U.S. Environmental Protection Agency (EPA)
Reviewing public comments is a legal requirement that is time-intensive and typically manual. The EPA was in need of a streamlined workflow to review and organize large volumes of comments about proposed environmental regulations.
An AI review system that augments human judgment without replacing it, reduces time spent on routine tasks, focuses human effort where judgment is critical, and maintains a full audit trail that can be accessed as needed.
When the EPA proposes a new rule, the public can comment on it using Regulations.gov. Each posted comment must be reviewed, categorized by topic, and considered before the proposed rule can be finalized. Depending on the proposed rule, there can be thousands of comments to review.
Before Cadmus built an AI-assisted workflow, each comment was reviewed by at least two humans who independently assigned topic codes. If the first two did not agree, they would meet to discuss and come to a conclusion. On the rare occasions that it was necessary, they would request a third reviewer to decide the best topic code. Then, Cadmus and their subcontractors had to manually transfer comments into third-party software that organized comments into topic-based reports. As a result of this process, Cadmus wasn’t able to provide the client access to the organized comments until the coding and transfer were nearly finished, which impacted the amount of time the client had to formulate their responses.
Cadmus implemented an AI-assisted public comment review workflow. The workflow is similar to the previous process, but it automates the most time-consuming steps.
Key elements of the new process include:
It is important to remember that every comment is still reviewed by at least one human even in this AI-assisted workflow.
The Cadmus team reviewed the results of the workflow to ensure accuracy and reliability. During these reviews, they found the new workflow saved weeks of effort, as well as the following stats:
This project shows that meaningful AI impact doesn’t require completely removing existing processes or cutting out expert reviews. The key is to start with a high-volume, rule-driven workflow where teams spend significant time on repetitive preparation, classification, or reconciliation work. At the same time, you should preserve the core human decision points, then introduce AI as a parallel reviewer or accelerator.
Our client work is often the first and largest of its kind—bar-raising, barrier-breaking, and goal-exceeding. Our dedicated experts are ready to help you achieve lasting results.
