A first-pass claims adjuster for Claude

It reads the claim
and makes the call.

Claimwise reads an incoming protection-plan claim and makes the decision a trained adjuster would make on the first read — then drafts the customer response. It judges a claim by its cause and incident date, not its symptom and filing date. Built for the program without a staffed claims team.

It adjudicates — it never asks "what do you want to do?"  APPROVE · REQUEST · DENY · ESCALATE, cited every time.

Watch it decide

Watch it make the call, in real time.

No narration, no edits. Three claims in, three different decisions out — cause read under the symptom, clause cited every time.

Three claims · three calls · 9:16
The problem

Claims arrive faster than you can read them carefully.

A small protection program, no staffed claims desk — and an inbox of claims that each take a careful read to call correctly.

So the reads get fast, and fast reads cost both ways. Approve what's actually excluded and you bleed margin on every miss. Deny what's genuinely covered and you lose the customer — and maybe earn a regulator's attention. The hard part was never the easy claims. It's reading the cause under the symptom, and the incident date under the filing date.

Approve the excluded → margin gone, claim by claim
Deny the covered → churn, complaints, scrutiny
=
A consistent, cited first pass → the human only touches what truly needs them
What it does

Every claim leaves with one decision.

No "here's what I found, what now?" REQUEST INFO is a decision — it routes the claim to pending-customer with a precise ask. DENY always cites the clause. ESCALATE is rare, reasoned, and never an accusation.

APPROVE

Covered, documented, in-window. Drafts the approval and the remedy — repair, replace, or reimburse.

REQUEST

Likely covered, one thing missing. Asks for exactly that — specific enough to finish in one reply.

DENY

Clear exclusion or out-of-window. Cites the exact clause — and names a path forward when one exists.

ESCALATE

Fraud, high value, or true ambiguity. Flags it, holds a safe default, asks one question — no accusation.

See it decide

Three real claims. Three different calls.

The symptom is a trap; the cause decides. So a cracked screen can be denied, and a "filed late" claim can still be paid — depending on what actually happened.

Real outputs — produced by the folder running in Claude. The logic was gate-tested on unseen claims, too.

Defect · Protect+ · $520
Laptop won't power on, no damage. Failed 4 days ago. Diagnostic: motherboard failure. Docs complete.
✅ APPROVE · replace
Covered defect, in-window, fully documented. Board repair exceeds the repair threshold → comparable replacement; prepaid label to follow.
Covered cause, within authority; repair > 70% of value → replace per the remedy matrix.
Accidental · Protect (base) · $410
"Cracked my screen when the phone slid off the couch." Photo attached, 2 days ago, in-window.
❌ DENY · cited
The base plan covers defects only; a drop is accidental damage (§Exclusions). Denied on cause — with the honest Protect+ upgrade path named.
Symptom (cracked screen) is irrelevant; the cause is excluded on this tier. Cite the clause, leave a door open.
Replacement · Protect+ · $900
Phone "stopped working." 3rd claim in 6 weeks; serial doesn't match the last unit we shipped.
🔺 ESCALATE
Fraud signal (velocity + serial mismatch) and over auto-authority. Paused, not denied; customer told only "in review." One question handed to the human.
Honest action on suspicion is escalate-and-verify — never accuse-and-deny, never rubber-stamp $900.
How it works

A folder you teach once. Then it adjudicates.

Drop in the folder

Put Claimwise into a Claude Project. Claude becomes the adjuster — identity, rules, examples, and the policy reference all loaded.

Set your policy, once

Fill the reference/ templates with your real plan terms, exclusions, evidence requirements, and fraud red-flags. Claimwise is only as correct as that file set.

Paste any claim

It returns one decision, the drafted customer response, and an internal one-line "why" — the clause plus the facts — you can audit or override in seconds.

The methodology

Folders as architecture. Each file does one job.

claimwise/
├── identity.md      # who it is, what it refuses
├── rules.md         # the decision logic (the heart)
├── examples.md      # worked claims + edge cases
├── README.md        # how to use it
└── reference/
    ├── coverage-policy.md   # tiers, exclusions, window
    ├── evidence-checklist.md# what each claim needs
    ├── decision-matrix.md   # cause × tier → outcome
    ├── fraud-signals.md     # flag, never accuse
    └── response-templates.md# the drafted replies

The logic short-circuits: fraud and high-value claims reach a human first, then eligibility, then the cause-to-coverage map, then the window, then documentation. The first step that produces an outcome wins.

Nothing is a black box. The policy is plain Markdown you can read, edit, and audit — so every decision cites a clause you can point to.

For builders

Want to build your own operator?

Claimwise is one worked example. The Operator's Handbook is the method behind it — how to turn Claude into something that decides instead of chats.

The 5-file structure, and how to write decision logic with real thresholds — not "use good judgment"
How to handle the edge cases that break most tools (cause-not-symptom, the buried exclusion), and keep the human gate where it matters
New operators as they ship from The Quiet AI — delivered with the handbook
Get the Handbook — free

A short, practical guide. No spam, unsubscribe anytime.

The folder itself is free and ungated — grab it on GitHub, no email required. This is just for the build-your-own guide.
No catch

It's a folder, not a SaaS.

Claimwise is free. No account, no claims platform to license, no per-seat pricing. It's a folder of plain Markdown you drop into Claude — running in a minute, and yours to fork and shape to your own policy.