Fun fact: Did you know ‘invoice' originates from the Latin word 'invocare,' meaning 'to call upon?'
But what is an invoice, exactly? In short, it documents the relationship between a seller and a buyer: the former of whom 'calls upon' payment.
Next in line is Accounts Payable (AP). AP is the total sum owed to the seller and found on the balance sheet. Alongside the AP, companies will log the amounts they owe themselves under the credit category, ‘Accounts Receivable’ (AR).
AP and AR data are common inhabitants of invoice PDFs. Accountants and analysts must aggregate invoice data to understand its financial position and represent it on their financial statements. However, compiling invoice data into spreadsheets or internal databases for further calculations is extremely repetitive and tedious. Did we mention that ‘invocare’ also means ‘unpleasant’ or ‘disagreeable’ – how appropriate.
Enter automation.
Automation offers an easier way for analysts and accountants to ‘call upon’ their invoice data. You can automate various parts of the AP process by:
Optical Character Recognition (OCR) is the technological glue for most automated invoice processing. OCR captures the data from invoices for downstream processing, saving analysts time by automatically extracting the data. But how does OCR work, and, more importantly, does it work?
OCR is a key ingredient in document processing workflows. It leverages predetermined rules to match characters against those stored in its internal database. One of OCR’s most prominent projects is digitising newspapers – media uniformly printed in set fonts and structures.
We’ve tested OCR for invoices extensively as part of developing our data extraction product, Transcribe. Because of its prescriptive nature, OCR technology tends to generate inaccurate information (or even gibberish) from poor-quality invoice scans.
OCR’s ineffectuality was an insurmountable obstacle for our founders’ previous company (one that built technology to automate tax calculations). Without any additional validation mechanisms, OCR is extremely limited, which can affect the AP process negatively. Let’s examine a few hypothetical scenarios and the limitations of each one.
Consider an invoice that might contain handwriting. The OCR engine has no prior handwriting templates (even if creating uniform templates for all handwriting styles was possible), meaning the invoice would need manual review. The result? A delay in the approval process.
Let’s say an invoice contains a radically different font from the templates on which the OCR technology was trained. If the OCR algorithms mistake an ‘8’ for a ‘0’, the resulting payment is an underpayment. Such an error would permanently damage the purchaser’s relationship with the seller.
In an even more serious scenario, an auditor can discover a mismatch between actual transactions and reported financial data during an audit. The finding might result in fines or penalties.
Firms sticking to the status quo (i.e. manual data extraction) will likely lose money because of inevitable errors and lost productivity. Not to mention, OCR is only one potential culprit of errors in automated invoice data processing (alongside system glitches, integration errors, human error, etc.).
No matter what, analysts must be able to call upon a clean, structured dataset when making essential decisions, such as:
It’s not enough to simply receive extracted data for manual review and approval. Case in point, – our client, Unigestion, used a previous supplier that delivered data with 30% accuracy. As a result, their employees spent too much time (and other resources) correcting data generated by a technology designed to improve productivity.
Accurate data should not be too much of an ask in 2024. After all, newer, AI-powered technologies can design toothbrushes, detect fake news and ward off thieves from your pet’s dinner. In comparison, is generating an Excel spreadsheet containing actionable invoice data too much of an ask?
Not at all. Newer AI technologies take the baseline of OCR technology and add a layer of Natural Language Processing (NLP). NLP allows the technology to ‘read’ the invoice document like a human. For example, a human would instinctively understand handwriting on an invoice means.
OCR would not. Yet, not only could AI-powered OCR understand it, but it can also copy, validate and structure it alongside all the other desired data.
AI + OCR for Invoices = A Dynamic Duo. The top two advantages of AI-powered technology include:
1. Speed
Eliminating the ‘manual review’ step will shave off hours per annum for any employee. By cutting mundane work from your employees’ daily schedules, they can focus on goal-smashing activities instead.
2. Cost Savings
Q: What do you get when you eliminate costly manual errors, boost employee productivity, improve the cost-effectiveness of your company’s current technology, improve relationships with suppliers and speed up core operational processes?
A: Substantial cost savings (> 75%).
There is no best way to add AI to existing automation technology, much less implement AI-powered automation. Instead, you might consider factors like:
Determine how fast you would like the invoice data. To trim unnecessary time off the process, consider accessing invoice data without manually uploading or downloading invoices to a UI. You can achieve this integration via REST API.
Determine whether you plan to use the invoice data for downstream AP processing. For example, if you’re using invoice data to generate and issue payment documents to suppliers, you’ll need the ability to customise your data in a preferred format. Flexibility will be a key concern when shopping for your AP technology vendor.
For more information, speak to a well-established automation vendor.
Ready to automate extracting data from invoices? You’re in safe hands with Evolution AI. We’ve worked with several leading global firms to capture data with an innovative approach:
Evolution AI’s multiple award-winning data extraction technology allows you to access your data your way.
We’d love to hear about your business’s use case and how we can help you meet your goals. Contact our financial data project team – book a demo or email hello@evolution.ai.
