Bringing AI-assisted medical coding into production
- Client
CompleteCare
- Description
Helped CompleteCare turn an internal AI experiment for medical coding into a production application.
3-4x
faster medical coding throughput
0
coding issues reported
56
seconds for processing
An opportunity to improve medical coding with AI
For more than 40 years, CompleteCare has helped healthcare organizations manage medical billing and revenue cycle operations.
CompleteCare saw an opportunity to reduce that administrative burden without compromising coding quality.
Medical coding presented persistent challenges:
- Reviewing complex clinical documentation
- Applying the appropriate codes
- Identifying missing information that can lead to denied claims or delayed reimbursement
We weren't looking for AI just for the sake of AI. We wanted to solve a problem we knew well, where getting the details right has real consequences for providers and patients.
Turning an internal AI experiment into a production application
CompleteCare began exploring AI-assisted coding internally. The early work showed promise, but building software that could reliably support real coding workflows required additional product development expertise.
CompleteCare partnered with thoughtbot to design, build, and deploy CareCodeAI.
The application analyzes clinical documentation and generates coding recommendations with supporting descriptions and rationale. It also flags potential documentation gaps that could affect medical necessity and reimbursement.
Rather than replacing coding professionals, CareCodeAI gives them information they can review and validate before making final coding decisions.
Designing for complex healthcare coding requirements
Medical coding isn't a single, uniform workflow. Requirements vary across specialties, procedures, and billing arrangements.
Working with CompleteCare's revenue cycle experts, thoughtbot developed a platform that supports professional, technical, and global coding workflows across multiple specialties.
The team also focused on making recommendations understandable. Coders can inspect the clinical rationale behind suggested codes instead of relying on unexplained results.
thoughtbot helped us take something we'd been working toward internally and turn it into a real product. They brought the product and AI expertise we needed while working closely with our team on the details that matter in revenue cycle management.
Transitioning AI reasoning models for better accuracy
As CareCodeAI moved into production, feedback revealed opportunities to improve coding recommendations. thoughtbot helped CompleteCare transition to newer AI reasoning models that provided more thorough clinical rationale.
Following the transition, CompleteCare received no new coding-related submissions through its in-product feedback system during the period evaluated. Although this wasn't a formal accuracy assessment, it was an encouraging change from earlier production feedback.
Measuring AI-assisted coding performance in production
With confidence in the tool’s impact and direction, CareCodeAI launched into production. Analysis of live production encounters showed a median AI processing time of 56 seconds, with high-volume users achieving 60–69 AI-assisted coding recommendations per hour.
These measurements reflect the application's processing performance. Coding professionals continue to review documentation, validate recommendations, and make final decisions.
Extending CareCodeAI into existing healthcare workflows
With CareCodeAI in production, CompleteCare and thoughtbot turned their attention to strengthening the underlying AI infrastructure and advancing electronic medical record integration.
The goal is to bring coding recommendations closer to the systems professionals already use, reducing the need to move between applications.
What began as an effort to improve an internal workflow has become a production platform that CompleteCare is continuing to develop in their mission to improve medical coding practices and outcomes.
