Onboarding Quality Agent AI

Invoked when: a file is submitted for credit review — QC audit is completed within 2 hours of submission

• –Checks every submitted file against the product-specific document checklist — the list of mandatory and conditional documents required for each loan product. Mandatory documents (identity, address, income, bank statement) must be present and legible for every application in that product category. Conditional documents (co-applicant KYC if a co-applicant is named, property documents if the product is secured, business documents if the applicant is self-employed) must be present if the application condition that triggers them is met. Missing mandatory documents are a hard QC fail; missing conditional documents where the triggering condition is present are also a hard QC fail.
• –Assesses document quality against the minimum legibility and currency standards — a document that is present but illegible (blurred scan, poor lighting, cut-off edges) fails the QC in the same way as a missing document, because the KYC Verification Agent AI cannot verify an illegible document. Currency checks confirm that the document is within its validity period — an Aadhaar card that is expired or an income document that is more than 3 months old fails the currency check for the applicable document type.
• –Cross-checks the data fields within the file for internal consistency — the name on the PAN must match the name on the Aadhaar (allowing for standard variations in transliteration); the address on the utility bill must be in the same city as the application-stated address; the income declared on the application form must be consistent with the income shown on the salary slip within a reasonable tolerance. Internal inconsistencies that are within the tolerance range are flagged as advisory; those beyond the tolerance range are a hard QC fail.

Invoked when: a file fails the QC audit — error tags are assigned to each failure before the correction instructions are dispatched

• –Tags every QC failure with a specific error category from the standardised error taxonomy — the 15 most common error categories that cover the majority of QC failures: missing mandatory document, expired document, illegible document, name mismatch (minor variation), name mismatch (significant variation), address mismatch, income document outside currency window, incomplete bank statement (missing months), missing co-applicant KYC, missing guarantor declaration, application form incomplete (specific fields), signature missing, photograph quality insufficient, and document fraud signal (referred to Fraud Detection Agent AI). Each error is tagged to exactly one category; where a single document fails on multiple grounds (expired and illegible), each failure is tagged separately.
• –Assigns a correction instruction to each error tag — the specific action the operations team must take to resolve the error. The correction instruction is borrower-facing where the borrower must provide a new or corrected document, and internal where the error is a data entry issue that can be corrected in the LOS without borrower contact. Borrower-facing corrections include a script for the operations team to use when contacting the borrower — ensuring that the borrower receives a clear, consistent explanation of what is needed and why.
• –Sets a resubmission deadline for each QC fail — 48 hours for errors that require the borrower to provide a corrected document, and 24 hours for internal data correction errors. Files that are not resubmitted within the deadline are escalated to the operations head and included in the Onboarding SLA Agent AI's TAT monitoring as a QC-hold delay.

Invoked when: an application is declined at QC or credit stage — rejection reason is logged and the borrower feedback loop is triggered

• –Logs the specific rejection reason for every application that is declined — whether at the QC stage (file rejected for failure to correct errors within the resubmission deadline) or at the credit stage (file passed QC but declined on credit grounds). The rejection reason log is the dataset that enables the borrower journey improvement cycle — identifying which rejection reasons are most frequent, whether they are concentrated in specific channels or geographies, and whether the rejection rate is improving or deteriorating over time.
• –Feeds the rejection reason data back into the onboarding process — where a specific error category is responsible for more than 20% of QC failures in a week, the agent flags the pattern to the operations head and the onboarding UX team. A systematic error pattern indicates a problem in the borrower-facing onboarding flow (borrowers are consistently submitting the wrong document type) or in the operations team's onboarding guidance (borrowers are being incorrectly briefed on the document requirements). The fix is a process change, not more error correction.
• –Ensures that every declined applicant receives a clear rejection reason communication — the specific ground for decline and, where relevant, what the applicant would need to do differently to reapply. The rejection communication is dispatched by the operations team using the agent's logged rejection reason as the basis; borrowers who do not receive a rejection reason have no path to remediation, which increases complaints and regulatory exposure.