October 30, 2022
•
8 minutes
There’s a new wind blowing in finance, and automation today is nothing like it was ten years ago.
Whether you incorporated an invoice OCR app back when they first came out or decided to stick with manual data entry after hearing about issues with accuracy, it’s time for a rethink.
Finance AI has been growing by leaps and bounds, and automated invoice processing today is nothing like the old-school software.
How Does Invoice Processing Work?
To understand the revolution in invoice processing we need to look at how the invoice processing software works— and what the difference is between the old and new.
First, let’s look at what happens when you process a document with legacy software that uses template-based methods to locate fields.
Invoice Processing with Legacy Software
- Upload. Scan any paper invoices, then upload them to your software service. This might involve selecting files and uploading them via a web interface, or, if your software has an API integration, you might use that.
- Extract your data. Behind the scenes, this is a multi-step process. The first is optical character recognition (OCR); the computer translates the scanned invoice into computer-readable text.
This is done by matching the letters in the scanned image to character sets stored in the database.
Once that’s done, it compares the invoice to a template and uses the comparison to locate key fields like payment due, bank details, invoice number and PO number. These are mapped to files in a database file. - Validate your data. This involves checking the numbers to make sure they make sense and are accurate. With legacy software, validation may be a manual check, or you may have a validation module that runs a few confidence checks.
- Export your data. In step 2, the invoice information was converted to structured data; the type of data computers work with and manipulate. With a click of the mouse, structured data is exported to your accounting system, where it can be used to pay invoices.
Notice the key to the extraction process is matching. And matching works great when two things look alike.
When the invoice you’d like to get processed looks like the model invoice, the invoice your processor uses as a template, you’re in for clear sailing.
Problems tend to occur when a supplier invoice doesn’t use quite the same format or uses a word or two that is not where it was expected. Then the software draws a blank, or might mismatch fields and give you extracted data that just doesn’t make sense.
You’d wish they could just make invoices with one standard format, but they don’t. You get a bunch of look alike, and your software performs beautifully, and then you get invoices from a more creative supplier and that is the end of that.
Of course, there’s another option. If you’re yet to go in for AP automation, your staff might still be copying out the data by hand.
There’s nothing wrong with manual data entry— except that it is slow, error-prone, and frustrating for whoever has to do it.
No one likes to be saddled with a job that a computer could do better, and invoice data entry is one of the more boring tasks in any AP department.
Invoice Processing with a New Invoice OCR App
Suppose you scrap the old way of doing it and go with a cutting edge AI-based solution— Affinda’s invoice extractor, for instance. In this video, Brendan explains why Affinda's invoice extractor is a great choice.
What’s going to change, and what would the procedure look like now?
What you do is much the same, but what the computer does is very different. More importantly, the results are different. Here’s what the procedure looks like for Affinda’s invoice extractor:
- Scan and upload your files. This could be dragging and dropping them into Affinda’s online invoice extractor, or taking advantage of the API to upload them in your own AP interface.
- Push extract, and Vega, the AI brain behind the extractor, goes to work. First on the scene is a new and improved intelligent OCR, producing a highly accurate text version of any scanned document.
Then the invoice itself is ‘read’ in much the same way a human would read it. Natural language processing is used to get at the semantic meaning of every word on the page, leading to a quick, extremely accurate extraction that can elucidate any field desired. - Data validation is run by a second automated process, and you’re given confidence levels for each extraction. Anything that doesn’t meet standards is flagged for manual review. An inaccurately parsed invoice can be fed back to the system for training, ensuring it doesn’t happen again in the future.
- Now it’s time for exportation. All you do here is push a button, then your processed invoices— or the data extracted from them— are exported to your accounting software.
With robotic process automation, you can set everything up for automatic payments, with just a quick manual confirmation step. Alternatively, if you’re working with the online interface, you can download the structured data extracted from your invoices in one computer-readable file.
What Changed in Invoice Processing?
