Mastering Accuracy: Doing more than just OCR for Receipts
When a receipt scanning app doesn’t get the receipt right 99% of the time, people stop using the scanning functionality. OCR still is not nearly 99% accurate with real-world receipts, so if your app relies on that alone, it is destined for failure.
There are many great receipt scanning applications available in the market. Some specialize in consumer use cases, some in business cases, some in both, and everywhere in between. There are a lot of lists out there that help people find the best tools, and one thing is certain: Expensify is widely regarded as a top receipt-scanning app, if not the top one. Don’t just take my word for it; check out our accolades from our friends over at Forbes: 'Best Receipt Scanner Apps (2023),' TechRepublic: '6 Best Receipt Scanner Apps for 2023,' or the all-knowing ChatGPT.
ChatGPT naming Expensify the top receipt scanning app.
Why does OCR only work 85% of the time?
The reason Expensify is the top receipt scanning app is that we have great OCR technology, but we don’t stop there. We have a much more developed platform that combines OCR to scan the receipt to get information quickly most of the time and then has additional checks to ensure we actually get the system right all the time (well, 99% of the time).
You see, even the best OCR in the world only gets a receipt right around 85% of the time. On top of building our own proprietary OCR technology, we scour the world to find the best OCR technology, and even today, there is nothing that reliably gets real-world receipts correct above 85% of the time. I’m not talking about an emailed receipt or a Home Depot receipt with no tip; those are easy to OCR. Everyone can process those quickly and reliably because we see them a thousand times over and create parsers to extract all the information.
When dealing with receipts for expenses and corporate cards, though, many receipts are scanned after being crumpled up and put in the pocket until the user gets back to their hotel room or has a handwritten tip that looks like it was written by a third grader. These are the receipts where OCR fails and gets the information wrong. All it takes is getting one receipt wrong to lose all trust in the receipt scanning reliability.
This is why we find so many people who rave about Expensify’s receipt scanning abilities. They have come from previous platforms where people might have had OCR, but never invested in getting the receipt right every time. And thus, those users never use the scanning functionality and are left to input all the information themselves.
Why hasn’t anyone copied Expensify’s receipt scanning?
Why don’t these apps go the extra mile? Well, because it’s actually really hard. Expensify has spent the better part of a decade making sure our receipt scanning is the best in the world. It takes a world-class system built of many moving parts.
A mobile app centered around scanning receipts. Expensify’s entire business model thrives on solving a pain point for a lot of people. The most important of those pain points remains allowing employees to take a photo of a receipt and us inputting the information for them. Thus, our entire app revolves around making that super straightforward from the time you sign up.
World-class OCR (optical character recognition). Expensify has built incredible technology around our own receipt scanning functionality. We have scanned millions of receipts and thus use our own proprietary technology to try to use OCR to get the receipt information.
A vast collection of parsers. On top of OCR, we program and continue to build our library that recognizes and extracts information from the receipts we see thousands of times a month. These receipts rarely change, so we know exactly how they are designed and thus have a vast array of email receipt types and top vendor receipts like (Home Depot, Amazon, Delta.com, etc.) that we build parsers to automatically detect and extract information.
A huge network of humans to check the accuracy. This is our not-so-secret, but never copied, sauce. After the receipts go through OCR/parsers, if we’re not incredibly confident the receipt is correct, we have a human check the inputs to make sure they match what our OCR technology has. This requires having a network of thousands of people working 24/7 ready to validate receipts whenever they come through. No one copies this because it’s super hard to load balance. Think of how many more receipts Expensify gets at the end of the month than in the middle with everyone rushing to get their expense reports in. But this is what makes Expensify the best receipt-scanning app around.
Proprietary matching of receipts to bank feeds. After we extract all the receipt information for the image, we then match receipts for users to their personal credit card and company card feeds. This is great for helping users reconcile their credit cards and understand where their spending is coming from.
AI-based receipt auditing for more validation. For our corporate customers, we take the receipt scanning a step further and let the admins know which receipts have been manually changed vs what our receipt scanning technology produces. This gives them an added ability to make sure their users are submitting the correct information.
If you combine all of these things, you've got yourself the world's best receipt-scanning app. Yet, only one company actually does those things, and that is us. Getting the receipt right all the time is what has continued to propel Expensify to the top of receipt scanning lists and all of financial software.
While we’ve mastered this for business use cases, we’re excited to bring much more functionality for consumer use cases as well in New Expensify. Check it out over at new.expensify.com and learn what we can do for your receipt scanning needs today.
