Reports Funnel diagnostic Pipeline hours Operations diagnostic Fix register

Where the funnel leaks — loan-level diagnostic of origination, April 2025 to August 2026

80,642 loans initiated in the 17-month comparable window, followed to their status on 14 Sep 2026. Source: Metabase loan dump (130,274 loans, all stage timestamps and actors). Sections 1–10 use this window because the current sourcing mix did not exist earlier; Section 11 extends the core metrics back to FY22. Rupee figures use the dump's loan_amount, which reconciles to the sanctioned amount in the Loan Detail Report (₹1,294 Cr vs ₹1,293 Cr for the same loans); actual disbursed on those loans is ₹1,201 Cr (93%). Cumulative disbursement since inception is ₹2,119 Cr.

80,642Initiated

37,076Logged in (46%)

17,049Sanctioned (46% of logins)

10,313Disbursed (60% of sanctions; ₹1,275 Cr sanctioned, ₹1,201 Cr paid out)

12.8%Initiated → disbursed

35 dMedian initiated → disbursed

The diagnosis — seven findings, in order of money

₹534 Cr of sanctions died after sanction. Of 14,329 sanctions old enough to judge (Apr 25–Jun 26), 36% were cancelled or rejected after the sanction letter — 5,205 loans. AP loses 63%, DNCR 42%, Rajasthan 39%; Tamil Nadu and Karnataka lose ~20%. Almost none show any activity after the letter: the sanction is issued before the customer has committed. Bringing every region to 22% adds ~2,050 disbursals, roughly ₹200 Cr over the window — ₹12 Cr a month. The largest single lever, and it sits with sales.
Nobody owns the wait. Files now spend 18% of the initiated → disbursed journey standing still on a query, up from 1% in FY22 — and that time is deducted from no one's turnaround: not the desk's, not U1's, not the RM's. Read one by one, 38% of queries and 39% of standing days are not requests for information — approvals travelling as queries, "call me" and "as discussed", system corrections, fee reminders, and 8,965 documents asked for a second time after QC had accepted them. Two credit gates, Biz UW and U3, agree 99% with the step before them and hold files for a day each. Sanction issues before commitment. Every one of these is the same design choice: a stage that can stop a file without owning the stop.
U2 is a queue, and half its length is queries. Login → sanction went from 7.0 days (Q2 FY26) to 15.7 (Q2 FY27); the increase is at Biz UW → U2, 0.9 → 6.2 days. Twelve regional credit heads carry 400–900 files each per half-year; the same people cleared U2 in a day for four years. Salaried is 44% of disbursed loans, the central CPU runs 15 days end to end against 31, and it sees 5–7% of files. Capacity and routing recover half the lost week; removing the avoidable queries recovers the other half at no cost.
41% of initiations were never loans. Nine digital feeds and two bulk lead-initiators produced 32,825 initiations and under 130 disbursals. They halve the reported login rate and make every funnel report since mid-2025 unreadable; the auto-cancel that closes them already exists. On the real book, duplicates are 10% (4,861 files, 92% same RM, 57% same day, 1,416 re-initiated within 90 days of a rejection) — a same-day block and a cooling period, not a dedupe programme.
The digital rails exist and are run backwards. e-KYC, PAN verification, crime check, Account Aggregator, penny-drop, e-Sign and e-NACH are all wired into the LOS. Only e-KYC is used (93%), and it cuts KYC queries by two-thirds where it is; the rest sit at 0–13% and are invoked after the desk has already asked — files with more rails completed carry more ADR rounds, not fewer. The seven LOS rules in Appendix B are sequencing changes and mandatory fields, not builds.
Sales capacity is spread too thin, and the first thing to hold it to is the answer time. 450 of 1,124 RMs (40%) produce fewer than one disbursal every two months — 4% of value; the top quartile delivers 65%. The RM's response to a query is 4 hours median, a day at p75, three days at p90, and it is the one RM number that moves the funnel this quarter. Genuine-customer conversion has been 24–28% every year since FY22: growth has been volume, and 942 of 1,678 RMs seen in the window have already gone quiet.
The first sanction is not the offer, and two products should not be on the menu. 46% of sanctions are re-issued before disbursal (MMR, Gujarat, Karnataka, DNCR above half); re-sanctioned files disburse at 73% against 50%, so the second letter is where the customer commits. Plot Purchase: 1,715 initiations, one disbursal. Home Improvement 11% sanction. APF is the fastest lane through credit (9 days) and the weakest book — ₹5.8L tickets at 12.0% with 2.1% NPA against 0.2% — and needs repricing before it scales.

Old vs new (Section 11, normalised ex-feed): the sanction leak, the re-sanction loop and real-customer conversion have looked the same every year since FY22 — growth has been volume, not conversion. The U2 queue, the feed and the query load arrived with FY26 volume. What is working: sanction → disbursal fell from 27 to 7 days over the same period, and real DSA partners (Urban Money, Andromeda, Ruloans, Basic, TENB) convert at 20% initiated → disbursed, ahead of direct at 18%.

Value at stake

Money leaves in four places: sanctions that die after the letter (₹534 Cr at sanction value in seventeen months), time (login → sanction doubled to 15.7 days — half of it a queue at U2, half of it files standing on queries that should not exist), files that should never have reached credit (a third of rejects are bureau facts; two-thirds of desk queries are documents the RM had), and phantom volume (41% of initiations were never loans). Ranked by rupees: the sanction leak, then the wait, then credit capacity, then data hygiene.

The structural question

The leaks share one design choice: the system has no concept of whose clock is running. Query time is deducted from no one's turnaround; two credit gates hold files without deciding anything; approvals travel as queries; a sanction issues before the customer has committed; a lead becomes a loan before a customer exists. Each is a stage that can stop a file without owning the stop. Until the wait has an owner, adding capacity at U2 staffs a queue that the ledger keeps refilling.

What moves first

Two things start Monday with people already in place and no development: the commitment gate at sanction with a named-file call in the four leaking regions, and the KPI change that puts query time on the raiser's clock and moves approvals out of the ledger. U2 capacity and salaried routing follow, smaller than first sized because the queue shrinks when a third of the queries stop. The system work is switching on rails that exist. From October, every number is reported on the deduped, ex-feed base, with the old figures alongside for one quarter.

How to read this document

Numbers. Every duration is a median unless marked — half the files are faster, half slower, and one very slow file cannot move it. Where a table gives p75 or p90, that is the value the slowest quarter or slowest tenth of files exceed: the tail, which is where customers are lost. Percentages of a stage are conversion; "case-days" are files × days above standard, so a two-file branch cannot dominate.

Colour. A coloured cell is a rating — green at or better than standard, amber near it, red beyond — and the thresholds are printed under each table. Colour is on the number itself, never in a separate column. Cohorts. Conversion is measured by following the files that entered a stage in a period through to today, not by dividing this month's sanctions by this month's logins.

Terms. Any dotted-underlined term (like this) shows its meaning on hover or tap; the full glossary is at the end. Reading paths. Fifteen minutes: the diagnosis, Section 0 and the ninety-day plan. One hour: add Sections 1–5, 13 and 16. Everything else is evidence and is there to be checked.

0 The pipeline — every stage the data records, and where customers leave27% of real customers become loans; three drops — U1 rejects, files never made into applications, sanctions that die — carry 70% of the loss, and none of them is the login desk.

Unique customers (PAN, else Aadhaar, else mobile — from the All-Customers export), real files only — 80,642 initiations become 47,787 files once the nine aggregator feeds and the two bulk lead-initiators are removed, and 42,926 customers once duplicates are collapsed on PAN, Aadhaar or mobile from the customer master. Each customer is credited with the furthest stage any of their files reached.