Behind the scenes, there’s been a big change in invoice processing. But the biggest difference is not in the procedure, it's in the results. Here are some of the changes you can expect to see in your results:
- Better accuracy
- Better speed
- More flexibility
- Custom fields are no problem
Sound too good to be true, like a wishlist someone dreamed up on an optimistic day? That was our wishlist when we started designing our new invoice processing app.
And it worked!
The idea is to use artificial intelligence to build an engine - a robotic brain of sorts - that understands human language. That's just what natural language processing (NLP) is. Then you teach that robot finance and how to process invoices.
That's just what VEGA does.
Your Invoice Processing AI Became Smarter
AI became much more powerful than it used to and it keeps getting smarter. This progress is exponential, so it's becoming faster, as well.
Before, automatic invoice processors struggled with different invoice formats. Today, it doesn’t matter if your supplier invoice comes in a format that you’ve never seen before.
Your AI OCR invoice app looks at the invoice format— just as your human staff member would— but, just like that human, it uses context to understand the fields, columns, rows, and much more.
If an invoice can be decoded by an ordinary human without extra background knowledge, your invoice processing app should be able to do it, too.
Vega does, btw.
Your OCR Invoice App is Trainable
If you have particularly specific invoices to process, your AI will need more context. Today, with the new OCR invoice apps, you can teach your app to understand that context.
That’s where flexibility comes in, powered by deep learning modules that are never finished getting smarter. Think continuous improvement, app-style.
If you test an AI engine, like Vega, it will first recognise quite a bit about human language. It has already been taught to expect lots of finance and invoice-like vocabulary.
But that's not all. There’s a lot more industry-specific machine learning that can happen. You can do that training yourself, or in the case of Vega, we can it for you.
The same goes for custom fields.
Modern AI invoice processor should be able to extract pretty much every field you could think of, but if you have another custom field you’d like extracted, it should be possible to set it up.
Your New Invoice Extractor Will Be Secure
A modern invoice OCR app needs to be super-secure. With today's AI-based apps, the security is taken to the next level, and you know your documents are 100% safe.
In case of Affinda, not only is your data kept in a secure location, it’s encrypted with state-of-the-art encryption to ensure no one but you has access to it.
Modern AI Invoice Extractors Merge With Your Apps Seamlessly
If invoice processing is at the core of you business, you need to make sure that your AI solution works well with the systems you already have. Integration without revealing that you are using a third-party solution for your business tools is a must.
White labeling whould not be an non-issue. Want to incorporate Affinda’s AI-based invoice processor into your own software?
Sounds great— let’s talk!
Affinda’s invoice OCR app is designed for easy integration. There's the API option, and Zapier integrations are piling up as we speak.
You might be looking at these characteristics of modern AI invoice extractors and thinking:
"Yeah, but that's not true for the AI solution I'm using right now"
We hate to break it to you, but, if that's the case, you should reconsider your choice.
That takes us to:
How to Choose and Test Your New Invoice OCR App
You might not be ready to spend money on a new and untried idea. We get that.
Software licensing gets expensive, and the solution you have already is functional, though not ideal.
We’re not suggesting you jump ship and sign a contract with an AI based service like Affinda without being absolutely, 100% sure that it will save you money in the long run.
We wouldn't want you to, either.
That’s why Affinda offers an extended free trial. You can access our invoice processor for free from your web browser, or if you’d like a more serious trial, contact us and we’ll get you a free API key.
We’ll make sure you get enough credits to thoroughly test our software.
Maybe it won’t work exactly as you’d like it to. Maybe you work in a specialized industry, and Affinda’s Vega doesn’t seem to understand all the jargon.
Don’t sweat it. You wouldn’t expect a new employee to perform flawlessly without training, and, as intelligent as our AI engine is, maybe you shouldn’t expect anything flawless in the first round either.
That's why you need to test it with a bunch of invoices. One or two won't do the trick. Dump a batch of 20-30 invoices and try then.
What you can expect is an engine that learns quickly— very, very quickly—- quicker than a human employee— and gets better and better with each invoice that is processed.
We’d be happy to train our AI for your special situation or industry, so you get reliable, accurate results you can count on every time.
So give us a call, or drop a note through our contact form. There’s nothing to lose, and a whole lot to gain.
