Premera Blue Cross

Reverse chronological. Each entry: what happened, what shifted, what came out of it.

---

Date: Jan 28, 2025, 4:00 PM EST People: Thomas, Michael ↔ David Braza, Romilla Batra (Colt expected but didn't join) Type: Video call Source: Transcript

Short pivot call. Dave and Romilla confirmed interest: "the short answer is probably yes, we want to expand our AI capabilities." Handed us off to the operational team.

Key points:

• 3 AI program domains at Premera: [unnamed], Claims, Services
• Colt runs UM domain (~12 projects)
• Chad Murphy (CCO, pharmacist background) is the business leader for UM operations — we haven't met him yet
• Jamie Halstead runs "the shop" alongside Colt, already on AI/automation path
• Romilla already briefed Chad about us
• Dave: "do some sort of contract with you all and expand our capacity"

What shifted: From "are they interested?" to "how do we plug in?" The handoff to Chad/Jamie is the real gate now.

Action items:

• Beth scheduling call with Chad, Jamie, Colt + Thomas, Michael
• Prep for that call — understand the 12 projects, find our lane

---

Date: Jan 5, 2025, 12:30 PM EST People: Thomas, Michael ↔ David Braza, Colt Courtright, Romilla Batra (Kristina Rose invited but couldn't attend) Type: Video call — product demo + discussion Source: Transcript Intro: Connected through Brad Flugel (Wharton healthcare professor)

Thomas demoed the product: patient queue, PDF upload → AI-generated clinical summary + timeline, AI flags (discharge delay, inconsistent documentation), and the dynamic criteria agent. Used a 30-page CHF 5-day stay as example.

Romilla's key context — the 3 UM milestones:

1. Initial prior auth (patient arrives, admit/observe/not covered?) — most important to them. This is where they want help first. Runs through InterQual to apply criteria.
2. Concurrent review (day 5 check — still meet criteria?) — what Thomas demoed. They review at day 5 because DRGs cover days 3-5 and they lack bandwidth for earlier.
3. Daily reviews — for contracts requiring daily authorization.

Romilla was clear: the initial prior auth determination is the sentinel event. Concurrent is important but secondary. Whatever we build has to go through evidence-based criteria (InterQual) — can't use AI-generated criteria or "we'll be in a hot spot."

Colt's deep technical context:

What they've already built:

• Already receiving live ADT feeds, processing them with AI
• Fully automated non-inpatient reviews to free bandwidth for inpatient (which is more complex)
• Auto-auth eliminates manual work on 25-50% of all requests
• AI accuracy equivalent or greater than human clinicians for approvals
• Can automate approvals end-to-end (including letter generation), but NOT denials (regulatory)
• Summary prediction creates prioritization hierarchies across admissions in multiple states
• ADT feeds update ~every 15 minutes — very chatty HL7 messages
• Arranging 80-90% of admissions where they have ADT feeds (primary markets: WA and AK). CCDs much sparser.
• Claims data gives broad (not deep) clinical history — diagnosis and therapy, not allergies/family history
• ADT feeds more important than CCDs for payer use case
• They've been building for ~2 years, deploying to production

What they need:

• Talent/capacity to apply rapidly changing AI technology — their bottleneck
• Had a 2-year roadmap, moved some year-1 items to year-2 due to capacity constraints
• Specific year-2 items: insourcing for imaging (ortho) and lab — auto-auth in specific domains, removing vendor expense
• Mentioned vendor "Caroline" for advanced imaging reviews as potentially disruptable

Architecture/integration points:

• They manage everything within their own environment and tools
• Suggested we might not need a UI — could work via APIs and data structuring behind the scenes
• Would need to connect to their engines (InterQual, MCG) via APIs
• Breaking contract logic into criteria that bands and prioritizes for nurses
• Multiple inter-system connections required: contracting standards, DRG/non-DRG, facility LOS, clinical criteria

Colt's questions about us:

• Volume of data we can handle?
• Predictive accuracy?
• Hallucination controls?
• Operating own infrastructure or could work in their environment?

Business model alignment:

Thomas positioned as FDE/consulting: "roll up our sleeves, build things for people, interview them, figure out issues." Not SaaS.

All three Premera people agreed:

• Dave: "the notion that you would create off-the-shelf software that we could take and install is probably highly unlikely for any payer"
• Colt: "reusable agents can be used across clinical parsing, case work, appeals"
• Dave: "the future is less vendors, less SaaS, more internal development... a lot of vendors are going to be disrupted and it's a good thing for American healthcare"

New people mentioned:

• Brad Flugel — Wharton professor, made the introduction
• Kristina Rose (kristina.johnsonrose@premera.com) — invited but couldn't attend
• Mark — vendor management at Premera (contracting goes through him)
• Talhah/Tal — CFO at Premera (funding decisions)
• Chad Murphy — CCO (mentioned, not on call)
• Jamie Halstead — (mentioned, not on call)
• Caroline — vendor for advanced imaging reviews

What shifted:

• We now understand Premera is NOT early stage with AI — they've been building for 2 years, have production systems
• Our value isn't teaching them AI, it's augmenting their capacity to move faster
• Initial prior auth is their #1 priority, not concurrent review (which is what we demoed)
• The UI is probably not the play — backend/API integration into their existing tools is
• Colt specifically floated accelerating year-2 roadmap items as the opportunity for us

Action items (from this call):

• Thomas to ping Beth to schedule end-of-January follow-up → became the Jan 28 call
• Premera internal: Dave, Colt, Romilla, Jamie, Chad to review roadmap and identify what we could accelerate
• Explore contracting through Mark (vendor management)
• Potentially involve Talhah (CFO) for funding

---

People: Thomas, Michael ↔ Dave Braza, Colt Courtright Type: Intro call(s)

Connected through Brad Flugel (Wharton). Dave and Colt had a brief (~30 min) initial call. Enough interest to schedule the demo.