Reducing MSME Loan Underwriting TAT From Days to Hours: An India Playbook

Most of an MSME file's calendar is queue, not analysis. On a ₹1.5 crore secured file, 33.5 working days becomes 10.5 — a 3x faster decision turnaround — by collecting statements through Account Aggregator, drafting the CAM in about 30 minutes, and fixing committee scheduling. Analysis moves to hours; the file does not.

Key facts

• RBI puts a clock on small tickets only. "Timeline for credit decisions for loans up to ₹25 lakh to MSE borrowers shall not be more than 14 working days," and above that "timelines shall be as per the Board approved sanction time norms" (RBI, *Frequently Asked Questions — Micro, Small and Medium Enterprises*, Q.23, updated 29 July 2025). Above ₹25 lakh, your own board norm is the only clock — and in most lenders nobody reports against it.
• Every application must be acknowledged and trackable. The same FAQ requires that "banks shall ensure that the acknowledgement and status of the application is sent automatically to the applicants", with "a unique application serial number for both physical and online applications" (Q.24). A Credit Proposal Tracking System is not optional, and it is also the cheapest TAT instrument you own, because it timestamps every stage.
• The statement-collection problem is solved at national scale. Sahamati's ecosystem dashboard reports 538.32 million cumulative consents fulfilled and 326.30 million accounts linked as at July 2026 (Sahamati, AA ecosystem dashboard). Bank statements no longer need to arrive as email attachments.
• GST data comes down the same pipe. RBI included GSTN as a Financial Information Provider "with a view to facilitate cash flow-based lending to MSMEs" (RBI/2022-23/140, *Inclusion of Goods and Service Tax Network (GSTN) as a Financial Information Provider under Account Aggregator Framework*, 23 November 2022).
• YuSight produces an end-to-end CAM draft in about 30 minutes, every figure traced to its source document and page. On the worked file below that removes four working days from the critical path and, more usefully, removes the source-hunting argument from the committee meeting.

Where do the days actually go in an MSME file?

Take one unit of work and hold it constant for the rest of this playbook: a ₹1.5 crore secured facility — ₹1 crore cash credit plus a ₹50 lakh term loan — to a private limited manufacturer with two banking relationships, three years of audited financials, property collateral, and two directors as guarantors. Not a ₹10 lakh Mudra file, not a ₹15 crore consortium proposal.

The stage-by-stage process is set out in the MSME loan underwriting process in India. This page is about the clock, so the column that matters is who owns it. There are only three owners, and confusing them is why most TAT programmes fail.

Now split that 33.5 days by owner:

Borrower-owned : 9.0 days = 26.9% Third-party-owned : 5.0 days = 14.9% (2.5 of which sits on the critical path) Lender-owned : 19.5 days = 58.2% of which analysis and drafting (stages 3-7, 9) : 11.5 days = 34.3% of which committee and documentation (10, 11) : 8.0 days = 23.9%

Thirty-four per cent of the calendar is the analysis block. That is the part software touches. It is a large share and worth attacking — but it also means that if you automated analysis to zero you would still have a 22-day file. Any vendor promising you "hours" end-to-end has not looked at the other 66%.

Why is document collection the biggest single block — and why does nobody automate it?

Nine days, borrower-owned, and it is the stage that receives the least engineering attention in every MSME lending programme we have seen. The reasons it takes nine days are mundane and completely fixable:

• The checklist is issued incrementally. The RM asks for financials, then remembers the GST credentials, then discovers a second bank account on the statement and asks for that too. Each round trip is two to three days.
• The wrong entity's documents arrive. A proprietor sends personal ITRs when the facility is to the firm; a director sends the holding company's audited accounts.
• Documents arrive unsigned, unstamped, or as photographs of a laptop screen.
• Nobody tells the borrower what "12 months of bank statements across all banking relationships" means, so they send the operating account and forget the OD account at another bank.

What compresses it:

Issue a complete, entity-mapped checklist at lead stage, generated from product, amount, constitution and security type — not typed by the RM. This is a rules problem, not an AI problem, and it is the highest-return hour of engineering in the whole programme.

Move everything that can be fetched to a fetch. Bank statements and GST returns should not be documents the borrower sends. They are data you pull with consent. More on this below.

