The situation
A nationwide titling and registration company processes the paperwork required to complete vehicle sales for dealers across all 50 states. The work is high-volume and detail-intensive, and it varies in ways that resist standardization: state-specific forms, dealer-specific conventions, transaction-specific compliance rules.
Growing the business meant processing more documents, which meant hiring more auditors with the experience to handle the variety of requirements. Those specialists were difficult to find, slow to ramp, and easy to lose. When they left, their institutional knowledge went with them. Dealers felt the strain of a longer turnaround as paperwork errors led to rejections, rejections meant resubmissions, and the gap between sale and final registration kept widening. The paperwork itself was the ceiling on growth.
The challenge
Automating this kind of work is harder than it looks. The documents are not well-suited to traditional extraction techniques since layouts and conventions vary from dealer to dealer, with mixed handwritten and typed fields. The rules behind them vary even more because every state has its own forms, edge cases, and compliance requirements. Much of the working knowledge needed to navigate these challenges lived only in the minds of long-tenured auditors, in patterns that an off-the-shelf system simply couldn’t apply.
The approach
Revel Labs started by defining a clear business outcome: lifting the paperwork cap on growth without sacrificing the accuracy and auditability the work depends on.
The work progressed in phases. First, the team tested the OCR capabilities of multiple cloud providers against the company’s actual document mix to see which performed reliably under real-world conditions. From there, the team built classification and extraction models tuned to specific document types and state requirements. These became the foundational pieces of the broader system to come.
The solution
Revel Labs built a system that combines traditional OCR with LLMs to handle the full range of documents the company processes. Classification and extraction models are in production today, reading content with greater than 90% accuracy. The broader workflow that ties them together is still being built.
Reading is only part of the work. Each document has to be checked against the relevant state’s forms, edge cases, and compliance rules. That institutional knowledge is now encoded into the system and applied consistently across volumes that no person could review manually. Auditors see only what genuinely requires their judgment while routine extraction and validation happen in the background.
The outcome
The deployed work is already on track to deliver measurable improvements:
- Auditor efficiency up more than 30%
- Rejected transactions down 75%
- Real-time validation catching errors before they reach an auditor
The paperwork ceiling is starting to lift for the company as well as for the dealers it serves.
The ongoing opportunity
The deployed work is the start. The phases still in build will extend it to:
- A combined OCR-and-LLM workflow that broadens the range of documents handled without human review
- A self-service model for dealerships, replacing shipped paperwork with submission and validation in flight
- A foundation flexible enough to absorb future OCR and LLM advances as they emerge
The paperwork that was the ceiling on growth is becoming the system that drives it.