What tampered financial documents actually look like in India and why they're getting harder to catch
This is not a Precisa product pitch. It is something we think more people in Indian fintech should understand.
Financial document tampering in India has evolved through roughly three generations in the past five years.
Generation 1: Basic PDF editing. Someone opens a bank statement PDF in an editing tool, changes a balance or a credit amount, saves it. Detectable by metadata inspection — the modification date doesn't match the creation date, or the editing software signature is visible in the file properties. An analyst who knows where to look catches this in under two minutes.
Generation 2: Print-scan-reprint. The edited document is printed and rescanned, which strips the original metadata. Harder to catch on metadata alone. Detection shifts to font consistency checks, pixel-level analysis of text rendering, and checking whether transaction reference numbers follow the bank's known format.
Generation 3: AI-generated statements. No original document is modified. A synthetic statement is created from scratch. Font is correct. Metadata is clean. Reference numbers follow the right pattern. Running balances add up. The only detection route is behavioural — does the transaction pattern make sense for this borrower profile?
Most Indian NBFCs and lenders have detection capabilities for Generation 1. Some have tools for Generation 2. Almost none have systematic Generation 3 detection built into their underwriting workflow.
The same generational progression is happening with GST returns and ITR documents. Generation 1 GST fraud ,manually edited PDFs , is well understood. AI-synthesised GST documents that pass format checks are not yet on most lenders' radar.
The fix is not a single tool. It is a cross-document verification habit: bank statement, GSTR, and ITR checked against each other at the transaction and declaration level. Documents can be fabricated individually. Fabricating all three with internal consistency across all signals is significantly harder.
