Content authored by practitioners with experience at Amazon India, Intuit QuickBooks, and the Tata Group. Meet the team →
NBFC business cases for bank statement analysis often rest on vendor benchmarks or loss estimates that a credit committee cannot verify, so approval stalls or the case is rejected.
Measure analyst time on 20 real files split by task, scale it by monthly volume, test it at 30 percent lower, and list error and fraud-loss categories separately with counts from your own records.
A template with line, source, measured or estimated, owner and review date; one total for analyst hours released; a separate loss-category list with rupee values only where traced to past cases.
An approval note with a defensible analyst-time figure, a sensitivity check and a named list of loss categories the committee can verify line by line.
A business case for bank statement analysis at an NBFC should start with time you can measure: how many files you read each month, how many minutes each one takes today and what share needs rework. Error and fraud-loss categories come second, named and counted from your own records. Use your own numbers throughout; no industry benchmark will survive a credit committee’s first question.
For background, see bank statement analysis use cases for NBFCs and the seven-dimension comparison in manual vs automated bank statement review.
How do you measure analyst time?
Time-stamp 20 real files end to end over one week. Split each file into five tasks:
- Intake: receiving the statement, unlocking passwords, checking the period is complete.
- Reading: going through every page, including scanned or multi-account files.
- Categorising: marking salary, business credits, EMIs, cash deposits and own-account transfers.
- Flagging: noting bounces, unusual credits and signs of alteration.
- Memo writing: computing average balance, obligations and income for the credit memo.
Record rework separately: files sent back because a page was missing, the format would not open or the period was short. Measure an average week and a peak week, such as the end of a quarter, because staffing is set by the peak.
Which error and fraud-loss categories belong in the case?
List them by name and count past cases from your own collections and fraud records:
- Misread income: salary or business credits overstated, for example one-off credits counted as income.
- Missed obligations: EMIs to other lenders not caught, which understates FOIR.
- Altered statements: balances or entries edited after the file left the bank.
- Undisclosed accounts: transfers to accounts the borrower did not declare.
- Circular credits: money moved in and out to inflate turnover.
Leave the rupee column blank where you cannot trace the loss to a case. Where you can, record the traced amount, for example a ₹4,80,000 write-off on a loan later found to rest on an altered statement. NBFCs already report frauds under the RBI Master Directions on Fraud Risk Management dated 15 July 2024, so those records are a starting point.
What does the template look like?
| Line | Source | Measured or estimated | Owner | Review date |
|---|---|---|---|---|
| Files per month | Loan origination system | Measured | Credit operations | Monthly |
| Minutes per file, by task | 20-file time study | Measured | Credit analyst lead | Quarterly |
| Share of files needing rework | Rework log | Measured | Credit operations | Monthly |
| Peak-week multiplier | Quarter-end volumes | Measured | Credit operations | Quarterly |
| Analyst hours released | Calculated | Calculated | Finance | Quarterly |
| Loss categories with case counts | Collections and fraud records | Counted | Risk | Half-yearly |
Keep one total, analyst hours released, and keep the loss categories as a separate list. Mixing a measured time figure with an estimated loss figure weakens both.
To check a vendor before you rely on the time line, use the same evaluation discipline as the pilot:
| Test | What to give the vendor | What a pass looks like |
|---|---|---|
| Time per file | 20 timed files from your study | Tool path, including analyst review, is at least 25% below your manual time |
| Rework | Files your team sent back for format | At least 90% open without manual handling |
| Obligations | Files with known EMIs | Missed-EMI cases from your list are caught |
Worked example: 1,200 files a month
Illustrative inputs. An NBFC reads 1,200 statements a month and the time study shows 35 minutes per file.
- Manual time: 1,200 × 35 = 42,000 minutes = 700 analyst hours a month.
- If review of the tool output takes 10 minutes per file: 1,200 × 10 = 12,000 minutes = 200 hours.
- Hours released: 700 − 200 = 500 hours a month.
Now assume your 35-minute estimate is 30 percent too high, so the true figure is 24.5 minutes. Manual time becomes 1,200 × 24.5 = 29,400 minutes = 490 hours, and hours released become 490 − 200 = 290 hours a month. At an illustrative loaded analyst cost of ₹600 an hour, 500 hours is ₹3,00,000 a month and 290 hours is ₹1,74,000 a month; use your own cost figure from payroll. If the case still holds at 290 hours, it does not depend on a precise estimate. If it only works at 500, the time study needs more files.
Note that the review time of 10 minutes is itself an input to measure in a pilot, not a vendor promise.
What about building in-house?
Some NBFCs consider building their own parser. The BSA build-vs-buy tool lays out the cost lines for that option, including keeping parsers current as bank formats change. Put the build option in the same template so the committee compares like with like.
How Terra Insight fits
TransactIQ is a managed bank statement analysis service on Terra Insight’s AWS Mumbai infrastructure, producing 40+ engineered credit signals, including EMI obligation aggregation across lenders and round-tripping detection, and reading degraded PSU and co-operative bank formats. Pricing is discussed in discovery, and evaluating lenders can request a representative-sample fit benchmark to measure the review-time input on their own files. See NBFC lending.