Block the file, visibly. A CPTS that shows the borrower exactly which four items are outstanding, updated live, converts an eight-day chase into a two-day one because the borrower stops waiting to be told.

What does not compress it: better OCR. If the document has not arrived, extraction accuracy is irrelevant. This is the single most common misallocation of a TAT budget.

Realistic after: 3.0 working days for the residue that genuinely must come from the borrower — audited financial statements, title deeds, board resolutions, KYC and photographs, the stock and book-debt statement.

Which India-specific accelerators actually compress the clock?

India has better underwriting rails than almost any market. They are also routinely oversold. Here is each one against the stage it hits, with the honest limit.

Two notes on Account Aggregator specifically, because it carries most of the load here.

First, the framework is a consent conduit, not a data source. The RBI Master Direction is explicit: "No financial information of the customer accessed by the Account Aggregator from the financial information providers shall reside with the Account Aggregator" (RBI, *Master Direction — Non-Banking Financial Company — Account Aggregator (Reserve Bank) Directions, 2016*, RBI/DNBR/2016-17/46, updated 6 September 2024). You still need an analyzer on the other end. The consent mechanics, purpose codes and failure modes are covered in the Account Aggregator framework explained.

Second, AA does not remove the analysis, it removes the waiting. A twelve-month statement that arrives in ninety seconds still has to be read for cheque returns, circular transactions, related-party flows and undisclosed borrowings. It arrives as clean structured data rather than a 340-page scanned PDF, which is why the analysis then takes minutes instead of days — but the reading is not optional.

And if your MSME originations flow through a digital lending app or an LSP, the collection design also has to satisfy the RBI (Digital Lending) Directions, 2025 (RBI/2025-26/36, 8 May 2025), which require the Key Fact Statement to "automatically flow to the borrower on the registered and verified email/SMS upon execution" and disbursement "always be made into the bank account of the borrower". The working checklist is in RBI digital lending guidelines for credit and compliance teams.

There is also the Unified Lending Interface, which aims to give lenders consent-based access to land records, satellite data, Udyam and other datasets through a common API layer. It is genuinely relevant to agri and small-ticket MSME TAT. We have deliberately not put a number against it here.

What does the RBI 14-working-day expectation actually require?

It requires a credit decision within 14 working days on MSE loans up to ₹25 lakh — not a disbursement, and not a documented facility. Three practical consequences that credit heads consistently get wrong:

1. The clock starts at a complete application, not at first contact. Which makes your definition of "complete" a compliance artefact. If your CPTS marks completeness generously, you will report compliance you do not have.
2. Above ₹25 lakh there is no regulatory clock, only your board-approved norm. Write one. A board norm of "30 days for secured facilities up to ₹5 crore" is a defensible commitment and it is the only thing that will get committee scheduling fixed, because it makes a deferral someone's problem.
3. The collateral-free floor changed this year. Banks must not take collateral on MSE loans up to ₹20 lakh, extendable to ₹25 lakh on track record, for loans sanctioned or renewed on or after 1 April 2026 (RBI, *Lending to Micro, Small & Medium Enterprises (MSME) Sector (Amendment) Directions, 2026*, RBI/2025-26/206, 9 February 2026). For TAT this matters enormously: below ₹20 lakh there is no legal search, no valuation, no mortgage, no e-stamping of security documents. The entire third-party block disappears, which is exactly why that segment can genuinely reach hours and the ₹1.5 crore file cannot.

Worked example: the ₹1.5 crore file, before and after

Same borrower, same policy, same credit standards. Four changes: consent-based collection, machine analysis with cited output, a 30-minute CAM draft, and a fixed committee calendar.