Initiations in LOS

80,642

−32,825 never a loan (feed / bulk lead) · −4,861 duplicate files of the same customer (10%)

Initiated

42,926

100% of customers

−8,101 (19% of the stage)

Application done

34,825

81% of customers

−1,946 (6% of the stage)

32,879

77% of customers

−3,781 (11% of the stage)

U1 decision

29,098

68% of customers

−9,545 U1 rejects (Biz UW: −0 real)

Biz UW

27,582

64% of customers

U2 decision

17,409

41% of customers

−1,724 (10% of the stage)

Sanctioned

15,685

37% of customers

−5,782 (37% of the stage)

Ops done (docs)

9,903

23% of customers

−317 (3% of the stage)

Disbursed

9,586

22% of customers

Drop	Customers lost	Where they sit now	Largest reason in the data	Fix	Owner
Initiated → Application	8,101	Soft Cancelled 81%	RM opened a file and never completed the application — customer not pursued or data never keyed	Auto-cancel initiations idle 15 days; mandatory fields before a file exists (Appendix B, Rules 6 and 4)	Prerak Mehta
Application → Login	1,946	Soft Cancelled 55% · ADR before sanction 22%	ADR loop with the 16-person login desk: co-applicant KYC, bank statements, income proof, login fee — 2.6 rounds per file, 70% of 175k query lines are known-missing documents	Document gate before the file reaches the desk; structured ADR codes (Appendix B, Rules 4 and 7)	Prerak Mehta · Vishal Valecha
Login → U1 decision	3,786	Soft Cancelled 69% · Soft Rejected 17%	File logged, then died in the U1 queue without a decision — median login → U1 is 2.7 days, p90 10.7	U1 SLA same working day; BCM queue-ownership rule; publish per-BCM ageing	Rohan Shah · zonal credit
U1 → U2 (U1 rejects; Biz UW adds nothing)	9,545	Rejected · 9,545 customers	Bureau / repayment history 32% of all reject codes, eligibility / FOIR 22% — knowable before login. Biz UW agrees with U1 on 99% of files	Bureau pull + hard-rule engine + FOIR at initiation (Appendix B, Rule 3); make Biz UW non-blocking or remove it	Rohan Shah · Prerak Mehta
U2 → Sanction	1,724	Rejected after U2 · median 13.5 days post-login	Property (technical/legal) is 36% of late rejects and is checked last — 82% of property rejects come after U2 at 19 days; U1/U2 overturn 14% (3% South, 22% MMR)	Technical initiated at login in parallel with U1 (Appendix B, Rule 5); monthly overturn calibration in MMR, MP, UP, Vidarbha, DNCR	Rohan Shah
Sanction → Ops done	5,784	Soft Cancelled 68% · still Sanctioned 17%	Sanction letter issued before the customer committed — AP loses 63%, DNCR 42%; 46% of sanctions get re-sanctioned because the first offer was not the real one	No sanction without processing fee or Accept-offer reply; right-first-time sanction amount; weekly named-file SUD call on anything older than 15 days	Vinayak Deousker · zone heads
Ops done → Disbursed	318	Soft Cancelled 44% · Sanctioned 41%	Negligible — ops has cut sanction → disbursal from 27 to 7 days	Hold; watch month-end bunching (28% of disbursals in last 5 days)	Vishal Valecha

Three drops carry 70% of all loss: U1 rejects (9,545 — mostly bureau facts knowable at initiation), the pre-application die-off (8,101 — files opened and never made into applications) and post-sanction cancellation (5,784 — sanctions issued before the customer committed). Login and ops, the two stages everyone watches, lose 1,946 and 318 customers respectively; login's cost is time, not conversion. End to end, 22% of real customers disburse — against 12% on the raw initiation count.

1 The funnel by cohortReported login rate fell 64% → 34% because of the feed; the real operation slipped 78% → 65% — two problems, two owners.

Each row is the loans initiated in that quarter and where they stand today. The last two quarters are still maturing at the disbursal end.

Initiated quarter	Initiated	of which lead-feed	Login % (all)	Login % ex lead-feed	Sanction % of logins	Disbursed % of sanctions	Initiated → disbursed
2025Q2	9,807	3,026	51%	73%	45%	52%	12.2%
2025Q3	12,188	4,733	49%	80%	45%	65%	14.6%
2025Q4	16,153	4,458	51%	70%	51%	57%	14.8%
2026Q1	19,748	9,316	38%	71%	45%	71%	12.0%
2026Q2	13,011	4,334	44%	65%	43%	70%	13.2%
2026Q3	9,735	2,836	47%	65%	45%	41%	8.7%

Reported login % collapsed from 64% to 34% in Q4 FY26 — but the ex-feed column shows the real operation moved 78% → 65%. Both stories are true and need different owners: the feed problem is a tech/product decision, the 13-point real decline is sales discipline at the login desk.

2 Where time is lost — median days per stageOps cut sanction → disbursal from 27 to 7 days; credit went the other way as volume rose, entirely at U2.

Standards used for colour: initiated → login 1 day, login → U1 2 days, Biz UW → U2 2 days, login → sanction 8 days, sanction → disbursal 10 days.

Login quarter	Initiated → login	Login → U1	U1 → Biz UW	Biz UW → U2	U2 → sanction	Login → sanction	Sanction → disbursed
2025Q2	1.1	2.3	0.8	0.9	0.1	7.0	27.3
2025Q3	1.1	2.6	0.9	1.9	0.0	9.4	17.2
2025Q4	1.0	2.9	1.0	3.2	0.0	13.3	15.1
2026Q1	1.3	2.9	0.9	5.2	0.1	12.9	10.1
2026Q2	2.7	2.8	1.8	6.2	0.0	15.7	10.7
2026Q3	3.3	2.2	1.0	4.9	0.1	11.0	7.2

Measured from the event log

Milestone gap — event log, real files, Aug 25–Sep 26	Median d	p75	p90	Files
Login → QC	0.2	1.2	4.0	30,163
QC → U1	3.0	6.2	14.0	18,913
U1 → Biz UW	0.9	3.0	8.2	17,384
Biz UW → U2	4.1	9.9	20.3	13,528
U2 → sanction	0.0	0.6	1.7	13,206
Login → sanction	14.0	23.8	39.9	13,745
Sanction → docket printed	7.9	19.1	38.1	8,716
Disb auth → OTC done	0.7	3.8	12.0	8,412
OTC → OTC compliance done	0.1	0.1	0.7	8,403
OTC compliance → payment	0.0	0.0	0.1	8,405
Sanction → disbursed	12.0	27.0	48.0	8,376
Login → disbursed	31.4	51.9	77.3	8,374

Every completed transition in the LOS from August 2025 to September 2026, recomputed from timestamps. The dump-derived figures above hold to within a day; the event log adds the QC → U1 wait and confirms that U2 → sanction and the payment chain are hours, not days.

Ops has done its job: sanction → disbursal is a quarter of what it was. Credit has moved the other way while volume rose — the process did not scale. Initiated → login tripled in the last two quarters (1.0 → 3.3 days) even with the feed loans excluded from the median.

3 U2 — who is carrying the queueTwelve regional credit heads carry 400–900 files each; three run a median above 7 days. The fix is capacity and routing, not effort.

Files that reached the business underwriter from March 2026 onward and have a U2 decision. Median and p75 days waiting at U2.

