Accounting, Tax & Bookkeeping// topic
Extraction and classification of source documents
In short
Turning invoices, bank statements, receipts and tax forms into structured, confidence-scored data a reviewer can check against the original page.
10 pages
definitions
- A credit note is not a negative invoice: where the sign really livesModel it as an invoice with a minus in front and three things break: allocation, tax reporting, and every aging report that reads the ledger afterwards.definition8 min
- Remittance advice: the document that says which invoices a payment coveredThe payer's account of what a payment was for. It is the least standardised document in the finance stack, and the one whose references most often point at something other than your invoice number.definition7 min
- The end-of-day bank statement message, and the fields a PDF never hadAn XML statement the bank sends the account owner, reporting booked entries and balances. For a bookkeeping pipeline it turns extraction into validation.definition7 min
- The partner and shareholder income statement, read as an extraction problemA United States form whose meaning lives outside its grid. The box holds an amount, the code letter beside it decides what the amount is, and the detail a preparer actually works from is in the attached statements.definition8 min
diagnostics
- A two-hundred-page client scan splits into the wrong number of documentsBoundary errors are wildly uneven across document types, so an overall accuracy figure hides them. Count documents in against documents out, per type, and fix the cause where the pages are produced.diagnostic10 min
- Dates and amounts come out wrong, but only for one client's suppliersA locale fault cannot corrupt every document, which is exactly why it survives. Sample the ambiguous cases only, sliced by supplier, and the fault appears at once.diagnostic10 min
- Extracted statement: closing balance does not equal opening plus movementsA bank statement proves its own completeness. Foot it page by page, and the size and sign of the difference tell you which row was dropped, duplicated, inverted or invented.diagnostic11 min
- Half the client bank statements will not open in the pipeline at allA pile of failed jobs at the first step is four different problems wearing one error message. Classify by failure class before anyone retries, re-requests or asks a client for a password.diagnostic11 min
- Line items extract cleanly but they do not sum to the invoice totalThe lines are right, the total is right, and the two disagree. Divide the difference by the subtotal before opening the document — the ratio usually identifies the fault on its own.diagnostic10 min
- One supplier arrives under six names, and none of them match the ledgerA supplier's name is the least stable field on its invoice. Match on the registration number, the remittance account and the address, and let name similarity break ties nothing else can.diagnostic11 min
Other topics in Accounting, Tax & Bookkeeping
See allClient document intake, chasing and portalsHow a firm actually gets records out of clients: secure upload, request lists, automated chasing, and a status view the whole engagement team can trust.8 pagesReconciliation, coding and month-end closeRules-first transaction coding, matching logic, and an exception queue that surfaces only what a human genuinely has to decide.3 pagesLedger and practice-stack integrationsWiring a build into QuickBooks, Xero, NetSuite, Sage and practice management — sync design, controlled write-back, rate limits, and what the APIs will not let you do.1 pageBuild vs buy and vendor integration riskWhen an off-the-shelf AI accounting tool is the right answer, when it is not, and the integration and lock-in risk carried either way.1 page
Working on something in this space?
Tell us where you are in a sentence or two. We'll tell you honestly whether we're the right team, and what a sensible first slice of the work looks like.
Start the conversation