Now the critical path, because the stages do not simply add:

Day 0.0 → 0.5 Checklist, entity verification, AA consent Day 0.5 → 3.5 Borrower document collection (3.0 days) Day 1.0 → 6.0 Legal search + valuation, running in parallel (5.0 days) Day 3.5 → 4.5 Analysis and CAM draft (1.0 day) Day 6.0 → 8.0 Committee (waits for the valuation report, not for the CAM) Day 8.0 → 10.5 Sanction, e-sign, e-stamp, charge creation Before : 33.5 working days After : 10.5 working days Saving : 23.0 working days ≈ 4.6 calendar weeks Speed-up: 33.5 ÷ 10.5 = 3.19x → 3x faster decision turnaround

And now the honest part. In the after case, analysis is 1.0 of 10.5 days — 9.5% of the clock, down from 34.3%. Of what remains:

• 3.0 days borrower
• 2.5 days of third-party legal and valuation that sits on the critical path
• 2.0 days committee
• 2.5 days documentation

None of that is an AI problem. The file reached 10.5 days because the analysis block collapsed and because someone fixed the committee calendar. Do only the first and you get roughly 22 days, which is a 1.5x improvement and a disappointed sponsor.

Where the "hours" actually is

Inside that 1.0-day analysis block, here is the touch time:

340 minutes = 5 hours 40 minutes of touch time Before, the same block was 11.5 working days of elapsed time

That is the days-to-hours claim, stated precisely: the lender-controlled analysis and drafting block goes from eleven and a half days to under six hours. The file goes from 33.5 days to 10.5. Both are true and only one of them is the sentence a vendor will put on a slide.

Why is committee scheduling not a technology problem?

Five days in the before case, and it is almost never five days of thinking. It is:

• The committee sits fortnightly, so a file that becomes ready on day 2 of the cycle waits eight working days.
• Papers circulate 24 hours ahead, so members read them in the meeting.
• One member asks for the GST reconciliation that is in annexure 4 of the CAM, and the file is deferred rather than the annexure being opened.
• Sanctioning authority is set by amount alone, so a ₹1.5 crore renewal with unchanged security goes to the same forum as a ₹1.5 crore greenfield exposure.

Four fixes, none of which involve software:

1. Sit weekly or twice weekly. Halving the cycle length halves the average wait, mechanically.
2. Circulate 48 hours ahead and require queries in writing beforehand. Deferrals collapse when questions arrive before the meeting.
3. Delegate by risk, not only by amount. Renewals with no material change, unchanged security and satisfactory conduct should not consume committee time.
4. Record the deferral reason in the CPTS. Within a quarter you will find that four reasons cause most of the deferrals, and three of them are checklist defects.

Where technology helps is narrower than it sounds but real: when every figure in the CAM links to the source page, the committee argues about risk rather than about provenance. In our experience the second kind of argument is what causes deferrals. The format Indian committees expect is in what a credit appraisal memorandum is.

What TAT is realistic, by segment?

Do not commit to one number across the book. The binding constraint changes by segment.

The pattern is consistent: TAT falls to hours exactly where no third party sits on the critical path. Remove legal, valuation and registration and the file becomes a data problem. Leave them in and you are optimising around a five-day floor you do not control.

How do you actually run this? A 90-day sequence

1. Instrument the clock (days 1–15). Turn on stage timestamps in your CPTS or LOS. Pull the last 100 sanctioned MSME files. Compute median elapsed days per stage, banded by the four segments above. Nothing else in this list is worth doing without this.
2. Fix the checklist (days 10–25). Generate the document list from product, amount, constitution and security type. Publish it to the borrower at lead stage with a live outstanding-items view. Measure stage 2 again after four weeks.
3. Turn on consent-based collection (days 20–50). Register as an FIU, integrate one AA, and route bank statements and GST returns through it. Keep the PDF upload path alive for the coverage gaps; do not make AA mandatory in month one.
4. Automate the analysis block (days 30–70). Document classification, entity mapping, bank statement and GST analysis, spreading and ratio computation, with every figure cited to its source page. Route exceptions to analysts, not documents — the design is in human-in-the-loop credit review.
5. Move the CAM from typing to editing (days 45–80). Generate the draft with citations; the analyst writes the judgement sections and edits the rest. Target: the analyst never rekeys a number that already exists in a spread or a bureau report.
6. Rebuild the committee calendar (days 50–75). Weekly or twice-weekly slot, 48-hour circulation, written pre-queries, delegated authority for unchanged renewals, deferral reasons coded in the CPTS.
7. Close the back end (days 60–90). eSign for sanction acceptance and standard documentation, e-stamping where the state supports it, and CERSAI or ROC filings prepared in parallel with committee rather than after it.
8. Re-measure and publish (day 90). The same 100-file analysis, same segments, same stage definitions. Publish the before and after by stage — including the stages that did not move.

