Any company that processes documents in bulk, knows the tedious task of processing is time-consuming and costly, but necessary. We are talking about things such as sorting documents for (digital) mailrooms, checking document quality , cropping documents, splitting multi-page documents, splitting multiple documents on a single picture, stitching multi-picture, correcting document orientation and extracting and validating data. In this blog we tell you more about how Klippa approaches automated document processing.
Let’s start by defining a use case to make it easier to understand how automated document processing works and what the business case is. We will look at a financial document processing case, specifically focussed on automated VAT returns. We use financial documents as an example because almost every company or person gets into contact with financial documents in some form. For this specific use case we will be talking about how to automatically convert and process receipts and invoices, but this could really be any type of document.
Now we ask you to use your imagination for a bit for a short thought experiment. Let’s say you are running the operations department of a major VAT returns company (since you are reading this blog, maybe you actually are ?). In this role you are responsible for making sure your clients get as much VAT returns as possible, since your company takes a percentage of the returns. This means it is your top priority to process receipts and invoices fast, at a low cost and with a high success rate. Your department checks 2 million documents a month from thousands of clients for tax return eligibility. Part is done manually and part automatically. You have implemented some kind of OCR API to extract data from receipts and invoices to help you automate part of the work. Automatically processing one receipt or invoice costs about €0.05, but manually processing or correcting one will easily cost 10 to 50 times more, depending on the complexity. So let’s say manual processing costs €0.50 per document. Below is an example of the type of documents customers typically send in. In this example they are all receipts, but these could be any type of documents and files:
About half of all documents are processed automatically, so that’s great, but the other half is giving you a headache because of the manual labor you put into it, not to mention peak volumes. Part of the problem is caused by the complexity of some documents, which makes the process hard to automate, part is caused by the (sometimes poor) quality of the data that customers provide. Customers send pictures from receipts and invoices, send PDF files, sometimes multiple receipts on one page, sometimes one that covers multiple pages. Some are blurry, some are rotated and some have hardly any contrast. Currently administrative employees are using tools to cut, rotate and optimize these documents so they can be processed afterwards. Needless to say, this takes a lot of time – and money.
The business case
What if more document correction and processing tasks could be automated? Instead of 50%, you could then process 75% or more of the documents automatically. Since your company is processing 2 million documents a month, 500.000 additional documents could be processed at only 10% of the cost. This saves you around €50.000 per month, or €600.000 per year. Sounds like a dream, right? Well, we are happy to tell you we can make it a reality.
How it works
Using our Klippa smart document processing tools, you can automate many types of document processing and correcting. This can be done via APIs, SDKs or our interfaces. Curious to know how it works? Klippa uses machine learning models and OCR to process documents. This ensures continuous improvement and high accuracy. Below is a 7 step summary that explains the process, which takes sheer seconds:
1. In the first step we enhance the image quality by automatically cropping documents, optimizing brightness and contrast and correcting perspective. That can be done in our API, but also via our mobile SDK. After cropping and perspective correction the same images look something like this:
2. All incoming documents are then checked for quality. In this step Klippa decides if documents are blurry, too bright, too dark or if there are any other problems that prevent automated processing or prevent processing completely.
3. Now we move forward with OCR. In this step we turn each image into searchable text. This creates a text file next to the image that we can later use for data extraction.
4. Now that we have the text on a document, we can decide whether or not a document is in the right orientation. Receipts and invoices should normally be vertical, more height than width, for instance. We can now automatically rotate documents to the correct orientation. That gets us something like this:
5. Because we have a text version of the document, we use our AI models to classify the documents using different variables. Take for example document type: we can classify up to 30 different document types such as receipts, invoices, contracts, tax statements and so on, but we can also categorize country of origin and language.
6. Now we can start to automatically extract any relevant data. In case of VAT returns these are data points such as the country of origin, date, total amount, VAT values and possibly things such as receipt line items. If requested we can even classify receipt and invoice line items into categories such as ‘food & drinks’, ‘electronics’, ‘alcoholic’, ‘household materials’ and 20 other categories. The response is generally a JSON file, but this can be customized to for example CSV or XML. In the image below you can see an example of the steps from picture, to text to JSON. Keep in mind that this is a simplified example of the JSON to keep the image relatively short, but we can extract many more data points:
7. In the last step there is the option to validate extracted data with third-party sources. For example the chamber of commerce database, VAT databases, Google Maps API and many more. These can even be internal databases with PO numbers or other values. This guarantees optimal data quality, prevents fraud and improves confidence.
In this blog we gave a brief overview of the types of challenges we solve. Do you have any document processing challenges in your organisation? Let’s talk and see how our tools and team can help you reach your goals. Whether you need improving output quality, saving processing time or saving money, Klippa is here to help.