Extracting data from investment statements
A custodian sends every client a statement each month, covering as many accounts as they hold and running from a few pages to a few hundred. Every firm lays them out its own way. Wealth managers and aggregation platforms ingest them in bulk, from client uploads, portal downloads and scans of posted paper.
A reader needs each position, its quantity, its cost basis and its market value, attributed to the right account. Portfolio systems reconcile those against their own books, and tax preparation works from the realized gains they imply.
What makes investment statements hard
The signs carry the meaning
A short position prints a negative quantity, market value and cost basis, with a positive gain. One dropped minus sign changes the answer.
Footnote markers sit on the numbers
The cost basis is printed -14,510.99 with a superscript c pointing at a footnote. Extraction has to decide which characters belong to the number.
One holder, several accounts
Three accounts share this document. A security can appear in two of them with a different basis, so every row has to carry its account.
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 investment statement, end to end
About this document: Fidelity's own sample monthly statement, published on fidelity.com so customers can see what one looks like. 28 landscape pages covering an individual brokerage account, a traditional IRA and a 529. It carries holdings tables for stocks, bonds, mutual funds and ETFs, plus three closing disclosure pages stored rotated 90 degrees. All the personal data is specimen: the account holder is John W. Doe.
Processing objective: The objective is to lift one complete position off page 7, signs intact. The row chosen is a short sale, the hardest on the page, and it reconciles: -13,710.00 minus -14,510.99 gives the 800.99 unrealized gain.
Demonstration fields: Security (short position), Quantity, Ending market value, Total cost basis, Unrealized gain
The document
4 of 28 pages
Page 1
Page 4
Page 7shown below
Page 18
Where each value came from
stock_holdings[2].description
{
"value": "ENSTAR GROUP LIMITED COM STK USD 1.00",
"page": 7,
"range": {
"end": 15952,
"start": 15905
},
"box": {
"xmin": 0.03841,
"ymin": 0.35575,
"xmax": 0.27168,
"ymax": 0.39248
}
}stock_holdings[2].quantity
{
"value": -100,
"page": 7,
"range": {
"end": 15968,
"start": 15961
},
"box": {
"xmin": 0.2716,
"ymin": 0.35574,
"xmax": 0.41537,
"ymax": 0.39246
}
}stock_holdings[2].ending_market_value
{
"value": -13710,
"page": 7,
"range": {
"end": 16002,
"start": 15992
},
"box": {
"xmin": 0.49652,
"ymin": 0.35573,
"xmax": 0.60613,
"ymax": 0.39244
}
}stock_holdings[2].total_cost_basis
{
"value": -14510.99,
"page": 7,
"range": {
"end": 16027,
"start": 16011
},
"box": {
"xmin": 0.60605,
"ymin": 0.35572,
"xmax": 0.70445,
"ymax": 0.39243
}
}stock_holdings[2].unrealized_gain_loss
{
"value": 800.99,
"page": 7,
"range": {
"end": 16048,
"start": 16042
},
"box": {
"xmin": 0.70436,
"ymin": 0.35571,
"xmax": 0.81469,
"ymax": 0.39243
}
}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.
