How Fractional AI Cut Costs by 84% for a PE-Backed E-Commerce Company

Overview

A PE-backed e-commerce company was previously working with a business process outsourcing (BPO) firm on an expensive document processing task. The e-commerce company partnered with Fractional AI to fully automate the task and drive significant recurring cost savings. In addition to cost reduction, the new genAI-powered system drove product improvements—operating significantly faster and more accurately than the BPO. 

Problem

The client’s workflow involved significant manual review of lists. Each list contained nuanced information—like item types, brands, quantities, and memos—that required structured mapping and validation. The process occurs annually at a large scale, incurring substantial BPO costs. 

Solution

Fractional AI designed and deployed an end-to-end automated system powered by large language models. The new process:

• Extracts key information from raw documents, including embedded notes
• Structures that data into standardized list formats
• Maps items with associated quantities, qualifiers, and annotations
• Evaluates items and lists to determine whether human quality assurance (QA) is required
• Automatically incorporates feedback from the QA process for self-learning

The system architecture was built for scale and reliability, with multiple levels of retries and timeouts to ensure continued operation even when services are unavailable. Delivered as fully contained infrastructure-as-code, the stack can be integrated easily and immediately.

Impact

The result was a faster and smarter system that dramatically reduced operational load:

• The new system is 84% cheaper to run with costs expected to continue declining
• List processing time dropped to 53 seconds (from 24+ hours)
• QA workload was immediately slashed, allowing internal QA time to be reallocated to higher-priority projects
• QA costs are projected to approach $0 as feedback is incorporated into the new system 
• The model outperformed legacy workflows in head-to-head evaluations, delivering more accurate results in a greater share of cases
• With the new system, the client now owns the intellectual property—rather than it being held offshore

Looking Ahead

With final handoff complete—including code, documentation, and walkthroughs—the client is set up to realize value for years to come. Additional automation projects are scoped and ready, positioning the client to capture the full promise of genAI.

‍