U2 (regional credit)	Regions	Files Mar–Aug 26	Median days at U2	p75 days	Reject %	Action
Ravi Sharma	RAJASTHAN/DNCR	904	4.3	9.0	25%	Hold; watch p75
Deepak Dixit	UP/DNCR	818	6.4	11.2	36%	Add a second U2 or delegate salaried files to central UW
Ramchandra Sarvankar	MMR/WEST MAHA	602	4.8	13.4	29%	Hold; watch p75
Ajay Kumar	MP/CG	595	5.9	12.9	25%	Add a second U2 or delegate salaried files to central UW
Safik S	TAMIL NADU	556	7.0	13.8	4%	Add a second U2 or delegate salaried files to central UW
Sachin Choudhari	VIDARBHA/WEST MAHA	527	7.1	19.1	25%	Add a second U2 or delegate salaried files to central UW
Amol Deshmukh	WEST MAHA	511	9.1	18.8	25%	Add a second U2 or delegate salaried files to central UW
Sudheer Kumar	KARNATAKA/TAMIL NADU	497	5.1	9.2	8%	Add a second U2 or delegate salaried files to central UW
Keyur Makwana	GUJARAT/WEST MAHA	401	1.9	7.0	17%	At standard
Jay Limbad	GUJARAT/WEST MAHA	388	6.0	14.2	21%	Add a second U2 or delegate salaried files to central UW
Sudarshan Reddy	TG/AP	361	5.0	8.7	11%	Hold; watch p75
Holagunda Bharateesha	AP/TG	228	4.0	6.8	14%	Hold; watch p75

Rohan Shah / zonal credit heads: the fix is capacity and routing, not effort. Salaried files are already meant to go to central underwriting; check they actually do. Consider a U2 SLA of 48 hours with automatic escalation to U3 on breach — the U2 → sanction step is already same-day, so the whole credit TAT is one queue.

4 Consistency of credit decisions14% of U1 recommendations are overturned — 3% in the South, 22% in MMR. Same policy, uncalibrated people.

Region	Files with U1 and U2 decision	U1 recommended, U2 rejected	U1 reject-rate spread across people (≥40 files)	Rejections that came after U2	Action
MMR	2,013	22%	22–52%	31%	Credit head to run a joint U1/U2 calibration on last month's overturned files; publish the policy gap
MP	1,356	21%	27–46%	36%	Credit head to run a joint U1/U2 calibration on last month's overturned files; publish the policy gap
UP	1,158	20%	13–63%	28%	Credit head to run a joint U1/U2 calibration on last month's overturned files; publish the policy gap
VIDARBHA	1,473	20%	26–59%	32%	Credit head to run a joint U1/U2 calibration on last month's overturned files; publish the policy gap
DNCR	1,852	18%	24–50%	25%	Credit head to run a joint U1/U2 calibration on last month's overturned files; publish the policy gap
WEST MAHA	1,753	18%	14–40%	34%	Credit head to run a joint U1/U2 calibration on last month's overturned files; publish the policy gap
CG	310	15%	—	32%	Credit head to run a joint U1/U2 calibration on last month's overturned files; publish the policy gap
RAJASTHAN	1,828	14%	26–59%	23%	Review the top-2 overturn reasons monthly
GUJARAT	1,655	11%	18–59%	19%	Review the top-2 overturn reasons monthly
AP	1,786	6%	13–35%	21%	Reference region — document what U1s here do differently
TG	738	6%	18–36%	21%	Reference region — document what U1s here do differently
KARNATAKA	1,946	4%	2–75%	17%	Reference region — document what U1s here do differently
TAMIL NADU	1,130	3%	0–35%	15%	Reference region — document what U1s here do differently

A U1 who rejects 52% and one who rejects 17% on the same product mix are not applying the same policy. The South's low overturn rate is the reference: same policy, better calibrated. A monthly joint review of every U2 overturn in the four worst regions is cheap and the fastest way to move this.

5 Post-sanction leakage — the biggest lever36% of mature sanctions die after the letter; AP 63%, DNCR 42%. The largest single lever, and it sits with sales.

Sanctions from Apr 2025 to Jun 2026 (old enough to have disbursed or died). "Lost" = status now Cancelled, Soft Cancelled, Rejected or Soft Rejected.

Region	Sanctions Apr 25–Jun 26	Disbursed %	Cancelled / rejected after sanction %	Lost sanctioned value ₹Cr	Median sanction → disbursal days	Action
AP	2,060	36%	63%	52	24	Stop sanctioning without a documented customer commitment (fee/e-sign) — RCA on the soft-cancelled book with zonal head
DNCR	1,301	58%	42%	71	18	Stop sanctioning without a documented customer commitment (fee/e-sign) — RCA on the soft-cancelled book with zonal head
RAJASTHAN	1,316	61%	39%	51	13	Weekly SUD call on every sanction older than 15 days; RM named per file
GUJARAT	1,277	63%	37%	52	24	Weekly SUD call on every sanction older than 15 days; RM named per file
UP	794	64%	36%	31	13	Weekly SUD call on every sanction older than 15 days; RM named per file
CG	229	66%	34%	6	31	Weekly SUD call on every sanction older than 15 days; RM named per file
MMR	1,428	65%	34%	79	14	Weekly SUD call on every sanction older than 15 days; RM named per file
VIDARBHA	1,000	65%	34%	33	20	Weekly SUD call on every sanction older than 15 days; RM named per file
WEST MAHA	1,175	66%	33%	55	19	Weekly SUD call on every sanction older than 15 days; RM named per file
MP	875	77%	23%	20	15	At reference level — hold
TG	567	77%	23%	17	6	At reference level — hold
KARNATAKA	1,518	79%	21%	38	9	At reference level — hold
TAMIL NADU	788	80%	19%	29	6	At reference level — hold

Rate and value point at different regions: AP has the highest exit rate (63%) on a ₹52 Cr pool, while MMR (₹79 Cr) and DNCR (₹71 Cr) hold the largest sanctioned value that died. Both belong on the call. AP: 1,247 soft-cancelled Home Loans after sanction, almost none reached ops. That is a sanction being issued before the customer has decided — most likely project/APF bulk sanctions. The fix is a rule, not a chase: no sanction letter without a documented customer commitment (processing fee, e-sign, or WhatsApp "Accept offer" — the button data already exists).

6 Sourcing — three businesses in one pipelineNine digital feeds and two bulk initiators produced 41% of initiations and under 130 disbursals. Real DSAs convert better than direct.

Source	Initiated	Logged in	Login %	Disbursed	Disbursed % of initiated	Disbursed ₹Cr
Direct	25,615	15,830	62%	4,504	18%	498
Real DSA	26,324	20,729	79%	5,728	22%	767
Lead feed	28,703	517	2%	81	0%	11

By partner

DSA / feed (≥300 initiations)	Initiated	Login %	Disbursed	₹Cr	Action
BFDL	9,725	1%	9	1.6	Move to a lead-CRM; create the loan only after a doc upload or RM call
Bajaj Finance Ltd	7,836	2%	30	3.9	Move to a lead-CRM; create the loan only after a doc upload or RM call
Urban Money Pvt Ltd	6,401	81%	1,339	165.4	Keep; benchmark for other DSAs
Andromeda Sales & Distribution Pvt Ltd	6,101	79%	1,302	197.7	Keep; benchmark for other DSAs
AUSHA DIGITAL PRIVATE LIMITED	4,151	1%	10	1.2	Move to a lead-CRM; create the loan only after a doc upload or RM call
Moneyview Limited	4,007	4%	25	3.1	Move to a lead-CRM; create the loan only after a doc upload or RM call
Ruloans Distribution Services Pvt.Ltd.	2,988	80%	611	102.5	Keep; benchmark for other DSAs
Basic Enterprises Private Limited	1,719	82%	377	52.8	Keep; benchmark for other DSAs
BFDL Green Channel	1,517	3%	7	0.9	Move to a lead-CRM; create the loan only after a doc upload or RM call
TENB FINTECH PRIVATE LIMITED	1,205	85%	238	30.6	Keep; benchmark for other DSAs
Paisabazaar Marketing And Consulting Private Limited	619	68%	64	9.7	Hold
Data Mine	609	0%	0	0.0	Move to a lead-CRM; create the loan only after a doc upload or RM call
Retention	605	99%	435	8.5	Keep; benchmark for other DSAs
Star powerz digital technologies Pvt ltd	470	74%	82	12.9	Hold
EBranding India	460	0%	0	0.0	Move to a lead-CRM; create the loan only after a doc upload or RM call
CAPITALNINE SERVICES PRIVATE LIMITED	444	80%	121	15.1	Keep; benchmark for other DSAs
Real Value Finloan Services Pvt Ltd	397	82%	66	10.7	Keep; benchmark for other DSAs

Prerak Mehta / Vinayak Deousker: create loans from aggregator feeds only after a first touch (document upload or RM call logged). Until then they belong in a lead CRM. This alone removes ~1,500 phantom initiations a month and makes the login-desk metric mean something again.

