We Tested 100 UK Bank Statements

That last point is the wedge. The running balance is an answer key the bank already wrote into the document. You don't need to hand-key a "correct" version first. The statement checks itself.

Why "we tested 100" needs a method before it needs a number

Most "we tested X statements" pieces lead with a percentage and bury the design. That's backwards. A conversion accuracy claim means nothing until four things are fixed: which statements were in the set, what counts as a pass, when the test ran, and which tool versions were involved. Change any one and the number moves.

So we'll describe the test the way we'd run it, and the findings we'd expect to hold qualitatively, rather than invent a tidy "97.4% across 100 files" line. If a figure isn't measured against a stated set and rule, it's marketing wearing a lab coat.

UK bank statements are a good stress test precisely because they don't agree on anything. A set of 100 drawn across the major banks and the app-only challengers covers more layout variety than most accountants see in a year.

What 100 UK statements would actually have to cover

A representative sample isn't 100 statements from one account. It's a spread across banks, statement types and quality levels, because each axis breaks converters differently.

• Bank spread. Barclays, HSBC, Lloyds, NatWest, Santander, Halifax, Nationwide, plus the app-first banks Monzo and Starling. High-street layouts and challenger exports follow different conventions entirely.
• Personal and business. Business statements carry more columns, faster-changing balances, and a denser run of card payments and direct debits.
• Text vs scanned. Some PDFs hold real, selectable text. Others are photos or scans of paper, where nothing is machine-readable until optical character recognition (OCR) reads the pixels.
• Length. Single-page summaries through to forty-page business runs where the running balance carries across many page breaks.
• Period. Quiet months with a handful of lines, and busy months with hundreds.

Skew the sample and you can manufacture almost any result. Load it with clean single-page Monzo exports and everything looks effortless. Load it with scanned Lloyds business statements and the same tool struggles. A fair 100 is built deliberately, then disclosed.

How UK bank formats actually differ

Here's what trips people up: there is no standard UK bank statement. The arithmetic backbone is shared — opening balance, the movements, closing balance — but almost every presentation choice varies.

Monzo and Starling tend to be the kindest to a converter because they were born digital and export tidy structures. The traditional banks were laid out to be printed and posted, so their PDFs encode position rather than meaning — and that's where most extraction errors begin. We go deeper on that gap in our extraction accuracy benchmark.

Text PDFs vs scanned PDFs

This is the single biggest split in any UK sample, and it deserves its own rule in the test.

A text-based PDF stores the characters directly. Conversion is essentially careful layout reading: find the columns, keep the rows in order, don't merge a wrapped description into a new line. A scanned or photographed statement stores only an image. Before anything can be extracted, OCR has to turn pixels into characters — and that's where a smudged 3 becomes an 8, or a faint decimal point vanishes and £1,240.50 reads as £124,050.

A serious test of 100 separates these two populations and reports them apart. Blending a clean text PDF with a phone photo of a creased statement and averaging the result hides exactly the failure that matters to the reader holding the worse file.

The pass/fail rule: does it reconcile?

Now the part that makes a test objective. For each of the 100 statements, the result is binary — it reconciles, or it doesn't.

The rule is the running-balance identity. Opening balance, plus every credit, minus every debit, should equal the closing balance the bank printed. Where the statement also prints a balance after each line, you can go further and reconcile transaction by transaction, pinpointing the exact row where the maths first diverges.

Why is this stricter and more honest than a field-level spot check? Because the most dangerous extraction error leaves no wrong cell to find. If a converter silently drops a transaction, every remaining cell can be perfect. A sampling audit sees nothing wrong; the spreadsheet just has fewer rows than the PDF. Only the running balance catches it — a dropped £640 supplier payment doesn't flag itself, but the closing balance won't tie, and that mismatch is the only signal you get.

Run that rule across the set and the headline metric falls out on its own: a reconciliation pass rate, the share of statements that balanced end to end. That number is meaningful precisely because the answer key came from the bank, not from a human re-typing the file. The method behind it is set out in our reconciliation accuracy benchmark.

This is the check Export Bank Statement runs on every conversion. It verifies each extracted transaction against the running balance and flags any statement that doesn't reconcile, so you catch a missing or misread line before you trust the numbers — not three months later in your accounts.

The failure points a 100-statement test would expose

Across a real UK sample, the same handful of failure points recur, and they cluster on the traditional banks and the scanned files.

1. Dropped transactions — a line lost at a page break or in a dense run. Invisible to spot checks, fatal to the balance.
2. Merged descriptions — a wrapped reference read as a separate transaction, inflating the row count.
3. Sign errors — a money-out amount parsed as money-in, common where a bank uses one signed column instead of two.
4. Decimal and OCR slips — on scanned files, a misread digit or a lost decimal point that throws a single amount wildly off.
5. Phantom rows from subtotals — carried-forward and page-total lines mistaken for real movements.

Each of these breaks the running-balance identity, which is exactly why the reconciliation rule is the right adjudicator. A statement can look clean and still fail; the maths is what tells the truth. If you want the bank-by-bank view of which layouts cause the most trouble, that's its own topic in Pillar 9 — Research & Benchmark.

What an honest result looks like

A credible "we tested 100 UK bank statements" report would publish, at minimum: the bank breakdown of the sample, the text-vs-scanned split, the date of the run, the converter and version, the exact pass rule (reconciliation to the printed closing balance), and the pass rate reported separately for text and scanned files. Anything less is a vibe.

We'd rather give you the method than a number we can't stand behind. The method is portable — you can run it on any converter, including ours, with your own statements as the sample. The most useful test of 100 is the one you run on the files you actually need to convert.

CTA: Convert your own UK bank statements and watch the totals reconcile, line by line, at Export Bank Statement.