01 · The intake
A claim is mostly unstructured evidence — with a coverage map laid over it.
The structured side of a claim — policy number, coverage limits, deductibles, named insureds — is the part that fits in a database. It is also a small fraction of what's actually in a claim file. The rest is unstructured: voice, image, scanned document, narrative. Every carrier that has tried to "AI-ify claims" has built a small army of brittle extractors against this pile, each one trained on its own labeling, none of them governed. Essence treats the unstructured intake as a substrate concern — governed at the surface, before any coverage map gets drawn.
What's actually in the file
Unstructured · most of the claim
FNOL
First notice of loss audioThe claimant's account, often from roadside. Tone, hesitation, fact pattern, witnesses named.
IMG
Photos of damageCurbside shots, EXIF-stamped, often partial, often poorly lit. Vehicle angles, point of impact, surrounding scene.
PDF
Repair estimatesBody shop printouts, scanned at varying quality. Line items, parts, labor hours, supplements.
DOC
Police and incident reportsScanned, sometimes faxed. Officer narratives, citations, diagrams of the scene.
TXT
Adjuster notes & medical narrativesFree-text observations, treatment summaries, claimant statements over time.
Most of the claims budget is spent translating this pile into something a coverage decision can be made against — and the translation layer is rebuilt at every carrier.
What fits in a database
Structured · the coverage map
POL
Policy number & named insuredsThe contract anchor.
LMT
Coverage limits & deductiblesPer-coverage caps, per-occurrence and aggregate.
END
Endorsements & exclusionsThe clauses that bend the coverage map.
JUR
JurisdictionState of loss · state of policy · regulatory regime.
The coverage map is the smaller, cleaner half of the file — and it's the half every existing claims system over-indexes on, because it's the half that fits in tables.
02 · The validator
The Coverage Map Validator surfaces a map. It does not determine coverage.
This is the doctrinal line that defines the AutoClaims vertical. Surfacing a coverage map — laying the policy clauses against the unstructured evidence and showing where things appear to align, conflict, or need additional information — is a substrate job. Determining coverage is a licensed-adjuster job. State licensing regimes, unfair claims settlement practices statutes, and bad-faith case law all turn on who actually decided. Essence makes the boundary structural: the validator is labeled, audit-recorded, and statutorily framed as adjuster review only.
What the validator does
Coverage Map Validator · adjuster review
- Surfaces the map. Lays policy clauses against the unstructured evidence and shows where they appear to align.
- Cites the spans. Every claim about evidence carries a hashed source span back to the original document, photo, or audio.
- Flags ambiguity. Conflicts, missing evidence, and clauses requiring interpretation are highlighted for the adjuster — not resolved.
- Records the chain. Every surface, review, and decision is a first-class event on the substrate.
- Stays inside its scope. The validator's output is a review map. It is labeled as a review map. It is logged as a review map.
What the validator does not do
Coverage determination · adjuster only
- Determine coverage. Coverage decisions are made by the licensed adjuster, on the substrate, recorded as a separate event.
- Approve or deny payment. Payment authority sits with the adjuster and any required supervisory chain — never with the substrate.
- Bind the carrier. No Aptiv output is binding on the carrier or the claimant. The substrate is review infrastructure.
- Substitute for licensed judgment. Bad-faith and unfair-claims-practices liability turn on who decided. The substrate makes the answer to that question unambiguous: the named adjuster.
- Operate outside its label. If a downstream system tries to use validator output as a determination, the audit trail shows it.
The licensed adjuster determines coverage. Always. The substrate's job is to make that determination faster, more consistent, and more defensible — by giving the adjuster a governed map of the unstructured evidence, with every claim cited and every step recorded.
03 · The specs
14 Aptiv Specs. Fully grounded. One labeled adjuster-review map.
The AutoClaims build is honest about the line between detection and determination — and the specs reflect that. Thirteen are approved outright. One — the Coverage Map Validator itself — is approved with notes specifically requiring it to be labeled as adjuster review map and never as coverage determination. That note isn't a defect; it's the doctrine made explicit at the spec level.
1
Approved with adjuster-review notes
0
Fabricated source claims
FNOL Intake Surfacing
Approved
Photo Damage Surface Mapping
Approved
Repair Estimate Extraction
Approved
Police Report OCR & Surfacing
Approved
Medical Narrative Extraction
Approved
Adjuster Note Indexing
Approved
Policy Document Parsing
Approved
Endorsement & Exclusion Mapping
Approved
Jurisdiction Resolver
Approved
Evidence Span Hash & Audit
Approved
Conflict & Gap Surfacing
Approved
Subrogation Signal Surfacing
Approved
Fraud Indicator Surfacing
Approved
Coverage Map Validator
Adjuster Review Map
04 · The gap
Substrate-level governance is categorically different.
Every claims AI program built outside a governed substrate ends up rebuilding the same translation layer — extracting the unstructured evidence, normalizing it, mapping it against policy, surfacing it for review, recording the chain. Each one is brittle, each one is bespoke, each one falls over when the regulator asks who actually decided. Essence resolves intent once, in Synergy, then governs intake across every spec in the cluster. The numbers below are the architectural consequence.
Aptiv Specs · approved
0
13 approved outright · 1 labeled adjuster review map
Coverage determinations by Aptiv
0
Determinations are the licensed adjuster's · always
Fabricated source claims
0
Hashed evidence spans · creator-reviewed · grounded in claim file
Time to admissible review
0
Aptivs come up admissible by construction · spec-defined, not bolted on
Claims AI programs
Each carrier rebuilds the translation layer in-house — bespoke extractors per document type, no shared substrate, no governed evidence chain. The unstructured side of the claim file gets handled with brittle pipelines that don't survive a litigation request.
Claims AI programs
"Detection" and "coverage determination" collapse into the same model output. When a regulator or a bad-faith plaintiff asks who decided, the answer is a model — and the carrier's defensible chain disappears.
Essence
Unstructured intake is governed at the surface — every evidence span hashed, every spec grounded in the claim file, every review map labeled as a review map. The validator stays inside its scope by construction.
Essence
Synergy resolves intent once, then propagates across the spec cluster. The licensed adjuster determines coverage — always — and the substrate makes that determination faster, more consistent, and more defensible by giving the adjuster a governed map of the evidence.
05 · What's next
AutoClaims is live.
All 14 Aptiv Specs are approved, with the Coverage Map Validator explicitly labeled as adjuster review. Carriers and TPAs evaluating now. The full briefing — architecture, patents, performance evidence — lives in the investors section on mindaptiv.com.