Stratum Collective

Precedent-based denial intelligence
for behavioral health

Demo Day 2026

0%

of appealed denials are overturned

0%

of denied claims are ever appealed

The gap between those two numbers is the market.

The AI Paradox

0%

of healthcare orgs use AI for RCM

0%

of providers now at 10%+ denial rates
up from 30% in 2022

More AI. Worse outcomes on the hard cases.
AI solves volume. Nobody solves evidence quality.

The Insight

How RCM AI works today

Keyword matching.
Surface similarity.
“This denial looks like that one.”

How denials should resolve

Like case law.
Each decision builds on prior rulings.
Precedent compounds.

Precedent Objects

Context PHP step-down · Anthem TX · ASAM 2.5 → 2.1 · Medical necessity

Evidence Warm handoff note + PHQ-9 delta + attending attestation + <72hr gap

Outcome Overturned on first appeal · 83% win rate for this pattern · v3.2

Governance Human-readable · Versionable · Editable · Full attribution chain

How It Compounds

L3

Reasoning
Generate defensible appeal strategies with win probabilities.

L2

Structured Knowledge
Knowledge graph over precedent. Decisions reference prior decisions — like case law.

L1

Retrieval
Surface similar prior cases with outcomes. Working at launch.

Each layer makes the next possible. The corpus gets more valuable at every layer.

Architecture

Materialized Aggregation Pipeline — privacy by structure, not policy

Tenant Layer

/api/precedents

Isolated CRUD per facility

precedent_objects

Raw POs · Per-tenant · PHI present

🔒 No marketplace access

⚙️ Pipeline

Aggregate
De-identify
Score

← PHYSICAL BOUNDARY →

Marketplace Layer

/api/marketplace

Reads only derived intelligence

marketplace_clusters

Aggregated patterns only
No raw POs exist here ✓

+ credit_ledger · contribution_scores

Single deployment. Data leaks are structurally impossible — marketplace routes physically cannot access raw patient data.

The Wedge: Behavioral Health

2–3×

higher denial rates than general acute care

Beachhead: SUD → IOP → PHP

Highest interpretation variance. Fastest precedent compounding. Weakest competitive coverage.

$127M–$235M

SAM in BH denial recovery

They build faster typewriters.
We build case law.

Precedent-based → ← Throughput Judgment-driven ↑ Rules-based ↓

Waystar · R1 · AKASA · FinThrive CombineHealth

Stratum

Business Model

Lite Sprint

$15–25K

Single cluster · 25–50 denials
5–10 precedents · 45–60 days

Full Sprint

$50–75K

Multi-cluster · 50–100 denials
10–20 precedents

Intelligence Platform

$120–200K/yr

500+ precedent library
Drift updates · Semantic search

Fixed fees. No contingency. Measurable improvement in 60–90 days.

Traction

ARPA-H

V11 Application Submitted
Three technical areas aligned to our architecture.

H1–H3

Hypothesis Validation
H1 (Knowledge Asymmetry): Validating · H2 (Speed): Designed · H3 (Corpus): Modeled

Pipeline

Pipeline
[Pipeline state — update before presenting]

Team

[Founder Name]

[Role]

[One-line credential]

[Founder Name]

[Role]

[One-line credential]

The Ask

[$ Amount]

[Round type]

Use of Funds

[Allocation breakdown]

[contact email] · stratumcollective.co