Niches · Document parsing & extraction

Turn bank and card statements into transaction rows

No company's docs show a full fit yet. The listings below do the job with a limit, or not at all.

0companies, full fit
13companies, fit with a limit
28companies carry the tag
4share of the category's work
2first-hand accounts

What counts as done

One row per transaction with date, description, amount, sign and running balance, totals that reconcile with the statement's opening and closing balance, and the page of each row, for the whole file in one job.

How agents ask for it: get the transactions out of this bank statement, convert my statements to CSV, reconcile these statements against my expenses

Does the job, with a limit 13

It does the job, but something in the docs or a user's account will get in an agent's way. The limit is named in each row.

ListingWhy
Azure Document Intelligence B
Microsoft Corporation
prebuilt-bankStatement.us extracts transaction details and fees, but only for US bank statements in the en-us locale. source
Datalab B
Endless Labs, Inc.
No statement model; an array schema can return rows with per-field page citations, but reconciling balances is left to the caller. source
Extend API + MCP B
CrowdView Inc. dba Extend
No prebuilt statement processor; an array schema with large-array handling and per-field page citations can return rows, but balance reconciliation is the caller's. source
Google Cloud Document AI BB
Google LLC
A Bank Statement Parser extracts account and transaction entities with page anchors, but supported countries and any balance reconciliation are not stated in the processor list. source
Invofox D
Invofox, Inc. (Delaware), with a registered branch in Spain
Bank statements are a named document type, but the field list, countries covered and any balance check are not in the reference, and the error list includes ERR_TOO_MANY_PAGES at 10 pages. source
LandingAI Agentic Document Extraction BB
LandingAI, Inc.
No statement model; an array schema plus Ground can return rows with their pages, but balance reconciliation is left to the caller. source
LlamaParse API + MCP B
LlamaIndex, Inc.
No statement model; an array schema with citations can return rows, but balance reconciliation is the caller's, and row-level page references depend on the citation mode. source
Mindee API C
Mindee SAS (Mindee Inc. outside the Eurozone)
The Bank Statement template returns beginning and ending balance, statement period and a list_of_transactions with dates, but countries covered are not stated, and balance reconciliation and the page of each row are not documented. source
Mistral OCR API B
Mistral AI (RCS Paris 952 418 325)
No statement model; an annotation schema with an array can return rows, but acc-38's number-field looping is the risk here, and reconciliation is the caller's. source
Nanonets API + MCP E
Nano Net Technologies Inc.
A dedicated /api/v1/extract/bank-statement endpoint exists, but its output fields, countries and any balance check are not described in the OpenAPI file, and it is synchronous only. source
Reducto API + MCP B
Reducto, Inc.
No statement model; an array schema with citations returns rows with their source locations, but balance reconciliation is the caller's. source
Unstructured API + MCP E
Unstructured Technologies, Inc.
No statement model; an array schema can return rows, with reconciliation left to the caller. source
Veryfi API + MCP C
Veryfi, Inc.
process_bank_statement_document extracts transactions, balances and account details and maps transactions to categories, but the page of each row and any reconciliation are not shown in the SDK README, and the API docs were not read. source

Doesn't do this job 7

Listings in the category that do something else, with the reason.

ListingWhy
Adobe PDF Services / PDF Extract API C
Adobe Inc.
No statement model; tables come out as cells, and turning them into reconciled transaction rows is left to the caller. source
Amazon Textract BB
Amazon Web Services, Inc.
Analyze Lending classifies BANK_STATEMENT pages in mortgage packets, but no documented model returns transaction rows with balances; rows would come from generic tables. source
Clairic BB
Cloudflare, Inc.
No statement model, and list items can only be text, number or date, so transaction rows with date, amount and balance cannot be returned as records. source
Docling B
Docling Project (LF AI & Data Foundation)
No statement model; template extraction returns values without the page of each row or any balance check. source
Nemotron 3.5 Parse B
NVIDIA Corporation
No statement or field extraction; tables are per page with no continuation across pages. source
OpenDocRouter B
LlamaIndex, Inc.
No extraction or statement model; Markdown only. source
Xiaomi-OCR-0 C
Xiaomi Corporation
No statement mode; KIE is per image with no row pages or balances. source

6 listings in the category were added after this reading and aren't marked yet.

What stops agents finishing

  • rows split across page breaks lose their date or balance
  • no prebuilt statement model outside the US

Around it

Before: file-storage.agent-upload, mailbox-access.find-and-read. After: accounting.reconcile-bank, personal-finance.categorise-transactions.

Where this comes from

Weight 4 of 100 in Document parsing & extraction: 2 of 46 accounts (Textract, Marker); held at 4 because Veryfi, Invofox and Textract list bank statements as products, a vendor claim not an account

The same as data: /api/v1/use-cases/document-extraction.statement-transactions.json, or the MCP tool get_use_case with id document-extraction.statement-transactions. Other jobs in this category: /api/v1/niches?category=document-extraction.

Draft, read 2026-11-15 from public docs and first-hand accounts, not yet reviewed by a person. A mark says what the docs show, not what a run measured.

For companies

Do agents find, use and choose your tools?

An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.