Bank statement descriptions split across rows: join them safely

Join description fragments only after identifying the transaction they belong to. A blank date alone is not a reliable boundary: some statements print a date once for several separate payments. Keep the original row order and give each confirmed transaction its own identifier.

Mark transaction starts before joining text

1. Number the extracted rows before sorting. Compare the PDF date, amount and balance columns with the rows around the broken description.

2. Mark the first row of each transaction. Attach a continuation only when its position and text match that transaction and it does not contain a separate amount or transaction marker.

3. Combine the description fragments in source order, separated by a space. Retain the original fragments and page numbers in a separate review sheet.

Fictional example

A fictional PDF shows 16/10 | TRANSFER NORTH | debit 42.00, followed by REFERENCE AB17 with no amount. These can become one description, TRANSFER NORTH REFERENCE AB17, after visual confirmation. The next line has no date but says CARD SHOP | debit 9.00: it remains a separate payment. Three physical lines therefore produce two transactions, totalling 51.00 in debits.

Decide whether a fragment can be attached

Use three outcomes in your review sheet: a confirmed new transaction, a confirmed continuation, or unresolved. A fragment’s indentation is supporting evidence, not a decision by itself. A bank reference may look like a number without being an amount; check which printed column it occupies. Conversely, a dated line containing only descriptive text can still belong to the preceding transaction. If the page layout does not settle this, preserve the separate fragment and mark its possible parent rather than silently assigning it.

In the fictional example, give the first two lines the ID T001 and the card line T002. Keep 42.00 on the amount-bearing source row and 9.00 on its own row. Do not fill 42.00 into every continuation and then sum the group: that would create a fictitious debit of 84.00. Joining text changes the description, not the payment amount.

Keep source order inside each group

Preserve an original-position field before regrouping lines, and use it to restore the source order. Sort each confirmed group by its preserved source position before joining its text, then restore transaction order. Check NORTH REFERENCE AB17, not merely the presence of all three words. A reversed reference can impair later matching even when the debit total stays at 51.00.

For a few transactions, manually copying the confirmed text into a review table may be simpler than building a query. For recurring files, save the boundary decisions with the source-row IDs; an automation can then reproduce a decision without hiding the evidence.

Keep a boundary sheet that separates text from money

Use the fictional lines above as a small test case. Amounts below are EUR; an empty amount on a continuation means “no separate amount printed”, not a zero-value payment.

Source fragment	Reviewed boundary	Output transaction	Debit retained (EUR)
TRANSFER NORTH with date and amount	New transaction	T001	42.00
REFERENCE AB17, no separate amount	Confirmed continuation	T001, append text	No additional debit
CARD SHOP, blank date, separate amount	New transaction	T002	9.00
Repeated DATE / DESCRIPTION / DEBIT heading	Page furniture	No transaction ID	No debit

The acceptance result is two transaction rows, two amounts and EUR 51.00 debited. Check all three: a correct total alone would not reveal a reference attached to the wrong payment. Before deleting a repeated heading from the working table, retain its page position so that a fragment below it can still be traced.

For a description continuing onto the next page, inspect the previous page’s final transaction and the next page’s account heading together. A new account, period or table section stops automatic attachment. If the first fragment is missing from the supplied pages, keep the orphan unresolved; do not create a parent transaction from its wording.

Preserve punctuation inside references such as AB-17. A printed hyphen can be part of the reference, so do not remove it merely because it falls at a line end. For a genuinely split word, record the source fragments and the confirmed reconstruction. Inspect a separate working copy of the problem page before combining its reviewed rows with the rest.

Checks and limits

Compare transaction count and amounts before and after joining. Grouping by date would wrongly merge both payments. In Power Query, grouping is a possible implementation after you assign transaction IDs; it cannot decide which fragments belong together. Review an orphan fragment at the top of a page against the preceding page.

Keep the original file unchanged and record any unresolved item before using the table.

Related checks

Identify transaction boundaries before joining text fragments. Diagnose a bank PDF before choosing a table extraction tool.

BankStatementScan is in preparation. The site offers an access request and a fictional example; it does not currently convert your bank files. See the planned service.

Sources consulted on 16 October 2026

• W3C — PDF reading order.
• W3C — Model for Tabular Data and Metadata on the Web.