Detect fraudulent customers using machine learning and forensics
Today, many digital marketplaces and financial service providers use a waterfall process for onboarding new customers, consisting of various checks such as email risk scores and device ID blacklists. These onboarding processes also likely require a proof of identity and a proof of address for all applicants or those considered high risk. Know Your Customer (KYC) regulations require companies to make reasonable efforts to establish the identity and address of their customers.
How are the majority of identities and addresses confirmed today? Documents. More specifically, the human review of documents. Typically anywhere from 40%-75% of new customers of a marketplace or financial service provider are required to submit documents to prove identity and address, in addition to documents for other data points such as income or assets.
However, the business practices of yesterday which rely on human review for mundane tasks have expired. Today’s leading businesses are replacing manual processes with faster, more reliable, and more scalable automated processes. Manual tasks such as reviewing documents are being replaced with software systems like Inscribe.
What does a document verification workflow look like?
- Can I trust this document?
- Is the information valid?
- Is the information correct?
Here at Inscribe, we’ve built a product, without humans-in-the-loop, to automate this three step process of document verification.
Can I trust this document?
Before processing an application you first need to know if you can trust the information on it. Our suite of 20+ fraud detectors examine the metadata and pixel-level information of documents to ensure the integrity of the information presented. Additionally, we compare documents to templates of known fraudulent and known legitimate documents. Detecting fraud in documents is Inscribe’s competitive advantage in the market for automated document verification.
Is the information valid?
Inscribe automatically classifies documents by their type (think bank statements vs. utility bill). When processing documents, it's important to know what type of document is being examined. The ability to determine if a document is a bank statement, pay stub, invoice, utility bill, etc. allows you to request your users to provide the correct document. It also allows Inscribe to treat documents differently based on their classification when detecting fraud and extracting information. Our trained machine learning (ML) models allow for 98% accuracy on document classification.
Additionally, it is important to understand the recency of the document. For compliance, it is often required that some data points, such as address, need to be verified using information issued within the last 3 or 6 months. Inscribe uses optical character recognition (OCR) and intelligent parsing logic to identify the relevant date within the document to ensure the examined documents are within the accepted recency window (3, 6, 9 months etc).
Is the information correct?
OCR technology isn’t new but reliably parsing key details such as name, and address has been a surprisingly difficult problem to solve; until now! Inscribe’s trained ML models allow for 95%+ accuracy on name, address, and date parsing. There are two ways for customer onboarding teams to work with Inscribe’s parsing capabilities:
- 1) Verification: where the customer's name, and address are provided to Inscribe along with the document. If the name given matches the name present on the document, Inscribe’s API responds with a binary verification (same for address and dates). Furthermore, it's essential for fuzzy matching to occur as not all names and addresses match exactly with the information presented on the document. Inscribe's solution fuzzy matches both names and addresses to reduce false-positives.
- 2) Parsing: if you don't have a name and address to provide to Inscribe along with the document, don't worry—Inscribe automatically extracts key details from documents that you can then use in downstream processes.
Value propositions from using an automated document verification system:
- Decreased fraud risk for channels with traditionally high-risk applicants
- Increased application processing speeds
- Higher application conversion rates
- Reduced operational costs
- Greater scalability
- Added fraud detection rigor
Interested in automating your document-dependent customer onboardings? Contact sales here