What should you measure, and where do the numbers lie?

Baseline before you start. One hundred files, timed by stage, medians by segment. Two weeks of work, and it is the only thing that will make your post-implementation numbers survive a challenge at ALCO.

FAQ

How can an NBFC cut MSME underwriting TAT?

Start by measuring where the days actually go, then attack the two biggest blocks: borrower document collection and CAM drafting. Route bank statements and GST returns through Account Aggregator so they arrive as data, generate the CAM with citations instead of retyping numbers, and fix the committee calendar. On our worked ₹1.5 crore file that takes 33.5 working days to 10.5.

Where does the time actually go in an MSME file?

On a ₹1.5 crore secured file, roughly 27% is the borrower collecting documents, 15% is legal search and valuation, 34% is analysis and CAM drafting, and 24% is committee scheduling and documentation. Actual credit judgement is a small fraction of the calendar. Most of the clock is queue.

What TAT is realistic for a secured MSME loan?

Ten to fourteen working days for a ₹20 lakh to ₹2 crore file with property collateral, once collection is consent-based and the analysis block is automated. Not hours — legal search and valuation are a five-day third-party floor you do not control. Unsecured files below ₹20 lakh are the segment where same-day is genuinely achievable.

Does RBI require MSME loans to be decided in 14 working days?

For MSE loans up to ₹25 lakh, yes — RBI's MSME FAQs state the credit decision timeline shall not exceed 14 working days. Above that, the timeline is whatever your board-approved sanction norms say. If you have never written one, that is worth doing, because an unwritten norm is never missed.

Does Account Aggregator remove the need for bank statement analysis?

No. It removes the wait and the format problem, not the reading. You still have to look for cheque returns, circular transactions, related-party flows and undisclosed borrowings. What changes is that the statement arrives as structured data in seconds rather than as a scanned PDF in four days.

Can GST returns be pulled through Account Aggregator?

Yes. RBI included GSTN as a Financial Information Provider in November 2022 specifically to support cash-flow-based MSME lending, so GSTR-1 and GSTR-3B data can be fetched under the same consent artefact as bank statements. The reconciliation judgement between the two is unchanged.

Why does committee scheduling take so long, and can software fix it?

It takes long because committees sit fortnightly, papers circulate late, and files get deferred over questions that could have been asked in writing. Software does not fix any of that. What it does fix is the provenance argument — when every figure links to its source page, the committee spends its time on risk instead of on where a number came from.

Will automation get us to same-day sanction?

For clean unsecured micro files with AA-fetched statements and no security creation, plausibly. For a secured facility, no — the legal search and valuation floor alone is about five working days. Be precise about which segment your target applies to before you promise it to a board.

What should we automate first to cut TAT?

The document checklist, then consent-based collection, then the analysis block. The checklist is a rules problem you can fix in a fortnight and it attacks the largest single block. Automating extraction before the documents arrive faster is the most common way to spend a TAT budget and see nothing move.

How do we prove the TAT improvement was real?

Baseline one hundred recent files with stage-level timestamps, banded by segment, before you start. Re-run the identical analysis at day 90. Publish the stages that did not improve alongside the ones that did — a report that only shows the wins will not survive its first challenge.

Key takeaways

• On a ₹1.5 crore secured MSME file, 33.5 working days becomes 10.5 — a 3x faster decision turnaround — but only if you fix collection, analysis, committee and documentation together.
• Analysis is 34% of the clock before automation and 9.5% after. Automate it alone and you get roughly 1.5x, not 3x.
• Document collection is the biggest single block and the least automated. The fix is a complete entity-mapped checklist at lead stage plus consent-based fetching — not better OCR.
• Account Aggregator and GSTN-as-FIP remove the waiting, not the reading. Coverage gaps mean you keep the upload path alive.
• eSign and e-stamping compress the back end; legal search and valuation do not compress at all. That five-day third-party floor is why "hours" is honest only for unsecured small-ticket files.
• Committee scheduling and borrower responsiveness are process problems. No platform fixes a fortnightly calendar.
• Measure by stage, band by segment, and baseline before you start.

See your first CAM in 30 minutes — [book a live demo](https://yuverse.ai/yusight).