Most PDF converters lose fidelity where it matters most.
We ran demanding documents (SEC filings, financial tables, dense decks) through Adobe PDF Services, ConvertAPI, Docling and BitMorph, then scored every output against the source. The results show a clear, measurable gap on the details that matter.
Adobe and ConvertAPI, in our test
Every one preserved.
The leading zeros are gone. The identifier no longer resolves.
The leading zero that gets lost
A CUSIP is nine characters. Values like 002824100 start with zeros. Hand that cell to Excel as a number and it becomes 2,824,100, and the leading zeros are gone on export. The same pattern affects ZIP codes, routing and account numbers, and phone numbers.
Measured across three formats
Integrity is the headline, but the same in-house engines win on structure and editability too. Every figure below is scored programmatically against the source document.
Financial identifier integrity
Leading-zero CUSIPs kept as text, not parsed to numbers.
Editable document structure
A real navigable outline, where commercial output is structurally flat.
Native editable objects
Editable auto-shapes commercial engines never emit.
Accurate and fast, not one or the other
The only other engine that preserves your identifiers is a heavyweight ML model that takes minutes per document. BitMorph matches its integrity in seconds.
Both ran on the same CPU machine (no GPU). The open-source ML engine is 20 to 100x slower across the corpus, unusable for an interactive tool or an API. It is the only other engine we tested that keeps the identifiers intact, so speed is the trade-off it asks for.
How we measured
No cherry-picking, no synthetic files. We ran each engine on public documents and scored the output against the source with the same script for every engine.
The corpus
Real, public documents that stress converters: SEC 13F filings, borderless government rate schedules, multi-column academic papers, and dense presentation decks.
The comparators
BitMorph vs Adobe PDF Services, ConvertAPI, and Docling. In our tests, Adobe PDF Services and ConvertAPI produced nearly identical output on every document, so we report them as one commercial baseline rather than two.
The scoring
Programmatic and identical per engine: leading-zero CUSIPs matched against the source, heading and link structure counted in the output, editable objects enumerated in the deck.
Reproducibility
Every number here is regenerated by a script in the repository and re-checked after any engine change, so the benchmark cannot silently drift.
Questions this benchmark answers
Which PDF-to-Excel converter keeps financial data intact?
On a corpus of SEC 13F filings, BitMorph preserved 100% of leading-zero CUSIP identifiers (39/39 and 100/100 on two filings). In our tests, Adobe PDF Services and ConvertAPI each altered 46 to 54% by stripping the leading zeros (002824100 becomes 2824100). Docling, an open-source ML engine, preserves them but runs 20 to 100 times slower.
Why do PDF converters change CUSIP numbers and ZIP codes?
Many converters hand each cell to Excel as a number. A value like 002824100 is stored as the integer 2,824,100, dropping the leading zeros. This affects any leading-zero identifier: CUSIPs, ZIP codes, bank routing and account numbers, and phone numbers. BitMorph detects identifier-shaped values and keeps them as text.
Which tools did you compare?
BitMorph against Adobe PDF Services, ConvertAPI, and Docling (an open-source ML engine). In our tests, Adobe PDF Services and ConvertAPI produced nearly identical output on every document, the same values, the same structure, the same results. Because we could not tell them apart in practice, we report them together as a single commercial baseline rather than two independent ones.
Is this benchmark reproducible?
Yes. Every figure comes from running each engine on public documents (SEC 13F filings, government forms, academic papers and real presentation decks) and scoring the output programmatically against the source. The scoring measures identifier integrity, editable structure and speed.
Built for your most demanding documents
Bring your hardest PDF: the 13F, the scanned schedule, the deck you lost the source for. See the fidelity for yourself.
Tests run in August 2026 on the public documents described above, using each tool’s standard PDF conversion. Results depend on the specific document and the version of each tool at the time of testing, and your own results may differ. Figures reflect what we observed, reported in good faith for comparison. Adobe and Adobe PDF Services, ConvertAPI, and Docling are trademarks of their respective owners; BitMorph is independent and not affiliated with, endorsed by, or sponsored by them.