7 RM productivity450 of 1,124 RMs produce fewer than one disbursal every two months — 4% of value. The top quartile delivers 65%.

RMs active in at least three months of the window, tiered by disbursals per active month.

RM tier (by disbursals / active month)	RMs	Initiated	Login %	Sanction % of logins	Disbursed	Median disb / month	₹Cr	Action
Top quartile	281	38,126	47%	56%	6,877	1.6	823	Protect and replicate — pair each with 2 bottom-quartile RMs
Second quartile	281	20,099	46%	43%	2,424	0.9	321	Coach on sanction conversion
Third quartile	281	11,252	47%	33%	816	0.4	109	90-day plan: minimum 1 disbursal a month or move to a lead-generation role
Bottom quartile	281	7,366	39%	22%	62	0.0	8	Exit or redeploy — 62 disbursals across 281 people in 17 months

Ashish Semwal / zone business heads: the bottom quartile's login rate is 57% and sanction rate 20% — they source badly, not just slowly. A 90-day floor of one disbursal a month, applied to the bottom half, either lifts ~₹15–20 Cr a month or frees the cost.

8 Product mixPlot Purchase: 1,715 initiations, one disbursal. Balance Transfer sanctions a third of what it logs.

Product	Initiated	Login %	Sanction % of logins	Disbursed % of sanctions	Median ticket ₹L	Action
Home Loan	42,758	33%	50%	55%	10.0	Fix the login step: most HL initiations are lead-feed; report ex-feed
LAP	10,588	58%	42%	68%	10.6
Balance Transfer	7,906	42%	32%	64%	11.4	Tighten sourcing — a third of logins get sanctioned
Home Construction / Extension	7,663	75%	56%	61%	10.3
Top Up	4,707	83%	42%	72%	5.5
Plot Purchase + Construction	2,934	78%	43%	54%	18.0
Plot Purchase	1,715	6%	4%	25%	10.0	Withdraw or re-spec — 1 disbursal from 1,715 initiations
Micro LAP	1,658	69%	43%	65%	7.7
Home Improvement Loan	713	27%	11%	50%	6.0	Tighten sourcing — a third of logins get sanctioned

9 What is sitting in the pipeline today1,755 initiations idle for more than a month; sanctioned stock is in better shape.

Open files by current stage and days since the last recorded action. Red where more than a quarter of the stage is older than 30 days.

Open stage (today)	Files	≤7 days idle	8–15	16–30	>30 days idle	Owner and rule
Initiated	3,867	808	489	815	1,755	Sales — RM to log or soft-cancel; nothing older than 15 days should exist
Login Done	501	426	23	50	2	Credit — U1 same day
ADR before Sanction	972	233	115	103	521	Sales — DRM clears ADR with customer
Sanctioned	1,222	502	263	263	194	Sales + Ops — SUD dashboard P1–P3
ADR after Sanction	151	22	3	19	107	Sales — collateral/legal ADR
Re-Sanction Required	35	15	12	3	5	Credit — same-day re-sanction

1,755 "Initiated" files idle for more than 30 days can only be dead. Auto-cancel initiations with no application after 15 days; the RM can re-initiate if the customer returns. Sanctioned stock is in better shape — 63% is under 15 days old, which is the SUD dashboard working.

10 Ninety-day planSeven moves in deliberate order: the first two need no hiring and no development, and the second shortens the queue the third then staffs.

#	Move	Owner	Metric and target	Worth
1	No sanction letter without customer commitment (processing fee, e-Sign or Accept-offer reply); weekly named-file call on every sanction older than 15 days in AP, DNCR, Rajasthan, Gujarat	CBO + zone business heads	Post-sanction loss 36% → 25% by December	~₹12 Cr / month
2	Own the wait. ADR standing time counts against the raiser's TAT, not the RM's, unless a stated due date is missed; approvals leave the ADR ledger for a deviation workflow with its own approver and SLA; free-text ADR replaced by a picklist; a document accepted at QC cannot be re-asked; Biz UW made non-blocking and U2's decision final both ways	CRO + COO + CTO	ADR-days as share of journey 18% → 8%; ADRs per login 2.3 → 1.2; queries with no information content 1,925 → 0; login → sanction 15.7 → 10 days from this alone	Half the lost week, at no cost
3	U2 SLA 48 h with auto-escalation; every Salaried Cheque / Cash income-program file to the central CPU for U1 and U2; technical initiated at login	CRO + zonal credit heads + CTO	Central share of files 6% → 40%; Biz UW → U2 median 6.2 → 2 days; login → sanction 10 → 7	The other half of the week
4	Switch the rails on in sequence: e-KYC, PAN and Account Aggregator before submit; penny-drop on the payee and e-Sign and e-NACH at docket; same-day duplicate block and 90-day post-rejection cooling period; feed and bulk leads held in a lead stage until first touch; idle-file auto-cancel extended from feeds to real files at 15 days	CTO + CBO	Ex-feed, deduped initiated → login 74% → 85%; ADR rounds 2.6 → 1.5; e-Sign / e-NACH at docket 0% → 90%	1,500 phantom files a month gone; the desk's workload halved
5	RM answer time as the first sales KPI — every ADR with a due date, response inside one working day; bottom-half RMs on a 90-day floor of one disbursal a month or redeploy; top-quartile RMs paired with the bottom quartile	CHRO + CBO	RM ADR response p75 24 h → 8 h; bottom-half median 0.3 → 0.8 disbursals / month; 6-month RM survival tracked	₹15–20 Cr / month or cost out
6	Right-first-time sanction — amount, tenor and ROI confirmed with the customer before the letter; monthly U1/U2 overturn review on files in MMR, MP, UP, Vidarbha and DNCR	CRO + CBO	Re-sanction share 46% → 25%; overturn 14% → 8%	~6 days off half the book
7	Withdraw Plot Purchase; re-spec Home Improvement and Balance Transfer sourcing; APF pricing and eligibility reviewed before the channel scales; reconcile and file the PMAY-ISS eligible claims (2,946 clean of 3,300 flagged) and make female co-ownership a sourcing default under ₹25L	CBO + CCO + CRO	Product sanction % of logins ≥40% or product off; APF NPA under 1%; ISS claims filed 0 → 2,946	Focus, book quality, up to ₹1.8L per customer

Reporting from October is on the deduped, ex-feed base, with the current figures alongside for one quarter. The order above is deliberate: rows 1 and 2 need no hiring and no development, and row 2 shortens the queue that row 3 then staffs.

11 The long view — FY22 to FY27Four things have looked the same every year since FY22: sanction rate, post-sanction loss, re-sanction share, and real-customer conversion at 24–28%. Growth has been volume.

Same metrics by financial year of initiation, normalised: the nine digital lead feeds and the two bulk lead-initiators (Admin System, Vivek Gajula) are excluded from every year, so FY26–FY27 are comparable with FY22–FY25. FY27 (*) is April–August 2026 (initiations to 31 August; September excluded) and immature at the sanction and disbursal end. Stage definitions are stable across this period; the Biz UW stage exists throughout.

