The bank-statement conversion that holds up in deposition.

What this is built for

Divorce discovery, lifestyle analysis, hidden-asset search, embezzlement reconstructions, fraud examinations, marital-estate inventories, plaintiff-side damages workups, and any other engagement where the conversion step was supposed to take an hour and ate three days because the PDFs came in seventeen different layouts and a quarter of them were scanned at 200 DPI with the staple still in frame.

You are not a typical SaaS user. You are billing hourly against an engagement, the output is going into an exhibit list or an expert report, and a quietly-wrong spreadsheet is worse than no spreadsheet — it is malpractice exposure. Every feature on this page is here because that risk is the binding constraint on what you can use.

Why forensic accountants pick pdftoexcel

• The reconciliation is the deliverable, not a confidence score. Every export is checked against the statement's own printed beginning and ending balance. If the rows don't sum within a cent, the file ships with an Unverified flag pointing at the first divergent row — never a silent miscount. ChatGPT and the existing PDF-to-Excel converters have no architectural invariant that forces the totals to close. We have one. That is the entire reason for this product.
• Scanned discovery packets are first-class. Opposing counsel hands you a 200-page PDF that is really a printout of a printout. Most converters either reject it or run cheap OCR with no arithmetic check and produce a spreadsheet that is wrong in subtle places. We route scanned files through Gemini vision OCR and run the same reconciliation on the extracted rows. Verified or honest flag, same as a digital PDF.
• Multi-year, multi-account, multi-bank in one engagement. Drop up to 10 PDFs in one batch and download a single ZIP of reconciled files. The Professional plan covers 1 500 pages a month — enough headroom for a typical mid-sized engagement without juggling per-file passes.
• Provenance you can defend. Each conversion records the parse path that produced it (deterministic fast-path, Gemini Flash, Claude Haiku, Claude Sonnet) and the reconciliation result, visible on the result page and in your dashboard history. If a deposition asks how the spreadsheet was produced from PDF page 47, the answer is not vibes.
• Confidentiality you can put in writing. No LLM provider trains on your data — both Anthropic (Commercial Terms) and Google (paid-tier Gemini API) exclude API inputs from model training. PDFs are deleted on successful conversion. Source files only persist if a manual reviewer needs to reconcile by hand, and even then for at most 30 days. We sign a DPA on request — read it at /legal/dpa.
• Inline edit before export. When the OCR misreads a "0" as an "O" in the description column, you fix it in the in-browser preview and re-export. No round-trip through Excel just to repair a single cell.

What "Verified" actually means

One specific arithmetic check: the sum of all extracted transaction amounts equals the difference between the statement's printed beginning and ending balance, to within one cent. Nothing else. It is not a legal certification of the statement's authenticity. It is not a representation that descriptions or merchant names are letter-perfect. It is not an opinion on whether the statement is genuine. It is the math.

That distinction is deliberately narrow. It is the only guarantee a software vendor can ship without lying — and it happens to be the exact guarantee that catches the silent extraction errors that make a downstream tracing exercise useless. You still review the output before it goes into an exhibit. We do not pretend otherwise. The full position is on the security page and in the Terms of Service.

How a typical engagement runs through this

1. 1. Upload the bundle. Drag the discovery PDFs onto /app/new. Up to 10 files per batch, up to 50 MB each, password- protected files prompt for a password and decrypt in memory.
2. 2. Conversion runs. The pipeline tries the deterministic fast-path first; if the PDF is from one of the 24+ banks we hand-tune for, it never touches an LLM. Otherwise it goes to Gemini Flash, escalating to Claude only if reconciliation fails. Most files finish in 5–15 seconds. Multi-year combined statements take longer.
3. 3. Review the result. The result page shows the Verified or Unverified status, the parse path, the row-level preview, and the running-balance column. If anything is flagged, the row is highlighted and you fix it inline.
4. 4. Export and archive. XLSX, CSV, or QBO with one click each, or a single ZIP for the whole batch. Reconciled rows stay in your dashboard for 90 days on the Professional plan, longer on Business — long enough to span a typical case timeline, short enough that retention isn't a discovery liability of its own.

The plan most forensic practices land on

Frequently asked, by forensic accountants

Can I cite the Verified badge in an expert report?

You can describe what it represents — that the sum of extracted transactions equals the printed ending minus beginning balance to within one cent — and that is what we publicly stand behind. We don't recommend representing it as a forensic certification of the statement itself, because it isn't one. The reconciliation is a check on the extraction, not on the source document.

What happens to discovery PDFs after I upload them?

Source PDFs are deleted from storage as soon as the automatic conversion succeeds. If the file fails the automatic pipeline and is sent to manual review, it is held in a private encrypted bucket for at most 30 days while an authorized engineer reconciles by hand, then deleted. Extracted rows live for 90 days on Professional. You can delete any document on demand. Full lifecycle on /security.

Do the LLM providers see the bank statements?

Roughly 70% of conversions complete on the deterministic fast-path with no LLM call at all. The fallback path sends page images to Google Gemini (paid tier) and, if that can't reconcile, to Anthropic Claude. Both providers contractually exclude API inputs and outputs from model training under their commercial terms. The full sub-processor list is at /legal/subprocessors.

Will you sign a DPA / BAA / NDA?

The standard Data Processing Addendum is published at /legal/dpa and incorporates the EU SCCs. Email hello@bankpdftoxls.com for a counter-signed copy. We are not currently a HIPAA-covered provider; engagements that need a BAA aren't a fit yet. Mutual NDAs are available on request.

Can your output be subpoenaed?

Anything we hold can in principle be subject to legal process, like any vendor. The practical mitigation is short retention and minimal data: source PDFs are deleted on conversion, reconciled rows expire, and account-level metadata is the only persistent record we keep.

What about a 600-page combined statement going back five years?

Single-file size cap is 50 MB. Pages above 200 in one PDF tend to take longer because the LLM fallback batches pages — we recommend splitting into one PDF per statement period when feasible, both for parse speed and for row-level traceability back to a specific statement.

About the operator

pdftoexcel is built and operated by Matrizexplícita Lda, a Portuguese limited-liability company. The product, the support queue, and the manual-review queue are run by Kyr Poskonoff — a single founder, intentionally. If you have a question about a difficult statement layout, a sensitive engagement, or a feature that would unlock your workflow, the email address answers within one business day.