Convert a Scanned PDF to Excel: A Practical Guide

Most guides to converting a bank statement assume you're starting from a clean PDF downloaded straight from your bank. That's the easy case. If you're trying to convert a scanned PDF to Excel — a photocopy from a filing cabinet, or a photo taken on a phone — you're solving a harder problem, and most of the advice written for digital statements simply doesn't apply. This guide covers what actually changes when the source is an image, what makes a scan good enough to convert reliably, and the steps that get you a usable spreadsheet even from a rough starting photo.

Why a scan is a fundamentally different problem

A digital bank statement PDF has a text layer: every character, its position and its font are stored as data. Extracting it is a matter of reading that data and rebuilding the table. A scanned or photographed statement has none of that — it's a picture. Software has to look at pixels and guess which characters they represent before any table structure can even be attempted. That extra guessing step is where nearly all the risk lives, and it's why how OCR handles a bank statement is really a separate topic from PDF extraction in general, covered in more depth in our PDF to Excel converter guide.

A digital PDF's accuracy is close to guaranteed, because there's nothing to interpret. A scan's accuracy depends entirely on how the image was captured — which means the outcome is partly in your hands before you ever open a converter.

What makes a scan good enough to convert

Resolution: give OCR enough pixels to work with

OCR needs enough detail per character to tell similar shapes apart — a 3 from an 8, a comma from a full stop. A flatbed scan at 300 DPI or higher gives it that. A phone photo doesn't have a DPI setting, but the equivalent is pixel count: aim for at least 2,000 pixels along the longest side, achieved by holding the phone close and using the camera's full resolution rather than a cropped or compressed preview.

Lighting: even and diffuse beats bright and direct

A single overhead light casts a shadow across whatever your hand or the phone itself blocks, and that shadow reads to OCR as ink where there isn't any, or blank space where there's text. Natural daylight from a window, or two lamps angled in from either side, spreads light evenly across the page and avoids both shadow and glare off the paper.

Skew: shoot straight, not at an angle

Photographing a page at even a slight angle tilts every line of text, and OCR engines are far more accurate on horizontal text than tilted text. Recent testing on document images found that a five-degree tilt alone can raise the OCR word error rate by 15% or more without correction, and the loss compounds further past ten degrees — a small, easily fixed mistake with an outsized effect (imagetotable.ai, 2026). Hold the phone parallel to the page, use a grid overlay if your camera app has one, and check the statement's edges line up straight in the frame before you shoot.

What OCR gets wrong on a poor scan

Independent research on document image quality confirms what anyone who has tried this by hand already suspects: blur, skew and uneven lighting are the dominant causes of OCR error on mobile-captured documents, more so than the OCR engine itself (Frontiers in Signal Processing, 2026). On a bank statement specifically, that shows up as a few repeatable failure patterns.

• Digit swaps. A 3 misread as an 8, a 5 as an S, or a 1 as a 7 — visually similar shapes are exactly where OCR guesses wrong most often, and a digit swap changes an amount without looking wrong on the page.
• Lost decimal points and minus signs. A faint dot or a thin dash is easy for compression and low resolution to erase entirely, turning £12.34 into £1234 or a debit into a credit.
• Merged or split rows. Creases, staple holes and photocopy artefacts can make one row's text bleed into the next, or split a single transaction description across two lines that OCR then treats as two separate rows.
• Garbled text from glare or shadow. A patch of glare from an overhead light or a fold-line shadow can wipe out several characters at once, producing a description column full of nonsense strings.

"With a scan, the question isn't whether OCR made a mistake somewhere — on a poor photo, it almost always will. The question is whether your process catches it before the numbers go anywhere important." — Priya Shah, Head of Product at Ledgerary

A real-world case: the shoebox of paper statements

Imagine someone sorting a late relative's paperwork for probate and finding a shoebox of paper bank statements going back a decade — no digital copies exist, and the bank's own archive only covers the last two years. Each page gets photographed on the kitchen table, one at a time, under a single ceiling light, some at a slight angle because the table wasn't quite big enough. Fed straight through a generic converter, several pages come back with garbled totals and a handful of rows merged together where a fold line ran through the transaction table. Reshooting the worst pages flat, in daylight from a nearby window, and running them through a converter built to reconcile against the statement's printed balance turns a stack of unreliable photos into a spreadsheet that can actually be handed to a solicitor with confidence.

How to convert a scanned bank statement to Excel, step by step

1. Flatten the page and light it evenly. Press the statement flat and light it from two angled sides rather than straight overhead, so no single shadow falls across the text.
2. Shoot straight-on and fill the frame. Hold the phone directly above the page, parallel to it, close enough that the statement fills most of the frame.
3. Check the resolution before you upload. Confirm the image is at least 2,000 pixels on its longest side, and retake it if it looks soft when zoomed in.
4. Let OCR run, then scan the description column. Skim the extracted text for garbled words — a reliable sign a particular row needs a closer look.
5. Check the balance reconciles. Confirm opening balance plus money in, minus money out, equals the closing balance printed on the statement.
6. Export to Excel. Download the verified table as Excel, CSV, JSON, or a ready import for QuickBooks, Xero or Sage.

Converting scanned bank statements safely

Because a scan or photo can contain far more sensitive material spread across a phone's camera roll than a single downloaded PDF, it's worth checking how a converter handles your files before you upload old financial paperwork to it. Our guide on whether it's safe to convert bank statements covers what to look for — Ledgerary, for instance, deletes uploaded files after processing rather than keeping them on a server.

Frequently asked questions

Can I convert a scanned PDF to Excel for free?

Yes, for occasional use. Ledgerary and most competitors offer a free tier that runs OCR on a scanned or photographed statement before extracting the table, so you don't need paid software for a one-off conversion. For regular use, check the free tier includes a balance check against the statement's own totals, not just raw text extraction.

How do I convert a photographed bank statement to Excel?

Photograph each page flat, straight-on and in even light, then upload the images or a combined PDF to a converter that runs OCR before extraction. A tool built for statements will detect the table structure, pull dates, descriptions and amounts into columns, and check the total against the statement's printed opening and closing balance before you export.

Why does OCR misread numbers on a scanned bank statement?

OCR guesses characters from pixels, and low resolution, skew, glare or a faint photocopy all increase the odds of a wrong guess. Numbers are especially risky because a single misread digit changes the value without looking obviously wrong, which is why a reconciliation check matters more on scans than on digital PDFs.

What resolution do I need to scan a bank statement for accurate conversion?

Aim for at least 300 DPI on a flatbed scan, or an image at least 2,000 pixels along its longest side from a phone camera. Below roughly 150 DPI, character-level OCR accuracy drops below 90% on standard printed text, and below 100 DPI a statement's smaller print becomes unreliable to extract at all.