How a risk management consulting firm cut complex document review from 2,700 hours to 80 with generative AI

Summary

A North American consulting firm specializing in outsourced risk management, claims recovery, benefits advisement, and actuarial services handles large volumes of complex, client-facing documents. Those documents are dense with legal and financial terminology, unstructured data, and inconsistent formats, which made manual review slow and error-prone.

The firm worked with Marlabs on a generative AI (GenAI) proof of concept (PoC) to test whether AI-powered querying and analysis could carry the load. Using Microsoft Copilot and ChatGPT-4o, the firm reduced the time needed to process complex legal and financial files from 2,700 hours of manual effort to 80 hours. The PoC is projected to save $175K–$225K a year, generated an 80 percent return on investment (ROI) within a few weeks, and gave the firm a validated path to scale its services.

Challenge

The firm, which has more than 100 employees, relies on its analysts to extract critical details from high volumes of legal and financial documents. Every file arrived in a different format, and much of the content was unstructured text written in specialized terminology.

Manual review was time-consuming, inconsistent from one reviewer to the next, and unable to scale as document volume and complexity grew. Analysts spent a large share of their time on repetitive extraction work instead of the advisory work clients hire the firm to do.

The firm needed to know whether AI could process and extract value from these documents reliably. Its objective was to evaluate AI's feasibility and build a strategy for accelerating, standardizing, and improving document analysis across diverse document types and use cases.

Solution

Marlabs designed a four-phase GenAI PoC that used Microsoft Copilot and ChatGPT-4o to query large volumes of the firm's client documents. The PoC tested how AI could improve speed, accuracy, and scalability in information extraction while keeping results compliant and relevant to the firm's domain.

AI Discovery and Readiness

The team began by assessing the firm's existing document processes and data structures to determine where AI would fit. This work evaluated the full document landscape and identified the bottlenecks that slowed review the most. Marlabs and the firm then aligned on the success criteria the PoC would be measured against so that the results would answer a clear business question.

Model Selection and Testing

With success criteria in place, the team tested multiple GenAI tools on their ability to handle complex, domain-specific content. Microsoft Copilot was integrated with the firm's environment, and ChatGPT-4o was put through performance trials against the same document sets. The team benchmarked the tools side by side to identify which performed best on the firm's legal and financial language.

Query Design and Interaction Simulation

Next, the team built and tested natural language queries that simulated real-world document analysis. Prompt engineering shaped each query around the questions analysts actually ask of client files. The team refined queries and responses through repeated rounds and simulated the firm's core use cases end to end.

Performance Evaluation and Insights

In the final phase, the team measured the accuracy, speed, and usefulness of AI-assisted responses. AI outputs were compared against manual review results to confirm accuracy, and time savings were analyzed across the document set. The team also evaluated how effectively the AI captured the insights analysts need, giving the firm evidence to support a long-term decision.

Results

The PoC demonstrated that GenAI can automate and strengthen the firm's ability to extract meaningful insights from complex client documents. Processing time for complex legal and financial files fell from 2,700 hours of manual effort to 80 hours, a change projected to save $175K–$225K annually. The PoC generated an 80 percent ROI within a few weeks.

Accuracy improved alongside speed. The firm increased precision in extracting critical details from unstructured text and reached 100 percent process accuracy in the PoC. Analysts spent less time on repetitive tasks and more time on strategic work, and client-facing teams received more timely and relevant insights. With high-volume document intake now handled consistently, the firm is equipped to scale its services and make faster, better-informed decisions for clients.

Impact

The PoC gave the firm measurable evidence that GenAI can carry its document review workload:

• Review time for complex legal and financial files fell from 2,700 hours to 80 hours.

• Projected annual savings reached $175K–$225K.

• The PoC generated an 80 percent ROI within a few weeks.

• Extraction reached 100 percent process accuracy on unstructured text.

• Analysts shifted time from repetitive extraction to strategic client work.

• The firm validated GenAI tools for long-term operational use.