Extracting data from auto loss runs
A fleet's insurer prints each auto loss run in its own layout, listing every collision, theft and liability claim on the vehicles. Commercial auto underwriters often request several years of history with each submission, often from several prior carriers. Brokers forward the runs as PDF email attachments.
Underwriting needs each claim's loss date, cause of loss, amount paid, recoveries and total incurred, along with each year's totals. A rating workbook loads them to measure claim frequency and severity across the fleet, which sets the renewal price and the deductible on offer.
What makes auto loss runs hard
One fleet, several coverage sections
A run often reports auto liability and physical damage as separate sections for one fleet, each with its own policy ID and year totals.
Recoveries can exceed the payment
Subrogation against another driver and salvage on a written-off vehicle can return more than the carrier paid, so total incurred sits far below paid.
Cause of loss in carrier wording
One carrier prints STRUCK AN ANIMAL where another prints a coverage code, and both claims have to land in the same category for rating.
Parse, extract, verify
Parse
Parse turns the document into reading-order Markdown and a hierarchy of blocks (text, tables, figures), and grounds each block to its exact location on the page.
Extract
You define the fields in a JSON schema, and Extract returns their values in a consistent, structured format, each linked back to where it appears in the Markdown.
Verify
Each value links back to where it appears, down to the character range in the Markdown, and the page and bounding box in the source document, so every result can be verified.
One auto loss run, end to end
About this document: The Automobile Physical Damage section of a claim report from the Texas Association of Counties Risk Management Pool, posted by Cameron County with an insurance request for proposal. Its 4 landscape pages list claims on the county's vehicles under five policy years. Each year opens with a policy header and closes with a Totals row. Page 1 is featured. The liability pages, which name private claimants, were removed before publication.
Processing objective: The objective is to follow one physical damage claim from its policy header down to the Totals row for its year. The claim is an animal strike, and its $13,043 total incurred is the largest of that year. It rolls into the $17,240 printed for the year as a whole.
Demonstration fields: Policy ID, Claim number, Cause of loss, Claim total incurred, Policy-year total incurred
The document
4 of 4 pages
Page 1shown below
Page 2
Page 3
Page 4
Where each value came from
policies[1].policy_id
{
"value": "RMP-00370-201107",
"page": 1,
"range": {
"end": 1011,
"start": 984
},
"box": {
"xmin": 0.21536,
"ymin": 0.2955,
"xmax": 0.35817,
"ymax": 0.31367
}
}policies[1].claims[1].claim_number
{
"value": "APD201102084-1",
"page": 1,
"range": {
"end": 1266,
"start": 1252
},
"box": {
"xmin": 0.02248,
"ymin": 0.33035,
"xmax": 0.18898,
"ymax": 0.34667
}
}policies[1].claims[1].cause_of_loss
{
"value": "STRUCK AN ANIMAL",
"page": 1,
"range": {
"end": 1421,
"start": 1405
},
"box": {
"xmin": 0.74533,
"ymin": 0.33053,
"xmax": 0.96859,
"ymax": 0.34685
}
}policies[1].claims[1].total_incurred
{
"value": 13043,
"page": 1,
"range": {
"end": 1341,
"start": 1334
},
"box": {
"xmin": 0.42521,
"ymin": 0.33045,
"xmax": 0.48495,
"ymax": 0.34674
}
}policies[1].totals.total_incurred
{
"value": 17240,
"page": 1,
"range": {
"end": 5882,
"start": 5875
},
"box": {
"xmin": 0.37613,
"ymin": 0.76985,
"xmax": 0.48543,
"ymax": 0.78819
}
}How to try it yourself
Everything on this page is reproducible. The sample, its schema, and the Parse and Extract API outputs are published. You can repeat the same two calls from whichever surface you already build on.
- Claude Skilldescribe the job; the agent writes it
- CLIno code at all
- MCPexpose ADE docs as an agent tool
- REST APIany language
- Pythonlandingai-ade
- TypeScriptlandingai-ade
Documents into data
Accurate
The best accuracy on the difficult documents other tools get wrong, such as messy scans, complicated tables, checkboxes and handwriting.
Auditable
Every answer cites its sources. Values come back with an audit trail (page, bounding box, character range) so a human can review them.
Agentic
Applies multi-step reasoning, and plans before it acts. Handles documents it has never seen before and adapts to variations.
API-first
Composable APIs that drop into any tech stack and can be called from anywhere. Easy to build on and embed in your pipelines and products.
Affordable
Cost-competitive rates with complete price transparency. Spend controls before you run, and every job reports its exact cost.
Enterprise-grade
Built for production and scale from day one. Fully observable and monitored, running securely in our cloud or yours.
Built for high-volume, high-throughput production
- Zero data retention
- Priority rate limits
- Enhanced support
- VPC and on-prem deployment
Available on Team and Enterprise plans. VPC and on-prem deployment require an Enterprise plan.
