Skip to content
woodfrog
AI
InsuranceUse case, no client result

Clear the claims document backlog in hours, not days.

Every claim arrives as a pile of papers: a form, bills, reports, photos, scans of IDs. Your handlers read each page and type what they find into the claims system. The pile grows faster than they can read it. A document agent reads every page first, so your people spend their time deciding.

One claim fileSchematic
4.4×Target return in 12 months
Week 1Fixed-fee audit
6 weeksTo the first live result
3Senior people on the work
TypeUse case: what we build, not a client story
ForInsurers, TPAs and brokers whose claims teams work through document-heavy files
EngagementWeek 1 audit, then build in about six weeks, fixed scope, fixed fee
Runs onClaude, with your claims system and document store, in your own cloud
Before and after

From a pile of papers to a file ready to decide.

Before
  • Handlers read every page by hand
  • Details retyped from scans into the claims system
  • Missing papers found days later, on first review
  • Files wait in a queue nobody can see into
After
  • Every page read as soon as it arrives
  • Details extracted, with the page they came from
  • Gaps and mismatches listed on day one
  • A person reviews a drafted file and decides
The problem

The backlog is a reading problem.

Most of the time in a claim is not spent deciding. It is spent reading. A handler opens the file, works through every page, checks the amounts against the bills and the dates against the policy, and types it all in. Then they find a missing paper, write to the claimant, and the file goes back in the queue.

A tall stack of paper files tied with string
Every claim is a pile of papers
What was broken
  • Documents arrive in every form: typed, scanned, photographed on a phone, sometimes in Hindi or another regional language.
  • Handlers retype the same details from paper into the claims system.
  • Checks are done by eye: amounts against bills, dates against the policy period.
  • Missing or mismatched papers are found late, so the claimant is asked twice.
  • Volume comes in waves, and the queue only grows when it does.
You might be here if
  • Your claims backlog is measured in days.
  • Your experienced handlers spend their time typing.
  • Claimants call to ask where their claim has got to.
  • You cannot see which files are stuck, or why.

Why it mattered. A slow claim costs twice: in handler hours, and in a customer who remembers how long it took.

The approach

Let the agent read. Let people decide.

We start with your files. In Week 1 we take a sample of real claims and write down what a handler looks for on each kind of document, and which checks they run. That becomes the agent's job: read, extract, check and draft. It never approves, rejects or settles a claim. Every decision stays with a person, who sees what the agent found and the page it found it on.

A person reviewing an insurance form with a pen
Every decision stays with a person
What comes in
Claim form
Bills, reports, photos
Policy record
woodfrog method
Read and extractEvery page, with its source
CheckAgainst the policy and each other
DraftFile note, gaps, next question
What changes
Gaps found on day one
No retyping
A person reviews and decides
Backlog cleared in hours
Deliberately not doneLetting the agent approve, reject or settle a claim
People decide every claim. Claude reads, extracts, checks and drafts.schematic
The solution

A document agent in front of every handler.

What we build
  • Intake that takes every document as it arrives, from email, upload or scan, and sorts it by type.
  • Extraction of the details your handlers retype today: names, dates, amounts, policy and claim numbers, each linked to the page it came from.
  • Checks your team already runs, written down and applied to every file: amounts against bills, dates against the policy period, names across documents.
  • A drafted file note for the handler, with a list of gaps and the next question to ask the claimant.
  • A review screen where a person accepts, corrects or overrides every field, and every change is kept.
  • A live view of the queue: which files are ready to decide, which are waiting on papers, and for how long.
Built on Claude by woodfrog, an Anthropic Certified Services Partner. Documents stay in your own cloud, personal data is masked where it is not needed, and nothing is decided without a person.
How it would runIt starts with one claim type, then adds more. Operate is open-ended and drawn at one week only to fit.
The impact

What we would measure.

Where a handler's time goes, before and afterReading and retyping move to the agent. Deciding stays with the person.
Schematic, no values
Bar lengths show the shape of the change, not a measurement. We measure your own split in Week 1.
What we would measure on a real engagementWhat we would measure
Time from documents in to file readyPer claim, by claim type
Claims waiting in the backlogAnd how long the oldest has waited
Fields corrected by a handlerTracked, so the agent is checked every day
Gaps found on day oneAgainst gaps found later in the claim
Return on the workTarget: at least 4.4× within 12 months, measured on your own numbers
Hands working through an insurance form on a clipboard
A file ready to decide
Clear the claims backlog.A short call, or record a message. A person replies within one business day.
Contact usCertified Services Partner, Claude Partner Network