Conversion

Metric (by year of initiation)	FY22	FY23	FY24	FY25	FY26	FY27*	Read
Initiations in LOS (all)	3,746	7,498	12,652	19,994	57,896	22,746	For reference only; every row below excludes the nine digital feeds and the two bulk lead-initiators
Real files (ex feed)	3,746	7,498	12,650	19,993	33,980	13,902	Feed and bulk leads: 0 → 0 → 2 → 1 → 23,916 → 9,865. The feed arrived in FY26
Real customers (deduped, ex-feed)	2,860	5,429	8,542	13,462	23,015	11,237	The true customer base grew 9× in five years
Login % of real files	86%	85%	79%	75%	78%	73%	86% → 75% → 78% → 71%: a slow slide that predates the feed, then a step down in FY27
Sanction % of logins	49%	49%	49%	47%	47%	44%	47–49% for five straight years. Credit conversion has never improved — this is the structural ceiling
Rejects as % of decisions	56%	57%	51%	49%	43%	49%	56% → 43% by FY26 — sourcing quality did improve; sliding back to 49% in FY27
Disbursed % of sanctions	53%	48%	54%	53%	62%	57%	Only FY26 crossed 60%
Sanctions lost after sanction (by sanction year)	44%	43%	45%	45%	38%	24%	43–45% for four years — a design problem, not a new one. FY26 improved to 38%
Re-sanction share	42%	47%	50%	54%	46%	45%	42–54% every year: the first sanction has never been the real offer
U1 recommended, U2 rejected	12%	11%	10%	11%	12%	20%	10–12% for five years, then 20% in FY27 — new credit heads not calibrated to the U1s
Real customers who disbursed	27%	24%	26%	24%	28%	21%	24–28% every year. The business converts the same share of genuine customers it always has — growth has come from volume, not conversion

Time and scale

Median days (by year of initiation)	FY22	FY23	FY24	FY25	FY26	FY27*	Read
Initiated → login	1.2	1.1	1.3	1.2	1.1	2.9	~1.2 days for five years; 2.9 in FY27 on real files alone — the desk is slower even after removing the feed load
Login → U1	2.9	3.8	3.0	2.9	2.8	2.3	Never below 2.3 days in any year
Biz UW → U2	0.7	0.8	1.0	1.0	2.8	5.3	0.7 → 1.0 → 2.8 → 5.1. U2 broke in FY26 when volume tripled
Login → sanction	5.9	7.8	7.7	7.9	11.1	13.0	5.9 → 12.4: doubled, entirely at U2
Sanction → payment release	75	100	49	28	15	9	100 → 9 days. The single biggest operational win in the company's history
Initiated → payment release	88	116	64	45	37	30	116 → 30 days despite credit slowing — ops carried it
Disbursals (first tranche)	834	1,482	2,632	3,748	7,677	2,544
Disbursed ₹Cr (sanctioned amount)	80	142	267	413	921	343
Active RMs	272	397	643	906	1,322	913	RMs 5×, customers 8×: customers per RM rose from 10 to 17
Median ticket ₹L	8.5	8.5	8.7	9.6	10.3	10.8

What is old and what is new, on a like-for-like base. Four findings are structural and have looked the same every year since FY22: sanction % of logins at 47–49%, 43–45% of sanctions dying after sanction, 42–54% of sanctions re-issued, and 24–28% of genuine customers disbursing. Growth has been volume, not conversion — nobody has ever moved these, which is why they are worth the most. Three are new: the feed (23,916 phantom initiations in FY26), the U2 queue (0.7 → 5.1 days) and the U1/U2 overturn jump to 20% in FY27. Login % of real files has slid from 86% to 71% over five years, independent of the feed. One thing improved beyond recognition: sanction → disbursal, 100 days to 9. The company has proved it can fix a stage when it decides to — the next one is U2, and the one after that is the sanction itself.

12 The credit organisation — each decision-maker's own funnel, FY23 to FY27Each sanctioning authority and each U2 has a followable book; the U2 slowdown is by name, and it is the queue, not the people.

Real files only. Who signs the sanction and the rejection is in the dump; from that, each authority's own book can be followed to today. "System" is the auto-generated sanction once the delegated approver has recommended — it is now 44% of all sanctions, so the U2 layer below is where most decisions are actually made.

A Sanctioning layer — CRO, U4 and the three zonal credit heads

Sanctioning authority (first sanction / first rejection by)	FY23	FY24	FY25	FY26	FY27*
System Auto-sanction (delegated authority)
Sanctions issued	—	1,796	2,303	4,738	1,264
Rejections issued	—	0	0	0	0
Login → sanction, median days	—	6.1	7.2	9.9	10.2
Own sanctions that disbursed (mature)	—	55%	59%	61%	73%
Own sanctions re-issued	—	48%	54%	48%	42%
Sanctioned ₹Cr (first-sanction amount)	—	145	212	451	130
Rohan Shah CRO (U4)
Sanctions issued	2,149	2,046	1,733	1,573	609
Rejections issued	426	543	642	457	261
Reject share of own decisions	17%	21%	27%	23%	30%
Login → sanction, median days	7.9	8.9	9.0	13.9	16.0
Own sanctions that disbursed (mature)	48%	51%	49%	69%	79%
Own sanctions re-issued	48%	52%	55%	42%	40%
Sanctioned ₹Cr (first-sanction amount)	222	283	299	288	123
Aniket Modi U4
Sanctions issued	362	397	598	1,096	235
Rejections issued	822	1,029	443	405	99
Reject share of own decisions	69%	72%	43%	27%	30%
Login → sanction, median days	7.7	8.0	9.2	15.9	15.0
Own sanctions that disbursed (mature)	47%	63%	51%	69%	81%
Own sanctions re-issued	34%	52%	49%	49%	41%
Sanctioned ₹Cr (first-sanction amount)	22	26	41	119	36
Navender Gahlawat Zonal credit — North
Sanctions issued	275	484	1,854	1,625	611
Rejections issued	52	41	228	324	317
Reject share of own decisions	16%	8%	11%	17%	34%
Login → sanction, median days	6.6	6.1	6.7	9.0	12.2
Own sanctions that disbursed (mature)	49%	57%	50%	59%	68%
Own sanctions re-issued	46%	45%	53%	53%	49%
Sanctioned ₹Cr (first-sanction amount)	23	41	179	191	81
Jay Limbad Zonal credit — West
Sanctions issued	124	25	147	820	988
Rejections issued	445	531	552	367	389
Reject share of own decisions	78%	96%	79%	31%	28%
Login → sanction, median days	5.8	9.7	9.1	12.2	12.3
Own sanctions that disbursed (mature)	52%	36%	65%	66%	68%
Own sanctions re-issued	57%	56%	69%	66%	53%
Sanctioned ₹Cr (first-sanction amount)	11	2	15	123	131
Krishna A Zonal credit — South
Sanctions issued	—	—	134	1,421	876
Rejections issued	—	—	2	24	15
Reject share of own decisions	—	—	1%	2%	2%
Login → sanction, median days	—	—	9.2	13.0	13.0
Own sanctions that disbursed (mature)	—	—	68%	79%	85%
Own sanctions re-issued	—	—	70%	46%	38%
Sanctioned ₹Cr (first-sanction amount)	—	—	14	167	125
Rajesh Singh Former U2/U3 — West
Sanctions issued	96	80	213	38	—
Rejections issued	518	136	234	35	—
Reject share of own decisions	84%	63%	52%	48%	—
Login → sanction, median days	8.9	7.9	7.3	6.1	—
Own sanctions that disbursed (mature)	47%	50%	48%	45%	—
Own sanctions re-issued	57%	55%	54%	42%	—
Sanctioned ₹Cr (first-sanction amount)	8	7	22	4	—

Three reads. Post-sanction survival differs by signatory, not just by region: Krishna A's sanctions disburse at 79–85%, Navender Gahlawat's at 59–68%, and the North book is where sanctions die — the zonal head's sanction discipline shows up directly in the leakage. Reject share is moving in opposite directions: Navender 17% → 34%, Jay Limbad 95% → 28% as his role changed from West-2 reject authority to zonal sanctioner, Krishna A under 2% throughout. Login → sanction has doubled under every signatory — 6–8 days in FY23–FY25, 12–16 in FY26–FY27 — which confirms the delay is in the queue below them, not in their own decision (U2 → sanction is still same-day for everyone).

B U2 layer — the regional credit managers, by year

Cell = median days a file waited at U2 that year; coloured ≤2 · 2–4 · >4. Beside it: files decided · own reject %.

