circuit

Improve Fraud Detection Using AI-Powered OCR

Typical frauds business owners face — duplicate receipts, shell companies, and altered invoices — and how an AI-powered OCR solution protects them.


Pop quiz time. What do giants like Amazon and Google have in common with Mr. & Mrs Jones' wholesale grocery across the street? Or Johnny Automobile's car repair business downtown?

If you ask the FBI, it's how vulnerable both are to being scammed out of millions by invoice and receipt fraud.

It doesn't matter if you're a mom-and-pop store or a multi-billion dollar concern. The great equaliser is the sheer volume of receipts and invoices.

You could have the hardest working, most experienced Accounts Payable team in the world, and the numbers game would *still *make it impossible for them to manually sift through hundreds per month and check for fraud, driving up the chances of a phoney invoice or two sneaking through - and that's all the opportunity criminals would need to con you out of thousands to millions.

The Real Bottleneck in Fraud Detection: What's Stopping Us?

Spoiler: it's not digitisation.

General purpose OCR has gotten good enough that you can rely on them to scan all your purchase orders, invoices, and receipts, digitise them and throw them into a PDF...but forcing your people to manually eyeball those 3 pieces of documents and compare each set for fraud is neither generating business value effectively nor ensuring due diligence.

The real bottleneck, then, is retrieving relevant, structured data from documents automatically to begin with. If you can optimise this, it would streamline your entire fraud detection workflow - all the way from purchase orders filed, to payments released.

That's where TAGGUN comes in.

TAGGUN is a next-gen AI/Machine Learning solution for receipt scanning, that combines state-of-the-art-OCR with deep learning, regular expressions, and fuzzy-matching for logically related fields - and operates as a secure, easy-to-use, easy-to-integrate REST API in the cloud.

You send in images or scanned PDFs of your purchase orders, invoices, and receipts, and get back structured data in real-time, in JSON, ready to be used. No queues. No pooling. No need for dedicated on-premise AI/ML hardware of your own.

Generating structured, easily-parsable data from just images of receipts is pretty neat, sure, but how does this help your fraud prevention teams? For that, let's talk about the 3 most common types of business fraud, and see how TAGGUN's features might help in each case.

The 3 Main Kinds of Fraud, and How TAGGUN Protects You From Them

1. Duplicate Receipts or Invoices

How traditional OCRs fail to detect duplicate receipts or invoices

The simplest kind of fraud is also the most likely to succeed, depending on how your internal bookkeeping works.

Most Accounts Payable systems that work with OCR tech will just accept a receipt or invoice, scan it, see if the numbers are where they are expected to be, in the format/layout they are expected to be in, and mark it as a successful operation if so.

Traditional OCR + template-based security measures that check for validity do not take into account the simple possibility that the original invoice, while valid, may have been submitted for reimbursement more than once by a shady supplier.

How an AI-powered OCR solution can help identify duplicate invoices

Where traditional systems may stumble, an AI-powered OCR solution like TAGGUN succeeds because, for each and every receipt added to the system, TAGGUN generates an MD5 hash - a sort of digital signature unique to each image/scanned PDF. These can be stored with even lower storage requirements than the actual image/scanned PDF.

TAGGUN generates these cryptographic hashes for you on the server side and sends them back as part of the JSON response. Having this value for each invoice means your AP/Fraud teams no longer have to manually pore through documents line by line to check for duplicates - this comparison process can be entirely automated with software, at a *much *faster pace than human beings ever can, using very little computing power. If the hash values are different, everything's good. If there is a match, it indicates that the current invoice is a duplicate, and can be automatically flagged by the system for further review.

An AI-powered OCR solution can offer business owners a lot more compared to a general-purpose OCR. Here's an article that covers this in detail.

2. Shell Companies

How business owners can be deceived by shell companies or fake vendors

The premise is simple: someone studies how your company works over time - the compliance services and subscriptions you use, your caterers and their food bills, and the suppliers you use for office goods...and then, they create a phantom/shell company, one that has no presence but for the mailing address, and send in invoices that look *exactly *like the legitimate ones you already pay in bulk per month, to collect disbursement from you for goods/services that were never sent.

