An onboarding error is not a quality problem — it is information. An error that appears in 18% of MSME files points to something specific: a step in the MSME onboarding journey that consistently fails, either because the borrower is not instructed clearly, the RM is not checking correctly, or the system is accepting documents it should reject. When errors are tagged consistently, the error register becomes an improvement roadmap — a prioritised list of exactly what to fix in the onboarding process to reduce rework, reduce processing time, and reduce the number of file errors that eventually become compliance findings. The Onboarding Quality Agent AI tags every error with a consistent taxonomy, produces a ranked error list updated weekly, and routes each error type to the team best positioned to fix it — operations, training, product, or technology.
The error taxonomy: why consistent tagging creates institutional knowledge
Most quality control processes note errors in free text: "bank statement missing August" or "EC too old." Free text is not analysable at scale. When the compliance team wants to know whether bank statement gaps are increasing, they cannot query a free-text field. When the training team wants to know which error types are concentrated in which branch or which product, free text gives them nothing to work with. The Onboarding Quality Agent AI tags every error with a standard error code (a category, a sub-category, and the specific field or document that failed), making every error analysable. "Bank statement gap — August 2025" becomes ERR-INCOME-BST-GAP. "EC expired — 8 months old" becomes ERR-PROP-EC-EXPIRED. The error code links the specific instance to the trend, the trend to the root cause, and the root cause to the fix.
The 15 most common errors: November 2025 error register
Bank statement has one or more missing months in the required 12-month window
The most common onboarding error across all products. Borrowers submit bank statements but not as a continuous 12-month set — they submit 12 individual monthly PDFs, often with one or two months missing. Alternatively, the statement covers 12 months but the start date leaves a gap (e.g., November 2024 to September 2025 covers only 11 months). Fix required at the point of document request: the digital onboarding agent must specify start and end dates explicitly.
Encumbrance certificate is older than 6 months at the time of sanction
For LAP and home loans, the EC must be dated within 6 months of the sanction date. Files commonly arrive with ECs that were obtained at application but take 3 to 5 months to reach sanction — by which time the EC has passed the 6-month threshold. The fix is a process trigger: if application-to-sanction TAT exceeds 4 months, a fresh EC must be obtained before sanction.
Bank account number or IFSC on NACH mandate does not match the bank account in the borrower's application or bank statements
A highly consequential error: a NACH mandate with wrong bank details will fail on first debit, triggering a bounce, a penal charge, and a customer service incident — for a borrower who believes their EMI is being deducted automatically. Common cause: RM fills NACH manually from borrower-spoken details rather than copying from verified bank documents. Fix: NACH bank details auto-populated from AA-verified bank statement, with manual entry disabled.
Property valuation report is older than 6 months at sanction, or the valuer is not on the approved panel
Similar to EC expiry but with an additional dimension: the valuer must be on the institution's approved panel, not just any registered valuer. Files sometimes arrive with valuations from valuers who were on the panel when the valuation was done but were removed subsequently. The QC AI checks the valuer against the current approved panel, not the panel at the time of the report.
Key Fact Statement is dated on the same day as or after the sanction date
The RBI requirement is that the KFS is provided before the borrower accepts the loan — meaning it must be dated before the sanction date. Files where the KFS is dated the same day as the sanction letter have an FPC compliance gap: the borrower's signature on the sanction implies acceptance before the KFS was technically in their possession. The fix is a system control: KFS must be generated and acknowledged before the sanction letter can be issued.
GST return filing has a gap quarter in the last 12 months — missing quarter visible in GSTN portal
MSME borrowers must have consistent GST filing to use GST returns as an income proxy. A missing quarter (filed late or not filed) creates a gap in the income verification chain. The QC AI checks GSTN filing status for all 4 quarters — not just whether the registration is active, but whether the returns are filed and up to date.
Monthly bank statement credits are more than 30% below declared monthly income in one or more months
A borrower who declares ₹2L/month income but has bank credits of ₹1.1L in 3 of the 12 months either has a seasonal income pattern (which should be documented) or has inflated their declared income. The QC AI flags this for human review — it does not automatically reject, because seasonal variation is legitimate, but it requires explanation and documentation.