U2 (regional credit) — median days a file waits at U2	FY23 days · files · reject %		FY24 days · files · reject %		FY25 days · files · reject %		FY26 days · files · reject %		FY27* days · files · reject %
Ravi Sharma NORTH	0.7	342 · 4%	0.9	425 · 9%	0.9	1,769 · 12%	1.9	1,761 · 14%	4.1	736 · 24%
Jay Limbad WEST 1	0.1	358 · 23%	0.1	634 · 12%	0.1	1,407 · 10%	0.7	1,418 · 12%	7.7	251 · 25%
Aniket Modi WEST 1	1.0	471 · 11%	1.0	896 · 12%	1.2	863 · 17%	4.1	539 · 17%	—
Rajesh Singh WEST 1	1.0	936 · 20%	1.1	1,055 · 17%	1.9	732 · 16%	—		—
Amol Deshmukh WEST 1	0.9	403 · 16%	1.1	491 · 12%	1.1	492 · 8%	7.1	912 · 18%	9.0	372 · 24%
Ramchandra Sarvankar WEST 1	—		0.8	236 · 11%	0.9	624 · 19%	1.9	1,106 · 23%	4.7	439 · 30%
Harshal Pardhi WEST 1	0.6	381 · 8%	1.0	621 · 6%	1.4	753 · 13%	4.0	634 · 15%	—
Deepak Dixit NORTH	—		—		2.0	478 · 18%	3.8	1,204 · 13%	6.8	679 · 39%
Sudheer Kumar SOUTH	—		—		1.0	282 · 5%	3.0	1,145 · 2%	5.1	363 · 9%
Navender Gahlawat NORTH	1.0	331 · 8%	1.1	754 · 6%	2.9	382 · 7%	4.2	210 · 3%	—
Amol Wakode WEST 1	0.9	348 · 8%	0.8	248 · 6%	1.0	206 · 5%	0.8	664 · 5%	3.3	115 · 19%
Ajay Kumar WEST 1	—		—		—		5.9	741 · 28%	5.9	485 · 22%
Sudarshan Reddy SOUTH	—		—		—		4.1	648 · 3%	5.0	264 · 13%
Krishna A SOUTH	—		—		0.9	142 · 0%	2.2	698 · 2%	—
Safik S SOUTH	—		—		—		7.9	341 · 0%	6.9	453 · 4%
Sachin Choudhari WEST 1	—		—		—		14.6	271 · 37%	6.3	392 · 21%
Pranav Jogdand SOUTH	—		—		—		1.0	469 · 6%	—
Keyur Makwana WEST 2	—		—		—		—		1.9	380 · 16%
Kesab Pal SOUTH	—		—		—		0.8	327 · 5%	—
Pradeep Jadhav WEST 1	—		—		—		0.1	181 · 1%	0.0	137 · 0%
Ganesh Reddy SOUTH	—		—		—		—		3.0	211 · 11%
Holagunda Bharateesha SOUTH	—		—		—		—		3.9	201 · 14%
Sanjay Singh NORTH	—		0.9	152 · 6%	—		—		—

The same people who cleared U2 in a day for four years now take 4–9: Amol Deshmukh 1.1 → 7.1 → 9.0, Ravi Sharma 0.9 → 1.9 → 4.1, Ramchandra Sarvankar 0.9 → 1.9 → 4.7, Jay Limbad 0.1 → 0.7 → 7.7. Nobody got slower — the queue got longer. Files per U2 roughly doubled between FY25 and FY26 with no new U2s in the West; the North added Deepak Dixit and Sudheer Kumar, the South added Safik S and Sudarshan Reddy, and the new joiners run at 5–8 days from day one. Reject % is also rising with the queue (Deepak Dixit 13% → 39%, Ravi Sharma 12% → 25%) — a tired queue rejects more. This table is the staffing case for U2 capacity, and it is by name.

C Central CPU route vs branch route

Route (U1 done by the central CPU vs a branch/regional U1)	FY23	FY24	FY25	FY26	FY27*
Central CPU
Files with a U1 decision	627	595	357	1,230	685
Share of all U1 files	11%	7%	3%	5%	7%
Sanction %	57%	56%	52%	62%	62%
Rejects of decisions	49%	46%	48%	35%	33%
Login → sanction, median days	6.0	5.1	5.3	7.1	7.3
Initiated → disbursed E2E, median days	157	72	54	24	15
Sanctions that disbursed (mature)	39%	43%	50%	73%	75%
Branch/regional
Files with a U1 decision	4,947	7,798	12,808	21,622	9,319
Share of all U1 files	89%	93%	97%	95%	93%
Sanction %	55%	58%	53%	52%	45%
Rejects of decisions	51%	45%	46%	43%	50%
Login → sanction, median days	8.0	7.9	7.9	11.8	13.0
Initiated → disbursed E2E, median days	113	64	45	37	31
Sanctions that disbursed (mature)	49%	55%	54%	62%	74%

Central CPU is identified as the six U1 underwriters who decide files from three or more zones (Pranav Jogdand, Amit Kale, Sunil Chavan, Kesab Pal, Arpit Raisinghani, Naresh Baraskar). The route was no faster than branch until FY25. From FY26 it pulled away — 24 days E2E vs 37, then 15 vs 31 — and its sanctions survive better (73–75% vs 62–74%). It still handles 5–7% of files. If salaried is a quarter of the real book, three-quarters of salaried files are not reaching it: the routing rule is the first thing to check in LOS.

D Salaried, self-employed and APF — with the real tags

Two of the new reports carry the fields: the QC Approval Report (Income Program, APF flag, U1/U2/U3 dates, technical dates, CIBIL — logins from 24 March 2026) and the Loan Detail Report (Occupation Type, actual disbursed amounts, LTV, FOIR, ROI, asset classification — the whole disbursed book). Salaried = income program Salaried Cheque or Cash; APF = approved-project number present (QC) or direct-authority-allotted purchase type (book) — the book measure is a purchase-type proxy; explicit APF tagging gives 2,308 loans at 1.9% NPA, the same conclusion. 1,059 QC files with no income program recorded are 83% U1-rejected and are left out.

Through credit (QC report)

QC report — logins 24 Mar to 12 Sep 2026, 3,645 files	Salaried	Self-employed	APF
Files	1,028	1,254	304
Median applicant CIBIL	734	740	693
Median application ₹L	10.0	14.0	6.3
U1 reject %	18%	10%	28%
U2 reject % of U2 decisions	16%	15%	4%
U3 reject % of U3 decisions	16%	16%	7%
Sanctioned %	47%	56%	55%
Login → U1, median days	6	7	4
U1 → U2	4	3	1
U2 → U3 (sanction)	1	1	1
Login → sanction, median / p75	19	19	9
p75	27	27	18
Login → technical initiated, median days	9	9	5
Technical reports negative %	24%	21%	1%

Salaried is not faster. Login → sanction is 19 days for salaried and 19 for self-employed, median and p75 alike; the salaried file spends longer at U1 → U2 (4 days vs 3). The salaried advantage exists only where the file goes to the central CPU (Section 12C: 15 vs 31 days E2E) — and salaried is 44% of disbursed loans while central handles 5–7% of files. Roughly five in six salaried files are still underwritten at the branch. APF is the fast lane — 9 days to sanction, U1 → U2 in a day, 1% negative technicals — because the project is pre-approved; its U1 reject rate (28%) is the customer, not the property. Technical is initiated 9 days after login on non-APF files, after U1 in 1,085 of 1,775 cases, and 21–26% of technical reports come back negative. A quarter of properties fail, discovered in week two. That is the single largest avoidable delay in the credit chain and it is a sequencing decision.

