inconsistently between code paths, creating silent pricing errors.
Carrier data-file exports hand-coded per carrier
Every new carrier or schema change required weeks of developer work. Business growth was bottlenecked behind engineering capacity.
Errors discovered downstream, at the carrier
Refund mismatches, missing required fields, and unreconciled bound premiums surfaced only after files shipped, sometimes triggering carrier escalations and correction cycles.
ePay reconciliation was fragile
Silent failures left payments in a holding state for days while staff worked from inbox notifications with no structured resolution workflow or audit trail.
Reporting consumed multiple person-days per month, per carrier
Each cycle: log in, filter, export CSV, open Excel, rename columns, compute metrics, reformat, re-save, then repeat for every carrier. An expensive, error-prone monthly ritual.
The Solution
Attri delivered a rebuilt policy-lifecycle platform and a Claude-powered agentic layer on top. Four specialized agents read, reason, and propose, then pause for a human reviewer before anything is sent or modified. The platform unifies premium logic, makes carrier onboarding a configuration exercise, and gives the operations team a conversational interface to their own data.
Report Query Agent
Translates natural-language requests ("Arch Specialty for July 2026, Policy Fees = $50") into structured filter parameters, runs existing report services, and returns the result inline with a download link. PII is tokenized before any data reaches the model.
Report Transformation Agent
Accepts column-level customization in plain English, generates a preview of the modified dataset, and exports to CSV / Excel / PDF, with explicit warnings when the reviewer is about to modify a carrier-facing file that must not be sent in altered form.
Payment Pattern Agent
Runs analytical queries across transaction-level data ("Which policies had first-installment cover premium but skipped taxes?"), returning structured results with a complete audit trail per session.
Anomaly Detection Agent
Proactively scans certificates before files ship: flags refunds exceeding paid premium, missing required fields, and cancellation factors set without a cancelled date. Issues are caught and resolved before they reach the carrier.
Additional Features
Django REST + React / TypeScript
Modern web application with PostgreSQL persistence and a canonical premium engine handling multi-state tax math and per-location parallelism: one code path, one truth.
Pluggable Report Architecture
New carriers onboard same-day via configurable report schemas. No code changes are required for new carrier formats, column variations, or schema updates.
Automated Daily ePay Pipeline
A Dramatiq + RabbitMQ pipeline runs the ePay reconciliation waterfall each night with a complete audit trail. Silent failures are structurally impossible: every step has a status.
LLM Provider Abstraction
The underlying model is a configuration choice, not a code dependency. Providers can be swapped without touching business logic, keeping the platform future-proof.
Report Query Agent
You
Arch Specialty for July 2026, Policy Fees = $50
Report Query Agent
Filtered the July 2026 Arch Specialty book to certificates where Policy Fees = $50, ran the existing report service, and prepared the export. Bound premium reconciles to cash received and no anomalies were flagged. The result is ready inline with a download link. PII was tokenized before processing.
Confidence: High
Ask a follow-up question...
Works with your existing tools
The Results
| Metric | Before | Now |
|---|---|---|
| Report-generation effort | Multiple person-days per cycle, per carrier | Single chat turn with a previewed download |
| New-carrier onboarding | Weeks of hand-coded schema work per carrier | Same-day via the pluggable report architecture |
| Bound-vs-cash reconciliation | Manual end-of-month spreadsheet exercise | One-click report with anomalies pre-flagged |
| ePay reconciliation | Silent failures, payments held for days | Automated daily waterfall with full audit trail |
| Error prevention | Caught downstream at the carrier after files shipped | Caught pre-export, anomaly agent flags before files leave |
| Auditability | Spreadsheet trails, manual logs, no session history | Full chat + tool-call + export log per session |
| Operating model | Admins doing repetitive data entry under monthly deadline | Human-in-the-loop approval over agent proposals |
Errors caught before they ship, no more carrier-side corrections or escalations.
Growth no longer gated by engineering, new carriers are configuration, not code.
The operations team queries its own data conversationally, no more monthly Excel rituals.
Every action logged, a full session audit trail for carrier-level accountability.
Human approval on every agent proposal, nothing is sent without sign-off.
Why It Worked
Errors caught before they ship
The anomaly agent proactively scans and flags issues before export, no more discovering problems after the carrier has received the file.
Human always in the loop
Every agent reads, reasons, and proposes a course of action. A reviewer decides. No file is exported, modified, or sent without explicit human approval.
One canonical premium engine
All premium logic (multi-state taxes, per-location parallelism, mid-term endorsements) consolidated into a single engine. One code path, consistent results.
Growth no longer gated by engineering
The pluggable report architecture means new carriers and schema changes are configuration exercises, not development sprints.
Full audit trail, every session
Every chat turn, tool call, and export is logged. Any action can be reviewed, replayed, or explained at any time, essential for carrier-level accountability.
LLM-agnostic by design
The underlying model is a config entry, not a code dependency. Providers can be switched without touching business logic, keeping the platform current as models improve.
Security & Compliance
Ready to Rebuild Your Policy Workflows?
See how Attri can rebuild your policy and carrier-reporting workflows with governed, auditable AI agents and a human-in-the-loop approval chain.
Ready to automate
Get in touch
See how Attri can automate your document workflows in minutes.