Reducing Back-Office Processing Costs with Agentic AI

How a Leading U.S. Life Insurer Modernized a 26-Year-Old Document Workflow

Executive Summary

A leading U.S. life insurance carrier ran its entire back-office document operation on a 26-year-old platform built on end-of-life technology, processing roughly 1,500 documents per day through fully manual indexing and routing. Andersen Consulting's AI & Advanced Analytics practice designed and scoped an agentic AI system built on AWS, using Amazon Bedrock with Claude models alongside Textract and Step Functions to automate ingestion, classification, and routing end to end. The scoped solution targets a phased replacement that removes approximately 75 person-hours of daily manual indexing while establishing a governed, secure, and observable agentic pipeline. During the scoping phase, the analysis showed a 4.4x to 5.9x annual return on infrastructure cost from indexing labor displacement alone, demonstrating the immediate financial case for Andersen Consulting's applied agentic AI approach and creating a scalable foundation for full back-office transformation.

Introduction: Context and Challenge

Across the Industry

Life insurers carry large blocks of in-force policies on aging administration platforms that are expensive to run and difficult to modernize. Rising security exposure from unsupported systems, growing correspondence complexity, and institutional knowledge locked in the heads of long-tenured staff have made back-office modernization an urgent operational and regulatory priority.

The Client’s Challenge

The insurer operated a document workflow system managing the full back-office lifecycle for a policy block of roughly 180,000 policies, running on Windows 7 and Windows Server 2003 with no vendor security support. Every inbound document, whether fax, scanned mail, web upload, or internal print job, required a human operator to open it, identify the policy number, assign a business area and work type, validate against the mainframe, and route it to the correct queue. This manual indexing step alone consumed approximately 75 person-hours per day at a cost of $683,000 to $878,000 annually. A second operator then pulled each item, swivel-chaired to the policy administration system to execute the action, and generated correspondence through an embedded mail-merge module carrying 700 to 800 templates when only 60 to 70 were needed. The platform combined end-of-life operating systems, a legacy colo network, and business rules that existed only as institutional knowledge, putting continuity, compliance, and cost squarely at risk.

Solution: Agentic AI Built on AWS with Claude Models

Approach Overview

Andersen Consulting mobilized a cross-functional team that paired applied AI engineers with insurance operations specialists. The team focused on the real decisions operators made each day rather than on replicating the old system feature for feature. The result is an agentic AI solution that lets the business process documents with no human touch unless a document fails a confidence threshold or triggers a high-risk approval gate.

Data Integration and Foundation

The architecture uses Amazon S3 as a universal document landing zone and Amazon EventBridge to route processing events with content-based filtering and replay. This creates a single governed entry point for every channel that previously fed the old platform through separate paths, and it gives the agentic layer clean, event-driven access to every document. All infrastructure is defined in CloudFormation, so the environment is consistent, auditable, and reproducible rather than assembled by hand..

AI and Insight Generation

Amazon Textract extracts structured fields, and Amazon Bedrock with Claude models drives the agentic core that handles classification, NIGO detection, and processing orchestration. The agents route simple decisions to lighter Claude models and complex reasoning to more capable ones, cutting inference cost while preserving accuracy. Consider an annuity address change. The Claude-powered agents now read the document, identify the policy number, validate it, classify the work type, and route it in seconds, a task that previously took an operator roughly three minutes of manual indexing per item. Bedrock Guardrails constrain PII handling, and every agentic decision is traceable, keeping outputs compliant and explainable.

User Experience and Workflow Integration

Documents that clear confidence thresholds flow straight through the agentic pipeline without operator involvement. Staff attention shifts to genuine exceptions surfaced through a review interface, so the people who know the business spend their time where judgment matters. Adoption is straightforward because the agentic AI removes work rather than adding steps.

Rapid Implementation

Andersen recommended scoping the first phase around the two highest-volume ingestion channels and the top five work types, delivering complete end-to-end agentic automation for those combinations before expanding incrementally. This phased path was endorsed by the insurer's leadership during scoping. The approach reflects Andersen's execution discipline: prove value on the highest-volume work first, then scale the agentic system with confidence.

Results and Impact

The scoped solution replaces a costly manual operation with an agentic AI pipeline that pays for itself many times over.

Financial Return: At roughly $127,000 in estimated annual infrastructure cost against $683,000 to $878,000 in current manual indexing labor, the solution delivers a 4.4x to 5.9x return from indexing displacement alone.
Operational Efficiency: Automated ingestion and classification through Claude-powered agents remove approximately 75 person-hours of daily manual indexing, freeing roughly 125 users for higher-value work.
Speed to Decision: The agentic pipeline moves documents from arrival to routing in seconds rather than the three minutes per item required under the manual process.
Risk and Compliance Alignment: The move off Windows 7 and Server 2003 onto a secured AWS environment with encryption, VPC endpoints, and least-privilege access closes a significant security exposure.
Institutional Knowledge Capture: Encoding NIGO business rules into the agentic system converts fragile tribal knowledge into governed, maintainable logic.
Correspondence Rationalization: Consolidating 700 to 800 templates toward the 60 to 70 actually needed reduces maintenance burden and error risk.

These outcomes convert a legacy liability into a modern, measurable, and scalable operation.

A New Standard for Back-Office Operations

The engagement shows what disciplined human and agentic AI collaboration delivers when the technology serves the operation rather than the reverse. Andersen Consulting and the insurer built a foundation that elevates back-office processing from a manual cost center into a governed, intelligent capability powered by Claude models.

"This gives our teams the ability to focus on the exceptions that actually require judgment, while the routine work handles itself," said the insurer's Chief Operations Officer. "It has changed how we think about scale and risk across our entire policy block."

Looking Ahead

With Phase 1 scoped, the insurer plans to expand agentic automation across additional channels and work types, then extend into downstream processing automation as policy administration API access is confirmed. Andersen Consulting expects to pair the platform rollout with change management support to help operations teams transition into exception-focused roles. Each step moves the organization closer to a fully AI-enabled back-office operating model built on agentic AI.

Deliver Value to Your Organization with AI in Weeks, not Years

Andersen Consulting is a global consulting practice providing a comprehensive suite of services spanning corporate strategy, business, technology, AI transformation, and human capital solutions. Andersen Consulting integrates with the multidimensional service model of Andersen’s global platform, delivering world-class consulting, tax, legal, valuation, global mobility, and advisory expertise.