How an AI-powered OCR protects businesses from fake vendors

An AI-powered OCR like TAGGUN automatically verifies the validity of any transaction and vendor on receipts/invoices, using a number of search engines - the VAT Information Exchange System (VIES) for VAT numbers in the EU, the fapiao for China, and IBAN, internationally. This information is included in the JSON response you get for each scan, like so:

This way, you can be sure that a legitimate transaction actually took place, and that you're being charged by a supplier who is legitimate - they are who they say they are and are properly registered in the country that they say they are.

If TAGGUN doesn't send you this verification, you can flag the transaction for manual review.

Getting away with faking VAT, fapiao, and IBAN information is incredibly difficult with the current authentication systems in place worldwide, so automatically having that information at your fingertips is peace of mind for you as a company.

3. Altered Invoices

How frauds use altered invoices to trick businesses into paying more for less

Let's say you, as a business, have ordered 10 high-resolution workstation monitors from a new supplier, Seller X, at $1000 each, for your dev team. Such high-value, one-off purchases make shady vendors like Seller X see an opportunity for something a little south of legitimate. Perhaps they decide to act on it and change the specifications of the monitor shipped to one from the same manufacturer, but a lesser model...but charging you the same per-unit value, betting you won't look into technical specs beyond the manufacturer name and cut and run with their profits.

How TAGGUN protects you

This scenario - like many others - is the perfect case for proper 3-Way Matching. And TAGGUN streamlines obtaining all the data you'd need for implementing it.

Essentially, every business transaction should ideally create three pieces of documents that should be compared for discrepancies.

  1. The Purchase Order (PO), which includes details such as the product or service being purchased, the quantity, the price, and the expected delivery date.
  2. The Delivery receipt, or a receiving report, which is a document that confirms that the product or service was received by the buyer. The receiving report includes details such as the product or service received, the quantity, and the date of the received shipment.
  3. Supplier's invoice, which includes details such as the product or service being sold, the quantity, the price, and the shipping date.

If the line item details (product description, model number, per-item costs) don't match up perfectly across all three documents, you flag the transaction and manually review it before releasing payment.

Implementing 3-way matching as a fraud prevention measure is labour-intensive and does not scale with high volume - suddenly your AP team no longer has to manually pore over 1 document, but 3 per transaction, essentially *tripling *their workload - resulting in higher chances of human error leading to losses via fraud, payment delays, late charges, and strained business-vendor relations.

But with TAGGUN's automated, ML/AI-based solution, you no longer have these limitations. TAGGUN enables you to generate business value quicker by streamlining the primary bottleneck - understanding context to quickly (real-time, in under 5 seconds), and accurately (accuracy rate of over 90%) obtain information from invoices and receipts for you. (To learn more about automating accounts payable workflow, you can check out this article.)

Of course, using TAGGUN does not replace your existing ERP software or your AP team; the technology only eases their burden. If you can get relevant data *this *fast, and in the industry-standard JSON format...your AP team no longer has to manually sift through piles of invoices and receipts, trying to spot inconsistencies, and your ERP system can simply use TAGGUN's JSON data as-is, instead of needing expensive third-party licences.

Only the discrepancies will be brought to your attention, and the only decision-making you'd have to do is how to deal with the shady vendor in question.

Automating Fraud Detection is the Only Viable Option

Risk of fraud is Just Another Monday™ of running a business of any kind, at any scale. Vigilance is critical, but manual fraud detection efforts might involve a greater cost than just man-hours and dollars.

Information gathering is labour intensive so personnel stress is always a factor, there are issues with scans of receipts/invoices being of poor quality, misaligned, with artefacts, or all at once; and that pile of invoices and receipts will only grow taller each day if your AP team can't get the data they need, at the pace they need.

TAGGUN helps you to automate the three biggest pain points of your fraud management workflow - gathering accurate information, having it processed in real-time, and having it be structured - saving you time and money, empowering your AP staff, and enabling them to focus on bigger, better things. Feel free to check out the pricing page and choose the plan that suits you best.




Continue Learning