Health plans are very good at finding care gaps. The hard part is the last mile: getting the member through to completed, documented care.
That was the theme of a recent Bright Spots in Healthcare episode sponsored by Arbiter AI. Host Eric Glazer opened with a story about a package marked “out for delivery” that never arrived. His point: a scheduled appointment, like a tracking update, is not the same as care delivered.
The panel brought four perspectives:
- Dr. Jay LaBine, Arbiter AI
- Erica Krieger, VP of Quality, Blue Cross Blue Shield of Michigan
- John Willis, Program Manager, Stars, Blue Cross Blue Shield of Minnesota
- Angela Spanos, RN, VP of Quality and Population Health, Fallon Health
Here are four lessons from the conversation.
1. Separate a care gap from a data gap
Not every open gap means care is missing. Sometimes the plan just doesn’t know the care happened.
Blue Cross Blue Shield of Michigan realized its risk adjustment team was already collecting millions of medical records a year. Using AI and natural language processing, the quality team scanned more than a million of those records for evidence of completed screenings. Colorectal cancer screening rates rose about 4% from this work alone, and the plan’s 2025 performance landed above the 90th percentile of MA plans. Krieger’s takeaway: collect once, use many times.
Fallon Health is attacking the same problem from the provider side. It recently completed its first direct EMR connection, pulling clinical data into its FHIR repository. That means cleaner gap lists, fewer medical record requests, and less provider abrasion.
2. Reach the members care management can’t
Care management capacity is finite. Willis noted that Minnesota’s program touches roughly 18–20% of members, the rising- and highest-risk groups. The other 80% still have open gaps.
Dr. LaBine shared how Arbiter tackled a version of this problem with Blue Cross Blue Shield of Michigan. The plan already had nurses calling its highest readmission-risk members. But a moderate-risk cohort of 40,000 members was too large to support manually. Arbiter was allocated 20,000 of them.
Arbiter’s AI agent, Eleanor, delivered the proven Coleman Care Transitions model at scale, using an 85/15 model: the agent handles 85% of the work, and nurses handle the rest.
| Metric | Result |
|---|---|
| Reach rate | 55% |
| Calls escalated to a care manager | ~5% |
| Members who opted into a follow-up call | 40% |
| Reach rate on follow-up calls | 88% |
| 30-day readmission reduction | 22% |
The readmission rate for reached members fell from about 16% to about 12.5%. Just as important, nurses spent their time on the members who truly needed them, doing top-of-license work.
3. Reaching a member isn’t completing the care
A scheduled appointment doesn’t guarantee the member shows up. Dr. LaBine described how Arbiter closes that gap in four steps:
- Reach — Connect with members at scale through personalized outreach.
- Book — Schedule the appointment, usually with a human in the loop, or directly in the provider’s system where integrated.
- Keep — Prepare the member with branded, personalized videos, visit prep, forms and education. Far more than a reminder text.
- Verify — Confirm completion through member self-report and claims, and reschedule anyone who missed.
The model works for annual wellness visits, PCP follow-ups and screenings like colonoscopy. It also gives plans a real-time view of where every member is in the journey.
4. Match the intervention to the problem
The panel agreed that the right response depends on why the gap is open.
Fallon Health saw home test kits returned at low rates: 14.5% for colorectal screening and 10.1% for A1C. Kits were arriving with no introduction. This year, Fallon added a provider portal message, an intro letter and two support calls. It expects at least double the return rate.
Michigan found a cohort whose blood pressure was controlled all year, then read slightly high at an in-home assessment. That cohort cost the plan about a point on a triple-weighted measure. An AI-driven self-monitoring workflow brought about 70% of them back into control. The rest were routed to their physicians.
Krieger offered a simple test for when AI fits:
- A clearly defined population
- A repeatable workflow
- A problem of scale that people alone can’t solve cost-effectively
AI doesn’t replace clinical care for complex, high-risk members. It extends the care team’s reach to everyone else.
The bottom line
Calls made and appointments scheduled are activity, not outcomes. As Dr. LaBine put it, “the measure that matters is completed care, and the Stars teams driving it are the unsung heroes of prevention.”
