In today’s fast-paced world of business, efficiency is key to success. That’s where Optical Character Recognition (OCR) and accounts payable automation come into play. By leveraging OCR technology in NetSuite, businesses can convert scanned or digital documents, such as invoices, into machine-readable formats, significantly streamlining the accounts payable process. In our recent blog post, we explored the difference between OCR technology and AI algorithms, both integral to automating the invoice scanning process. To learn more on this topic, read the full blog post here.

Since OCR plays a crucial role in extracting information from invoices and speeding up your AP processing, we often get asked, “How accurate is your OCR?”. When you ask this same question to other solution providers, you might encounter one of two responses. Some might offer a vague answer, claiming industry leading accuracy. Alternatively, you might receive a response with a misleading percentage of accuracy with no detail. Neither response is particularly helpful, as both seem designed to downplay the significance of this topic.

In reality, the answer of OCR accuracy is influenced heavily by multiple factors, and no two companies will experience the exact same result. In this blog post, we are going to take a closer look at OCR accuracy, exploring how it’s measured, and most importantly, how SquareWorks’ OCR stands apart from other solutions.

How is OCR Accuracy Measured?

OCR systems have come a long way in improving accuracy, with some systems claiming a 99% accuracy rate. However, it’s important to note that many solutions claiming this level of precision are primarily measuring their ability to extract characters on a page. This means that only one character in 100 may be misread or skipped, resulting in a mere 1% error rate. While this high level of accuracy seems promising, it’s crucial to examine what this means in the context of AP automation. The evaluation of OCR accuracy involves multiple layers, and understanding these layers is crucial to delivering a comprehensive accuracy score.

Unpacking the Layers of OCR Accuracy

1. Character Extraction

Recognizing individual characters on a page is the first step. To a person, it’s very clear what an “A” looks like, even when the style, size, or quality changes. To a computer, this task is much more challenging. Imagine trying to describe the logic of what an “A” looks like. There are so many variations it’s not feasible. Instead, the computer needs to be trained using machine learning by showing it countless examples of what an “A” might look like. With enough examples, a computer can learn to recognize characters on a page with high accuracy.

2. Word Formation

Next, the focus shifts to transforming characters into words, sentences, and paragraphs. The system actively constructs words, sentences, and numerical expressions by analyzing the arrangement and relationships between characters. This step is vital for comprehending the context of the information on the document, ensuring accurate extraction and meaningful interpretation.

3. Contextual Understanding

Even if a computer can extract all words from a page with 100% accuracy, this information is useless without context. How do you know which words make up the vendor’s name? Is that an invoice number or a PO number? Different approaches are taken to bring context to the words and paragraphs found.

Some AP automation providers will attempt to define the logic manually. For example, if you see numbers to the right of the words “Invoice Number,” that’s likely the invoice number. This appears to work at first but falls short as more variations need to be accounted for. You will quickly run into situations where existing vendor invoices suddenly stop extracting information as expected.

At SquareWorks, we have avoided this pitfall entirely and lean heavily on machine learning to drive contextual understanding. By training on countless numbers of example invoices, we’re able to identify context without the overhead and risk other providers bring to the table.

4. NetSuite Alignment

Finally, the information extracted needs to be entered into NetSuite. However, even if a system can extract information accurately from a page, that information might not line up with your data in NetSuite. For example, what if the vendor’s name is different from what you have stored in NetSuite, or what if you’re missing a PO number in NetSuite?

Misalignment between data on the invoice and your data in NetSuite is common. To combat this problem, SquareWorks is able to learn and adjust based on your usage. We also provide you with a comprehensive rules engine to override data mapping logic if needed.

As we advance through these levels, it’s important to note that OCR accuracy often decreases due to various influencing factors. Upon evaluating the layers of OCR accuracy, it becomes evident that the focus extends beyond character extraction. Instead, it involves integrating all these layers to determine the overall accuracy rate.

What is SquareWorks’ OCR Accuracy?

While some OCR solutions focus solely on character extraction, SquareWorks goes beyond this by focusing on the various layers of OCR. Further, we analyze the frequency with which you need to modify values and manipulate what’s already populated on the screen. Our OCR accuracy falls within a range of 90-99%, factoring in variables such as the utilization of purchase orders, scan quality, and the effectiveness of information extracted in your NetSuite instance. When considering alternative solutions in the market, we encourage you to dig deeper into the numbers provided for a more comprehensive understanding.

Opting for a purpose-built, NetSuite-native solution, such as SquareWorks Automate, can make a significant difference. Our OCR accuracy goes beyond character extraction, providing a holistic approach of how well our solution processes data. Over time, our technology adapts as your organization processes more invoices, gradually reducing the need for manual data entry and review. To learn more about SquareWorks OCR and customer accuracy rates, please reach out to us here.