Prior auth review cut from 4.2 days to 6 hours: a BlueCross BlueShield AI automation case study

Challenge

The prior authorization team handled 42,000 requests monthly but averaged 4.2 days per review. Urgent requests required a 72-hour turnaround, but the team met that deadline only 61 of every 100 times. Hiring more staff to fix this would have cost $2.8 million a year.

Approach

We created an AI system to sort incoming requests based on urgency, complexity, and completeness. Simple cases went straight to automated approval, while complex ones were sent to senior reviewers. We trained the model using 18 months of past decisions and set up a feedback loop so reviewers could immediately fix any wrong classifications.

Outcome

We cut average review time from 4.2 days to 6 hours for cases handled by AI. Urgent requests were on time 97 of every 100 times. Without adding staff, the team processed 18,000 more cases monthly, saving $2.8 million annually in labor. The AI triage managed 3,400 out of 5,000 cases completely on its own.

Every month, we handle 5,000 cases. And guess what? We cracked 3,400 of them without needing a single hand to jump in.

So, we took a deep dive into 18 months of past decision data to teach the AI triage system how to handle requests. In the end, it managed to take care of 3,400 out of 5,000 incoming requests all by itself, thanks to the clear clinical rules we put in place. That meant the reviewers could focus their energy on the 1,600 cases that really needed human attention.

Catching and Fixing Data Errors by Keeping Close to the Process

So, here’s what we did: when the AI spit out recommendations, our reviewers jumped in to fix any hiccups immediately. Then, we fed those corrections straight back into the model. That little hands-on touch pushed accuracy from 84 of every 100 at launch up to 91 of every 100 by week 16. And we’re not stopping, we keep updating the model as new data comes in.

How we made sure compliance never slipped through the cracks

The compliance team kept missing the 72-hour deadline for urgent requests. So, we built a triage system that bumps those urgent ones straight to the front of the queue. No matter when they come in, they skip the line. That way, the fast-tracks actually stay fast.

Results

• 4.2 days to 6 hrs Average review time
• $2.8M Annual headcount cost avoided
• 18,000 Additional cases per month
• 97 of 100 Urgent request compliance

Related case studies

• We automated the pharmacovigilance data pipeline, cutting adverse event processing from 18 days to 4 hours.
• We consolidated 14 regional patient encounter definitions into one standard in 11 weeks, cutting reconciliation labor costs by $2.1 million.