On the book (Loan Detail Report)

Loan Detail Report — loans first disbursed FY25 to Aug FY27	Salaried	Self-employed	APF
Loans	5,055	6,495	1,975
Disbursed ₹Cr (actual)	603	778	152
Median disbursed ticket ₹L	10.2	10.4	5.8
Median ROI %	13.0%	14.5%	12.0%
Median LTV / FOIR	53%	50%	67%
FOIR	49%	48%	40%
Median CIBIL (latest)	736	738	728
Initiated → first disbursal, median days	41	36	35
Sanction → first disbursal	20	15	20
Still partially disbursed %	18%	15%	2%
NPA % (asset classification)	0.1%	0.3%	2.1%
Foreclosed %	3.6%	4.0%	0.7%

APF is cheap money with ten times the NPA. ₹5.8L median ticket at 12.0% against 14.0% for the rest, and 2.1% NPA against 0.2%. It is the fastest segment through credit and the weakest on the book — MHADA/direct-allotment borrowers, CIBIL 728. Pricing and eligibility for APF need a separate look before the channel is scaled; the fast lane should not become the default. Self-employed is the better book on every yield and risk line and is also faster to disburse (36 vs 40 days). Sanction → disbursal for salaried is 20 days against 15 — the salaried customer waits longest after sanction, which is where the earlier "salaried takes as long as self-employed" finding lands.

Bureau at login

Applicant CIBIL at login (QC set)	Files	Share	U1 reject %	Sanctioned %
<650	626	18%	40%	36%
650-700	613	18%	44%	25%
700-750	1,120	32%	34%	37%
750+	1,104	32%	25%	49%

One in three logins has a CIBIL below 700 and 40–44% of those are rejected at U1 — the case for the bureau gate at initiation (Appendix B, Rule 3), in the company's own numbers. Above 750, U1 rejects one in four; below 650, two in five.

E Every file, by employment type — from the customer master

The All-Customers export carries Employment Type, PAN, Aadhaar and mobile for every applicant, so the segment split can be done on the whole book rather than the QC and disbursed subsets in D. Salaried = Salaried Cheque or Cash; Self-employed = SENP or SEP.

Customer master Employment Type — real files initiated Apr 25–Aug 26	Salaried	Self-employed	Home-maker / pensioner	All real files
Files	19,486	25,551	2,381	47,787
Unique customers (PAN / Aadhaar / mobile)	17,484	23,020	2,287	42,926
Median applicant CIBIL	724	728	714	726
Median sanction ₹L	11.0	12.2	3.9	11.4
Login %	78%	76%	80%	76%
Sanction % of logins	48%	43%	63%	46%
Rejects of decisions	43%	47%	27%	44%
Initiated → disbursed %	23%	20%	27%	21%
Initiated → login, median days	1.8	1.3	0.9	1.4
Login → U1	2.1	3.0	3.1	2.7
Biz UW → U2	3.8	3.8	1.1	3.3
Login → sanction	11.7	12.1	9.1	11.8
Sanction → disbursal	15	12	10	13
Initiated → disbursed E2E, median	37	34	27	35
p75	58	54	48	56
Post-sanction loss (mature)	35%	36%	46%	36%

Salaried is 41% of real files. On the whole book it converts a little better than self-employed (sanction % of logins, reject share, post-sanction loss) and is not faster anywhere in credit — login → sanction is the same, and Biz UW → U2 is slower. The central CPU (Section 12C) handles 5–7% of files at half the E2E: the routing gap is the largest single salaried lever. Home-maker and pensioner applicants (2,400 files) sanction at the highest rate and are worth a look as a co-applicant strategy — they are mostly the female co-owner PMAY requires (Section 13H).

13 Eradicating the queues — ADR, gates, sequencing and the tranche stockEight document types answered within a day are the login checklist the app should enforce; three credit gates agree 99% with the step before them and should go.

Built from the ADR report (61,568 queries since April 2025, joined to the loan dump for region, branch, RM and phase), the QC Approval report (stage dates and statuses for logins since 24 March 2026) and the Loan Detail report (the disbursed book on 31 August). The organising test: a query answered the same or next day was information the RM already had — so the app should have refused the file without it.

A The login checklist — what the app enforces instead of the desk asking

Query type	ADRs Apr 25–Sep 26	Resolved ≤1 day	Raised by login desk	Raised before login	What the app enforces instead	Mechanism
Co-applicant KYC/docs	21,406	74%	57%	37%	Co-applicant Aadhaar + PAN captured via e-KYC before submit; relationship field mandatory	App gate
Bank statement	18,202	69%	53%	38%	Account Aggregator pull, or PDF upload with 6/12-month check; ABB computed in-app	App gate + AA
Applicant KYC	13,888	77%	66%	46%	Aadhaar OTP e-KYC and PAN verify at initiation; no manual upload path	App gate
Salary slip / employer	11,706	65%	47%	37%	For Salaried income program: 3 slips + Form 16 upload mandatory; employer name from EPFO/PAN	App gate (salaried only)
Geo check-in	8,723	76%	52%	49%	Residence and office geo check-in by RM in app; timestamped photo	App gate
Login fee / cheque	8,651	76%	62%	48%	UPI/payment-link fee at application; login blocked until paid	Payment gate
Consumer app / e-KYC	8,391	75%	64%	51%	Customer consumer-app completion required before file reaches desk	App gate
Selfie / live photo	5,210	71%	56%	45%	Live selfie with liveness check at initiation	App gate
ITR / business proof	5,502	70%	55%	41%	For Self-employed: ITR (2 yrs) or GST/Udyam upload; assessed-income cases flagged for PD	App gate (SENP only)
Property papers	10,314	63%	33%	26%	Sale deed / agreement / plan upload for purchase cases at login; drives technical initiation	Login checklist (purchase)
PD / verification	6,240	61%	31%	24%	Credit-owned; PD scheduled from geo check-in same day	Credit process
Loan statement / RTR	3,902	52%	30%	19%	Bureau pull at initiation replaces most RTR asks	Bureau at initiation

51% of all ADRs are resolved the same day and 70% within one; for the top eight document types it is 71–78%. Those eight are ~90,000 of the 175,000 query lines and 16,277 ADRs are raised before login at 1.7 rounds a file. Every one is a field the app can refuse to submit without. Property papers, PD and loan statements resolve slower (58–66%) and are legitimately credit's — leave those with credit and let the desk stop asking for the rest.

B By region — where to automate first

Same-day and ≤1-day are the RM's speed of answering. High answer speed + high volume = the customer had the data = automate. Low answer speed = RM discipline, not a form problem.

Region	ADRs	Loans	ADRs per login	Rounds per loan	Same-day	≤1 day	Before login	Top query	Action
TG	3,656	1,153	2.27	3.17	53%	68%	26%	Co-applicant KYC/docs	Automate + fix RM discipline — resolution slower, rounds high
GUJARAT	7,018	2,402	2.02	2.92	55%	72%	35%	Co-applicant KYC/docs	Automate first — desk queries answered next day; move the top-3 types into the app gate
RAJASTHAN	7,560	2,530	1.98	2.99	52%	70%	27%	Bank statement	Automate + fix RM discipline — resolution slower, rounds high
TAMIL NADU	4,364	1,616	1.98	2.70	58%	76%	26%	Co-applicant KYC/docs	Automate first — desk queries answered next day; move the top-3 types into the app gate
KARNATAKA	7,307	2,281	1.89	3.20	56%	74%	28%	Co-applicant KYC/docs	Automate first — desk queries answered next day; move the top-3 types into the app gate
DNCR	5,850	2,422	1.63	2.42	52%	68%	24%	Co-applicant KYC/docs	RM discipline — queries sit with sales for 2+ days
WEST MAHA	5,081	2,079	1.59	2.44	45%	61%	19%	Co-applicant KYC/docs	RM discipline — queries sit with sales for 2+ days
VIDARBHA	4,126	1,744	1.56	2.37	42%	63%	25%	Bank statement	RM discipline — queries sit with sales for 2+ days
MP	3,511	1,497	1.56	2.35	36%	55%	23%	Bank statement	RM discipline — queries sit with sales for 2+ days
CG	820	335	1.47	2.45	43%	64%	29%	Bank statement	RM discipline — queries sit with sales for 2+ days
MMR	5,585	2,486	1.45	2.25	46%	67%	28%	Bank statement	RM discipline — queries sit with sales for 2+ days
UP	3,181	1,453	1.37	2.19	44%	61%	21%	Bank statement	RM discipline — queries sit with sales for 2+ days
AP	3,205	1,167	0.94	2.75	57%	72%	26%	Co-applicant KYC/docs	Automate + fix RM discipline — resolution slower, rounds high

