Multiline invoice descriptions: avoid creating extra items

A supplier can print an item name, size, packaging and delivery reference on four lines. A generic table export may treat them as four rows. The reverse mistake is just as costly: joining the next item to the current description.

The target is not one spreadsheet row per visual text line. It is one row per invoiced item, with enough context to explain how the text was grouped. The examples here are synthetic; they describe a review method, not a tested claim about an extraction engine.

Recognise a continuation without guessing

Consider this printed block:

The two middle lines look like continuation text. The next populated quantity and amount suggest a new item. That is evidence to review, not an infallible rule: a service can legitimately have no quantity, and some layouts print amounts on a separate line.

Keep the page image or source PDF near the table. If there is no clear anchor, mark the grouping as uncertain instead of merging silently.

Use more than one anchor

A product reference, a quantity, a right-aligned amount and a separator can jointly suggest where an item starts. Do not rely on indentation alone. Product descriptions can be indented and service lines may not follow the same pattern.

Preserve useful detail in the description. “Archive boxes” and “Archive boxes, natural board, pack format” may mean different purchases. But a repeated invoice header, page number or subtotal is layout information, not part of the product name.

When a row crosses a page boundary, review both pages together. A header inserted between the two halves should not split the item or join itself to the description. Keep references to both pages if the eventual export supports that situation; otherwise explain the limitation.

Native PDF import can still be a good starting point

Microsoft documents multiline-row cleanup in its PDF connector, including transformations such as filling values down and grouping adjacent rows. Those tools can be useful when the structure is sufficiently regular. The correct transformation depends on the document; this article does not claim that a universal sequence works in every Excel version.

Tabula is another option for text-based PDFs. Its stated scope excludes scans. If text selection itself is unavailable, do not confuse a blank table import with an invoice containing no items.

Check the result with a small review list

Count the intended item blocks on the source, then compare the exported row count. Check the boundary between two similarly named items, the start of each page and all service lines with blank quantities. Finally inspect whether a description contains an invoice header or a total.

Amounts alone are not enough. Two different items can have the same price, and a wrong merge may leave the sum unchanged. A good review keeps descriptions and source context visible alongside numeric checks.

The InvoiceLineLab example shows prepared line notes without receiving files. For the underlying column design, read invoice headers versus line items.