How much of your week goes on typing in things that are already written down?
Supplier invoices. Receipts from the glovebox. A customer form filled in by hand. Someone reads the paper and types the same details into a spreadsheet or accounting system. It is slow, it is boring, and every retyped number is a chance for a mistake.
There is a technology built for exactly this. It is called optical character recognition, or OCR, and it is one of the most useful bits of AI a small business can put to work.
Optical character recognition
OCR simply means teaching a computer to recognise letters and numbers wherever they appear.
If you have ever pointed your phone at a sign and watched the words appear on screen, you have already used it. A computer starts with a photo or a scan. To the computer that picture is just millions of tiny dots called pixels. OCR is the process that turns those dots back into words and numbers you can search, copy and use.
The clean up
Photos of paperwork are rarely perfect. Before reading anything, OCR tidies the image:
- Greyscale. It drops the colour so busy backgrounds and coloured letterheads do not confuse it.
- Noise removal. It wipes away specks, coffee stains and creases that could pass for a full stop.
- Straightening. It rotates a crooked page so the lines run level.
- Contrast. It darkens the text and lightens the page so the letters stand out.
This is why a flat, well lit photo reads far better than a crumpled receipt shot at an angle.
Lines, words and characters
Next, OCR slices the page into pieces. First into lines of text, then into words, then into single characters. Each small block can then be studied on its own.
The recognition step
Here is where the AI comes in. Older systems compared each character against a library of letter shapes, a bit like cookie cutters in every font and size.
Newer systems use machine learning. A neural network is trained on millions of labelled examples, from crisp printed letters to scrawled handwriting. Over time it learns the quirks of each letter. That is how it can read a font it has never seen before.
The common sense check
Reading letters one at a time still produces errors. So OCR checks its own work:
- Dictionary checks catch words that do not exist and suggest the likely fix.
- Language models look at the surrounding words to pick the reading that makes sense.
- Pattern rules know what a date or phone number should look like. If a number contains the letter O, something is wrong.
These checks are what lift accuracy from pretty good to almost perfect.
Paperwork in a small business
Once the text is out of the image, the time savings start. Banks have used OCR for years to turn mountains of forms into spreadsheets without someone typing all day. The same idea now works at small business scale:
- Invoices. Pull out the supplier, date, amount and invoice number without retyping them.
- Receipts. Snap a photo and have the spend recorded, rather than keeping a shoebox for tax time.
- Forms. Turn handwritten intake or job sheets into records you can search and sort.
- Old files. Scan archived documents so you can find them by keyword instead of digging through folders.
Many of the tools you may already use, such as accounting software and phone scanning apps, have some of this built in. It is worth checking before you buy anything new.
Where it struggles
OCR is not perfect, and it helps to know where it trips up:
- Handwriting. Messy cursive is still the hardest case, though context helps a lot.
- Unusual fonts. Decorative or very old styles can throw it.
- Poor photos. Grainy, dark or blurry images lose detail that no software can recover.
- Curved surfaces. Labels on bottles or crumpled pages distort the letters.
So keep a person in the loop, especially for totals and anything that ends up in your books. A quick check of what the software read is still far faster than typing it all yourself.
Where to start
Pick the one type of paperwork you retype most often. Run a week of it through an OCR tool and compare the results against what you would have entered. If it gets most of it right, you have found a few hours a month you can take back.
Try this. This week, photograph five receipts or invoices flat and in good light, then run them through the scanning feature in your phone or accounting app. Count how many fields came out right without any typing.