AI document extraction for bookkeepers

Dynamite Docs
Source → result
Bank statements: A bookkeeper’s reviewSource reference: BANK-SEP p2 r4. Date: 2026-09-04. Description: Office supplies. Illustrative example, not extracted customer data.SOURCE DOCUMENTBank statementsSource referenceBANK-SEP p2 r4Date2026-09-04DescriptionOffice suppliesA bookkeeper’s reviewfx Values from your documentASource…BDateCDescription12345BANK-SEP p2 r42026-09-04Office suppliesCARD-SEP p1 r82026-09-09Supplier refundDocument fields / RowsCheck the rows before closing the month.
Source reference
BANK-SEP p2 r4
Date
2026-09-04
Check the rows before closing the month. Illustrative example.

Prepare month-end transaction rows from bank statements, card statements, and receipts. Keep the source beside the table while you check dates, signs, and balances.

Try a document extraction

Answer in 30 seconds

Prepare month-end transaction rows from bank statements, card statements, and receipts. Keep the source beside the table while you check dates, signs, and balances.

A month-end folder rarely contains one tidy format. Bank PDFs use different debit and credit conventions. Card statements mix purchases, refunds, fees, and payments. Receipts may show a tax amount that the statement does not. Dynamite Docs extracts draft fields and rows from those documents so you can review the underlying evidence before you move data into a workbook.

Start with one statement and record its period, opening balance, closing balance, and expected transaction count. Keep a separate table for each account and currency. A clean-looking result can still omit a continuation row or turn a refund into an expense. Source review is the step that makes the table usable.

Documents to prepare

Bank statements

Extract posting date, narration, reference, debit, credit, and running balance. Keep opening and closing balances as document fields rather than treating them as transactions.

Card statements

Separate purchases, refunds, interest, fees, and payments. Preserve card and statement identifiers without exposing full account numbers in a shared workbook.

Supporting receipts

Capture merchant, receipt date, tax, tip, currency, and total. Use the receipt as supporting evidence for a statement row; two records of the same payment are not two expenses.

An illustrative month-end transaction register

Fictional values. The reference is a reviewer-assigned source locator. This table is a working register, not a reconciled ledger.

Dynamite Docs
Inside the output
Bank statements: A bookkeeper’s reviewSource reference: BANK-SEP p2 r4. Date: 2026-09-04. Description: Office supplies. Illustrative example, not extracted customer data.A bookkeeper’s reviewfx Values from your documentASource referenceBDateCDescription12345BANK-SEP p2 r42026-09-04Office suppliesCARD-SEP p1 r82026-09-09Supplier refundDocument fields / RowsEditable cells,…
Source reference
BANK-SEP p2 r4
Date
2026-09-04
Description
Office supplies
Check the rows before closing the month. Illustrative example.
Source referenceDateDescriptionDebitCreditReview note
BANK-SEP p2 r42026-09-04Office supplies125.00Receipt attached
CARD-SEP p1 r82026-09-09Supplier refund24.00Check credit sign

Review against the source

Dynamite Docs
The review
Bank statements: A bookkeeper’s reviewCompare the first and last rows on every page to catch repeated headers and missing transactions.. Recalculate the movement from opening to closing balance using the statement’s sign convention.. Check posting dates against transaction dates and keep currencies separate.. Illustrative example, not extracted customer data.SOURCE DOCUMENTBank statementsSource referenceBANK-SEP p2 r4Date2026-09-04DescriptionOffice suppliesReview before exportSource referenceBANK-SEP p2 r4Compare with the source on the leftAccept after reviewCheck the rows before closing themonth.Original stays…
Source reference
BANK-SEP p2 r4
Date
2026-09-04
Description
Office supplies
Check the rows before closing the month. Illustrative example.
  • Compare the first and last rows on every page to catch repeated headers and missing transactions.
  • Recalculate the movement from opening to closing balance using the statement’s sign convention.
  • Check posting dates against transaction dates and keep currencies separate.
  • Record unmatched receipts and unclear narrations as exceptions before assigning ledger codes.

Prepare a reviewed handoff

Dynamite Docs
The handoff
Bank statements: A bookkeeper’s reviewExtract one account at a time. Correct rows beside the statement. Prepare the month-end handoff. Illustrative example, not extracted customer data.FROM SOURCE FILE TO A RESULT YOU CAN USEBankstatementsReview workspaceAbookkeeper…review1Extract oneaccount at a time2Correct rowsbeside thestatement3Prepare themonth-end handoff
  1. 01Extract one account at a time
  2. 02Correct rows beside the statement
  3. 03Prepare the month-end handoff
Check the rows before closing the month. Illustrative example.
  1. Extract one account at a time

    Upload a representative statement and review the proposed columns. Preserve the account period and identifiers so another reviewer can identify the source.

  2. Correct rows beside the statement

    Inspect page breaks, wrapped descriptions, decimals, and refunds. A saved layout correction can guide the next statement, but its amounts still need a fresh extraction and review.

  3. Prepare the month-end handoff

    Export accepted rows to Excel or CSV. Add your ledger mapping and exception notes in the receiving workbook, then verify the imported row count and totals.

Use the accepted output

Excel supports separate sheets for accounts and exception work. CSV provides a flat import table. Review the destination’s date format, decimal separator, and debit/credit convention before using the rows in your bookkeeping system.

Review Excel, CSV, and JSON export formats

Processing choices and allowances

Choose hosted AI, your own provider keys, or local Ollama on an eligible plan. Each successfully processed hosted page uses one PE. Free and Hobby BYOK use the same PE allowance; Pro and Ultra BYOK cost zero PE, with provider charges and limits still applying. Failed pages cost zero PE.

Check current plans, page limits, and capabilities and review storage and data-residency policies before processing sensitive records. Local model inference is a separate choice from browser-local file storage.

Workflow boundaries

  • Extraction does not reconcile a bank account, select ledger codes, or post journals.
  • A matched amount alone does not establish that a receipt and a bank transaction belong together.

Questions about this workflow

Can I combine statements from different banks?

Review each statement separately first. Standardize the accepted columns in your receiving workbook and preserve the account, currency, and source reference when combining rows.

Does this replace reconciliation?

No. It prepares reviewed transaction data. You still compare it with the ledger, investigate differences, and approve any accounting entries.

Loading Dynamite Docs… This page is taking longer than expected. Reload page.