Veryfi Fraud Framework for CPG & FinTech

Digital image forensics for detection of traces of forgeries etc.

Helen Birulia avatar
Written by Helen Birulia
Updated over a week ago

Veryfi is addressing Fraud for CPG and FinTech verticals

Fraud detection is vital for CPG and FinTech verticals to protect financial assets, preserve customer trust, reduce operational costs, ensure compliance, prevent brand damage, and promote fair competition. Veryfi is investing in robust fraud detection processes, to enhance customers security posture and maintain a competitive advantage in the market.

CPG – Brands, Loyalty Marketing

FinTech – Billpay

Fraud can also make it more difficult for legitimate customers to claim and redeem rewards points, which can lead to dissatisfaction and a decline in engagement with the program.

According to a study by the Collinson Group, a global loyalty and benefits company, fraud accounts for an average of 15% of loyalty program costs. This includes costs associated with rewards points that are fraudulently claimed or redeemed, as well as costs associated with fraud prevention and detection.

The impact of fraud on brand loyalty programs can be significant;

  • It can lead to financial losses for the company,

  • Reduced trust in the program among customers, and

  • a decrease in customer loyalty.

Fraudsters create counterfeit invoices or manipulate payment details to divert funds to their own accounts, leading to financial losses for bill pay providers and their customers.

Counterfeit Invoices: Fraudsters may create entirely fake invoices that appear legitimate, using the names, logos, and branding of reputable businesses or service providers. These invoices can be sent to unsuspecting individuals or businesses, who may unwittingly make payments for goods or services that were never delivered.

Altered Payment Details: Another form of fraud involves altering legitimate invoices or payment details. Fraudsters may intercept or modify invoices in transit, changing the payment beneficiary's bank account information to divert funds to their own accounts. This can lead to legitimate payments being misdirected and funds being lost.

According to our most recent Financial Professional Census, the average estimated cost of invoice fraud to middle markets businesses is an eye-watering $280,000 per year.

Finance teams are identifying, on average, one invoice fraud attack every month, notwithstanding those which slip through unnoticed. In total, the 2,750 businesses surveyed uncovered more than 34,000 cases of invoice fraud in the space of a year.

Veryfi offers a two stage approach to fraud

Veryfi Lens and Veryfi API

Lens Document Capture Framework

Supercharge your mobile app with the world’s fastest, accurate, machine-powered data extraction framework. In minutes your engineering team can enable expense management capabilities, support for paper receipt, invoice or bill scanning and back-office paper digitization by dropping in Veryfi Lens into your native or hybrid mobile app. It’s that easy and fast.

Contact with your intent and please let us know how you want to use the Veryfi Lens product.

Veryfi Lens for Document Capture supports native mobile platforms, cross-platform SDKs and browser-based web applications.

Adding Veryfi Lens to your mobile solutions will delight your users with innovative camera features that increase document quality, accuracy and prevent fraudulent submissions.

Signal Type


Is Blurred

If captured image drops below a Low quality threshold, a key “is_blurred” boolean value indicating this state is returned in the notification.

Document Classifier

Helps you identify what type of document was captured into memory.
Use “is_document” and “document_type” keys to base your next step decision on.

Computer Screen Detection

User tried to capture a receipt from their computer screen (LCD). This is typical of when user manipulates the receipt in photoshop. The key “lcd_screen” returns a list of probabilities that it was a screen for the last 30 frames.

Detected Objects

What objects Lens saw during the last 30 frames. These objects are returned in the “detected_objects” along with the probability score of it being that object.

Device Angle

The angle in degrees the phone was in during capture of the document. See key “device_angle”. A capture process is typically 0-30 degrees and anything closer to 90 could signal the user it taking a photo of a screen.

Example JSON

Note: signal/s arrive on veryfi LensUpdate delegate.

"status": "captured",
"is_blurred": false,
"is_document": true,
"document_type": "receipt"
"device_angle": 90,
"lcd_screen": [0.8, 0.7, 0.9, ...], // Last 30 frames
"detected_objects": [
[ // Frame 1
"dining table":0.109375
], […] //Frame 2

Veryfi API fraud is an extension to Veryfi data extraction

Signal Type


Example JSON

Duplicate Detection
Duplicate spike alert

Existing duplicate detection looks at: total, invoice #, date and vendor and some other assets etc.
The next generation of duplicate detection will look at OCR_Text and run phonetic algorithms to compare. But even at this stage duplicate detection can be customizable per your specific use case.


Document Classifier

User tries to submit a handwritten document or one created in MS Word.

"is_document": true,    "document_type": "receipt",

Secondary (Lens was 1st) to check how many pages in the submitted document are blurred.

  "is_blurry": [
0: false,
1: true

User has added a new line item by hand and/or modified the total by adding a 0 to the total to exponentially increase their spend claim.

  "meta": {
"handwritten_fields": [
0: "date",
1: "subtotal",
2: "total"],
  • Tax + Line items = Total

  • Repeating line items (SKUs or Descriptions)

  • Total Items sold != SUM of Line Items

  • etc.

   "reconciliation_checks": {
"totals_match": true,
"has_dupes": false,
"count_items_sold": true
"warnings": [
"code": "WSBTL",
"message": "Subtotal for Line Items 638.13 doesn't match subtotal on document 638.18"
"code": "WLIPRBL",
"message": "Line item #5 has a problem. Quantity x Price != Total"

Detects Moiré pattern due to interlaced scanning.

 "is_lcd": false

Did this answer your question?