Content authored by practitioners with experience at Amazon India, Intuit QuickBooks, and the Tata Group. Meet the team →
Lenders compare bank statement analysers on accuracy before checking definitions, so a tool that reads every row correctly can still produce ABB, FOIR and income figures that do not match the credit policy.
Take five statements with trusted manual calculations, ask the vendor to show each metric with the rows used, compare each rule with policy, and record the rupee difference and whether the rule is configurable.
A definitions table covering average balance method, obligations source, income rule, cash deposit treatment and own-account transfers, with policy rule, vendor rule, rows included and configurability.
A list of metrics that match policy, metrics that can be configured to match, and metrics with a rupee gap the credit head must accept or reject before accuracy testing begins.
Two analysers can read the same bank statement perfectly and still report different average balance, obligations and income, because they define those metrics differently. Before you compare accuracy, test each definition against your own credit policy, with the vendor showing exactly which rows went into each number. A tool that misreads nothing but counts the wrong rows will still move loans across your cut-off.
Background on what these metrics tell an underwriter is in how to read a bank statement for credit risk and bank statement analysis accuracy signals.
Which definitions differ most?
FOIR (fixed obligations to income ratio), ABB (average bank balance) and income rules are not set by RBI; each lender fixes them in its board-approved credit policy. Vendors pick defaults, and the defaults vary on five points:
- Average balance: daily end-of-day average across the period, or sampled dates such as the 5th, 15th and 25th of each month. Some policies also exclude the last 3 days before a salary credit.
- Obligations: only EMIs the tool detects in the statement, declared obligations from the application, or bureau obligations added from outside the statement.
- Income: salary credits only, all credits, or all credits less one-offs such as loan disbursals, refunds and maturity proceeds.
- Cash deposits: counted as income, counted with a haircut, or excluded.
- Own-account transfers: money moved between the borrower’s own accounts, which inflates both credits and debits if not removed.
Ask the vendor for the written rule for each, not a verbal description in a demo.
How do you run the test?
- Choose five statements with a manual calculation you trust: at least one salaried borrower, one self-employed borrower and one statement with cash deposits and own-account transfers.
- Give the vendor the statements only, not your figures.
- Ask for each metric with the rows used, so you can see what was included and excluded.
- Compare each rule with your policy, and record the rupee difference on each file.
- Ask whether each definition can be set to your policy, and whether the setting applies to every file or must be repeated.
| Metric | Your policy rule | Vendor rule | Rows included or excluded | Difference on test file | Configurable |
|---|---|---|---|---|---|
| Average balance | Sampled 5th, 15th, 25th | Daily average | All days vs 18 dates over 6 months | Record in ₹ | Yes or no |
| Obligations | Detected EMIs plus declared | Detected EMIs only | Declared loan without statement EMI missing | Record in ₹ | Yes or no |
| Income | Salary credits only | All credits | Refunds and transfers included | Record in ₹ | Yes or no |
| Cash deposits | 50 percent haircut | Full value | Cash credits | Record in ₹ | Yes or no |
| Own-account transfers | Excluded | Included | Transfers from own savings account | Record in ₹ | Yes or no |
As a buyer’s test, the table is a pass when every row is either a match or configurable to match, and a fail when an unconfigurable difference moves a decision across your cut-off.
Worked example: when FOIR flips a decision
Illustrative figures. A salaried applicant has net salary credits of ₹62,000 a month. The statement shows EMIs of ₹18,400. The application declares a further personal loan with an EMI of ₹5,500, paid from another account, so actual obligations are ₹23,900.
- Tool rule, detected EMIs only: ₹18,400 ÷ ₹62,000 = about 30 percent.
- Policy rule, detected plus declared: ₹23,900 ÷ ₹62,000 = about 39 percent.
At an illustrative policy cut-off of 35 percent, the first figure approves the loan and the second declines it. Neither tool misread a row; the definition decided the outcome. This cut-off is for illustration only and is not a recommendation.
The same check applies to income. If the tool counted a one-off ₹40,000 refund as income in a six-month window, average monthly income rises by about ₹6,667, which also lowers FOIR.
What should happen after the test?
Keep the completed table in the vendor file. Once definitions are aligned, move to an accuracy and workflow pilot on closed cases, as set out in the two-week pilot plan. Under the Reserve Bank of India (Digital Lending) Directions, 2025, issued on 8 May 2025, the lender stays responsible for credit assessment done through a service provider, so the definitions in use should be documented and approved.
How Terra Insight fits
TransactIQ is a managed bank statement analysis service on Terra Insight’s AWS Mumbai infrastructure. It produces 40+ engineered credit signals, including salary consistency and EMI obligation aggregation across lenders, and every signal is rule-traceable so the credit team can see why a figure is what it is. Evaluating teams can request a representative-sample fit benchmark. See the bank statement analyzer page.