The .docx file.
The one document format that stores its tables as tables. That makes extraction far more reliable than from a PDF of the same report.
What it actually is
A .docx is a ZIP of XML, and its tables are stored as genuine table structures with rows, cells and merges. This is the key difference from a PDF: nothing has to be inferred from the geometry, because the grid is written down.
The prose around a table is worth keeping too. A sentence reading "figures exclude the Northern region pending migration" is an exclusion that belongs in the methodology, and it exists nowhere in the numbers.
The awkward case is a document whose numbers are in the sentences rather than in a table, which is common in board reports. Those are read as context rather than promoted to data, because a figure quoted in prose has no denominator attached.
At a glance
- Really a
- ZIP archive of XML parts
- Tables
- Stored as real structures, including merged cells
- Prose
- Kept as context, not discarded
- Embedded objects
- Linked spreadsheets are detected and named
- Read here as
- Extracted. Tables lifted out, formatting discarded.
- Status
- Ready
- Fidelity
- Data extracted
- Largest file
- 250 MB
Tables are lifted out of the document; the prose around them is kept as context.
Upload one
The analysis runs before there is anything to pay for. You see what it found, and the evidence behind it, first.
No account needed to start. You only pay when you like what you see.
What goes wrong, and what is done about it
01
Merged cells that were doing two jobs
A merged cell spanning three columns is often a group heading rather than a value, and flattening it repeats the heading into three data cells that then look like observations.
Merged spans are read as spans. A merged cell above a group of columns is treated as a header level rather than as a row of data.
02
Numbers that only exist in a sentence
Board and management reports frequently state the important figure in prose and put only supporting detail in the tables. Extracting tables alone gets everything except the thing the document was written to say.
Figures stated in the surrounding text are captured as context and reconciled against the tables, and any disagreement between the two is reported as a finding.
If you have the choice, send something else
If the tables in the document were pasted from a spreadsheet, send the spreadsheet as well. The document has the numbers; the workbook has the formulas that produced them.
Questions people ask
Is this better than uploading the PDF version?
Yes, meaningfully. The Word file stores table structure explicitly, while the PDF stores glyph positions from which the structure has to be inferred. Send the .docx where you have the choice.
What happens to images and charts in the document?
A native chart's underlying data is read. A chart pasted in as a picture is read as a picture, in the interpreted tier, and says so.
Can it handle a document with twenty tables?
Yes. Each one becomes its own source, and how they relate to each other is worked out from their contents rather than assumed from the order they appear in the document. Tables that share a key are matched; ones that do not are kept separate.