PDF was never designed to prove anything

Open your inbox. Search for "invoice" or "receipt". You'll find dozens of PDFs — supplier invoices, hotel receipts, train tickets, restaurant bills. And like everyone, you trust them.

Yet the format underpinning much of your company's financial life was never designed for that. It's written in its own history.

1991: a printing problem, not a trust problem

In the early 90s, Adobe co-founder John Warnock launched an internal project called "Camelot". The goal was simple: let anyone open a document on any computer and see it exactly as its author intended. Same fonts, same layout, same colors.

PDF was born in 1993 to solve a visual fidelity problem. Not an authenticity problem. Not a proof problem. Nowhere in the original spec is there any requirement to guarantee that the document's content has not been modified.

PDF is a presentation format. Full stop.

The misunderstanding that conquered the world

PDF became so reliable visually that we ended up believing it was reliable, period. A PDF "looks serious." It doesn't shift like a Word file. It looks printed, frozen, final. Our brains took the shortcut: what looks frozen must be authentic.

That shortcut is wrong. And the irony is total: the company that created the format also sells the tool to modify it. Acrobat Pro lets you change an amount, a date, or a merchant name in a few clicks — not to mention dozens of free online editors that do the same thing in a browser.

Modifying a PDF requires no technical skill. Detecting that it has been modified, however, does.

What this changes for your expense reports

Take a hotel receipt in PDF, attached to an expense report. What guarantees that the amount shown is the original one? Nothing. Absolutely nothing.

Without an electronic signature — and which hotel signs its receipts? — a PDF carries no native integrity guarantee. The only traces of a modification are buried inside the file: editing metadata, software history, inconsistent internal structure. Clues no human reviewer can catch by eye, across hundreds of expense reports a month.

And the problem just changed scale. With generative AI, you don't even need to retouch a real receipt: you can produce a visually perfect fake one in seconds. Visual checks, already fragile, are becoming obsolete.

Regulators have ruled: PDF alone is no longer enough

European tax authorities figured this out before anyone else. The French e-invoicing reform, whose reception obligation applies to all companies from September 2026, rests on a simple principle: a PDF invoice emailed as an attachment is no longer a compliant invoice. Structured formats (Factur-X, UBL, CII) transmitted through Approved Platforms are taking its place, with machine-readable data and built-in consistency checks.

Read between the lines: the State itself has acknowledged that PDFs don't have sufficient probative value. If they did, the reform wouldn't exist. The movement is European, with the ViDA directive gradually extending these requirements to cross-border flows.

The blind spot: everything that isn't an invoice

The e-invoicing reform covers B2B invoices. It doesn't cover taxi receipts, restaurant bills, hotel notes, parking tickets. The bulk of expense report evidence remains, and will remain for years, PDFs or images with no integrity guarantee whatsoever.

That's where the economic paradox closes in. Each receipt carries a small amount: €15, €40, sometimes €200. Manually checking each one costs more than the fraud you'd find. So you sample, you check 5 or 10% of reports, and accept the rest on the strength of a format that, as we've seen, proves nothing.

Expense fraud isn't a widespread dishonesty problem. It's a cost-of-control problem. As long as verifying costs more than defrauding, fraud is structurally winning.

Flipping the equation

That's the bet we made with EVA: if human review can't be profitable on small amounts, then review must be automated, exhaustive, and designed from the ground up to stay economically relevant.

EVA analyzes 100% of expense reports before payment, where traditional audit samples after the fact. The analysis combines:

Plugged via API into your existing expense management system, on 100% European infrastructure.