CERSAI encumbrance search not present in the file, or CERSAI search is older than 30 days before sanction
The CERSAI search must be completed close to sanction to catch any new charges registered since the EC was obtained. A CERSAI search done 3 months before sanction is stale — a new mortgage could have been registered since. Institutions sometimes forget to do a fresh CERSAI search closer to sanction when the original search was done at the time of EC procurement.
One or more pages of the loan agreement requiring a signature are unsigned in the digital record
eSign completion is checked by the QC AI against the list of required signature pages in the agreement template. A common failure pattern: the eSign flow was completed but a page added later (such as a revised EMI schedule after a rate change) was not re-signed. The QC AI checks every required signature page against the completed eSign record.
Address on Aadhaar differs from declared address by more than the permitted deviation — different city or state
Borrowers whose Aadhaar address reflects a previous residence are common, particularly urban migrants. The permitted deviation is within the same district. A different district requires an address proof document alongside the Aadhaar. Many files arrive without the supplementary address proof, relying only on the Aadhaar with the old address.
ITR computation sheet present but the NSDL acknowledgement number is missing — submission not verifiable
Some borrowers provide the ITR computation sheet (printed from the CA's software) without the NSDL filing acknowledgement. The acknowledgement is the evidence that the ITR was actually filed — without it, the computation sheet is a declaration, not a verified filing. The QC AI checks for the acknowledgement number and cross-validates it against NSDL if the API is available.
Processing fee charged does not match the fee schedule and no approved waiver document is present
RMs sometimes offer a processing fee concession to close a deal without obtaining a formal waiver approval document. The loan is processed, the lower fee is collected, and the waiver document is never filed. The QC AI flags any fee that does not match the schedule and checks for a waiver approval from the appropriate authority — the fee concession itself is not the issue; the missing waiver document is.
Under-construction property for which RERA registration number is not present or RERA status is not verified
For home loans on under-construction properties, RERA registration of the project is mandatory under RBI guidelines. Some RMs submit applications with the builder's RERA application number (pending registration) rather than a confirmed registration number. The QC AI checks the RERA portal for active registration status — a pending application does not satisfy the requirement.
Photograph on submitted identity document does not match the applicant photograph in the file — face verification failure
The QC AI runs a face verification check between the borrower's selfie (from digital onboarding) and the photograph on the Aadhaar or PAN. A significant mismatch — more than what would be explained by age difference or image quality — triggers a flag. This check catches cases where a document belonging to a different person is submitted in support of an application.
Borrower's signed consent for credit bureau enquiry is dated after the bureau pull date
Bureau enquiry requires prior consent. Files sometimes show the bureau pull date (visible in the CIBIL report) preceding the consent document's signature date — meaning the institution pulled the bureau before the borrower signed the consent. This is a procedural and regulatory compliance gap. The QC AI compares the bureau pull timestamp in the CIBIL report with the consent document's signature date.
The NACH mismatch error is not a quality problem — it is an EMI bounce waiting to happen, with a penal charge attached, and a borrower who will call support confused about why their automatic payment failed
The NACH mandate with the wrong bank account number passes every other check in the file. The KYC is clean. The income documents are complete. The property title is clear. The only thing wrong is 11 digits in the NACH form. That error is invisible until the first debit attempt — which fails, generates a bank return code, triggers a penal interest charge on the borrower's account, and creates a service call that takes 20 minutes to resolve and leaves the borrower questioning whether the institution's processes are reliable. The Onboarding Quality Agent AI finds the 11-digit error before the file reaches sanction. The bank account on the NACH is compared to the bank account on the verified bank statement. If they do not match, the file is held until the NACH is corrected. A NACH mismatch costs the institution a first EMI bounce, a penal event, a service call, and a damaged first impression — all for an error that takes 30 seconds to correct if it is caught before disbursement.