Project One
We were brought in to help a large healthcare provider explore how AI could improve the accuracy and speed of pre-authorization decisions.
What we did
Our first step was a short discovery phase: we mapped the existing process, identified the highest-impact decision points, and agreed on success metrics that mattered to both the business and compliance teams.
Strategy and engineering were aligned from day one. We ran two-week sprints where each sprint had a clear hypothesis, a small experiment, and a measurable result.
Key outcomes
- Proved we could predict approval likelihood from structured claim fields with acceptable accuracy
- Integrated with their staging environment and showed we could surface recommendations in under two seconds
- Delivered a working prototype that operations could pilot in a single region within three months
- Built guardrails so the system would defer to human judgment when confidence was below a threshold
Technical approach
The technical architecture was designed for safety and explainability from the start. Every recommendation was tied to a set of contributing factors that clinicians could review. Governance, versioning, and audit trails were part of the first release, not an afterthought.
Post-launch we stayed involved to monitor model performance and retrain when new policies or code sets were introduced. Today the system handles a substantial share of pre-auth decisions in that region, with human reviewers focusing on the cases that need judgment.
