Financial institutions are required by compliance regulations to perform KYC (Know Your Customer). Before opening an account, they must verify a series of documents such as ID cards, address proofs, and bank statements, manually comparing each one—efficiency is miserably low. Traditional banks spend **$2,200** per case, and errors accumulate: assuming 95% accuracy per field, only 60% of applications pass directly after 10 fields.

LlamaIndex co-founder Jerry Liu recently published a tutorial showing how to automate this process with an AI agent. The core idea is simple: first use LlamaParse to extract structured data from various documents, then use Claude to cross-verify consistency across documents.

It’s divided into two steps:

**Step 1: Customer Identification Program (CIP)**. Extract name, date of birth, address, and ID number from government-issued driver’s licenses. LlamaParse returns extraction results along with confidence scores and source text references for easy auditing.

**Step 2: Customer Due Diligence (CDD)**. Extract information from utility bills and bank statements, then have Claude act as a compliance analyst to compare whether the identity information in the three documents matches. Claude can handle format differences like "J." corresponding to "Jason" and "Street" corresponding to "St", and can also explain its judgment reasons.

The tutorial uses three interesting samples: an official driver’s license (with photo, watermark, barcode), a synthetic utility bill with a multi-column layout, and an **image-based PDF bank statement without a text layer**—the last one directly filters out most simple extraction tools.

The tutorial includes complete Python code and a sample document generation script, which can be run directly on GitHub. The link is below.

---

**Further Reading**

- Tutorial link: https://github.com/jerryjliu/llamaparse_use_cases/blob/main/kyc/tutorial.md

- LlamaParse: https://cloud.llamaindex.ai