≤1-day resolution by region × query type green ≥75% · amber 65–75% · red <65% · small number = ADR volume	Co-applicant KYC/docs		Bank statement		Applicant KYC		Salary slip / employer		Geo check-in		Login fee / cheque		Consumer app / e-KYC		Selfie / live photo		ITR / business proof
TG	73%	1,611	66%	984	75%	845	63%	526	78%	392	78%	411	71%	346	64%	168	68%	806
GUJARAT	79%	3,428	77%	2,524	81%	2,305	73%	1,442	79%	1,231	79%	1,384	75%	1,067	75%	607	82%	868
RAJASTHAN	74%	2,111	73%	2,692	78%	2,027	70%	1,442	77%	912	83%	1,630	77%	1,603	71%	564	76%	225
TAMIL NADU	77%	2,181	74%	1,180	78%	954	68%	915	81%	272	72%	714	84%	110	78%	525	70%	540
KARNATAKA	76%	3,336	64%	1,702	79%	788	68%	1,574	82%	1,138	74%	612	78%	939	73%	501	50%	230
DNCR	74%	2,041	71%	1,929	80%	1,718	66%	1,282	75%	729	81%	835	76%	788	69%	662	74%	527
WEST MAHA	78%	2,141	68%	1,123	77%	381	64%	893	74%	399	66%	406	76%	426	77%	395	66%	371
VIDARBHA	56%	463	62%	1,176	77%	854	61%	741	73%	639	64%	467	72%	610	65%	232	63%	314
MP	59%	609	56%	825	70%	449	50%	403	66%	324	65%	228	66%	437	67%	223	55%	105
CG	75%	153	60%	206	77%	132	62%	128	68%	79	74%	62	66%	128	73%	51	55%	22
MMR	62%	902	69%	2,086	78%	1,710	61%	1,450	75%	1,888	72%	1,255	74%	941	70%	843	68%	715
UP	69%	844	62%	860	75%	666	54%	419	76%	287	76%	257	75%	493	62%	192	58%	293
AP	74%	1,439	71%	814	72%	967	69%	404	81%	349	73%	375	75%	420	68%	155	73%	447

Gujarat, Rajasthan, Karnataka, Tamil Nadu, AP and DNCR answer 70–85% of desk queries within a day across every document type — the data exists at sourcing and the app gate will remove most of their ADR volume outright. MP, Vidarbha, UP and West Maha answer at 55–65% and carry rounds of 2.2–2.4: the gate helps, but the residual is RMs sitting on queries for two days or more, which is a supervision problem. TG and Karnataka run 3.2 rounds per loan — the highest — with fast answers, which means the desk is asking in instalments rather than once. UK and PCH: 51% of queries pre-login, 2.2–2.5 rounds, slow answers — every problem at once, on small volume.

C The desk itself

Login desk (CPA) — who raises, how many rounds, how fast RMs answer them	ADRs	Loans	Rounds per loan	Same-day	≤1 day
Rikin Tiwari WEST 2	5,056	2,322	2.18	67%	84%
Naveen NORTH	3,984	2,161	1.84	66%	81%
Kapil NORTH	3,772	2,379	1.59	64%	79%
Haripriyan B SOUTH	2,537	1,564	1.62	61%	79%
Sagar Holkar WEST 1	2,365	1,781	1.33	71%	85%
Vamsikrishna Nalla SOUTH	1,967	1,045	1.88	53%	69%
Shivam Shukla NORTH	1,654	1,102	1.50	62%	78%
Keshav Verma WEST 1	1,366	954	1.43	46%	71%
Shubham Pawar WEST 1	1,046	757	1.38	51%	75%

Rikin Tiwari (West 2) raises 2.18 rounds per file, Sagar Holkar (West 1) 1.33 on similar volume — same job, 60% more asks. The answer rate to Rikin is the highest on the desk (84% within a day), so Gujarat RMs are not slow; the desk is querying in pieces. A single consolidated checklist query per file, issued once, is the standard; per-CPA rounds should be on the desk lead's daily read.

D Branches and RMs with the worst rounds

Median RM runs 2.6 rounds per file, p90 3.4. Below 2 is achievable today — the best RMs in every region already do it.

Region	Branch (≥80 loans with ADR)	Loans	ADRs	Rounds per loan	≤1 day
KARNATAKA	HOSKOTE	185	780	4.22	79%
KARNATAKA	RAMANAGARA	129	527	4.09	67%
GUJARAT	ANAND	134	517	3.86	78%
KARNATAKA	HUBBALI	115	429	3.73	67%
TG	KUKATPALLY	148	548	3.70	68%
RAJASTHAN	UDAIPUR	88	321	3.65	70%
GUJARAT	JUNAGADH	171	620	3.63	80%
KARNATAKA	DAVANAGERE	136	491	3.61	74%
RAJASTHAN	BHILWARA	253	913	3.61	76%
RAJASTHAN	SIKAR	102	362	3.55	70%
TG	MAHBUBNAGAR	139	493	3.55	74%
TG	KARIMNAGAR	93	324	3.48	62%

Region	RM (≥30 loans with ADR)	Loans	ADRs	Rounds per loan	≤1 day
GUJARAT	Kalpesh Parmar	37	208	5.62	73%
KARNATAKA	Nandeeshgowda B	38	172	4.53	64%
GUJARAT	Alpesh Chavada	43	192	4.47	82%
KARNATAKA	Rajesh L	42	184	4.38	77%
TG	Panchalingala H	30	131	4.37	76%
RAJASTHAN	Harbansh Singh	40	174	4.35	70%
KARNATAKA	Nithishkumar D	38	155	4.08	70%
GUJARAT	Dhaval Vaza	35	142	4.06	74%
KARNATAKA	Aravinda V	57	230	4.04	75%
TG	Durgam Harish	43	165	3.84	71%
VIDARBHA	Mangesh Navhate	48	183	3.81	60%
TG	Naveen Kumar	32	122	3.81	66%

E Three gates to remove

Gate	Evidence (QC report, logins since 24 Mar 2026)	Cost	Decision
QC approval	Login → QC 1 day, QC → U1 3 days (p75 7). 46% of QC approvals are the same day as login, by the same 16 desk people	A second pass on the same file by the same desk	Merge into login. One desk action, one timestamp
B1 — business underwriter	Agrees with U1 on 1,977 of 2,011 recommendations and 1,217 of 1,219 rejections (98.9%); 34 recommendations overturned, 2 rejections reversed. U1 → B1 median 1 day, p75 4	334 files waiting at B1 today; ~1 day on every file	Remove. If sales needs a view, make it a notification, not a gate
U3 — sanction / rejection authority	Sanction is auto-generated ("System", 0 days). Named U3s see rejects almost exclusively: Navender 96% reject, Rohan 94%, Jay 93%, Aniket 91%. 206 of 212 U2 rejections confirmed; 37 of 1,407 recommendations overturned (2.6%)	A rejection waits for a zonal signature: median 15 days login → U3 reject	U2 decision final both ways; U3 review on sampled or high-ticket files only

Login → QC → U1 → B1 → U2 → U3 → sanction becomes login → U1 → U2 → sanction. On the QC cohort this takes the median from 19 days to about 11 before any staffing change, and 334 files come off the B1 hold today.

F Sequencing

Step	Where it sits today	Where it belongs
Personal discussion (PD)	Done 3 days after login (p75 6), never before login, before U1 in 81% of files — on the critical path. Co-applicant PD done on 21% of files	Geo check-in at sourcing schedules PD the same day; PD outcome captured in app before login
Property documents	Sale deed / agreement uploaded 2 days after login (p75 8); 34% never uploaded; technical initiated before the deed is up in 27% of cases	Mandatory at login for purchase cases; upload triggers technical automatically
Technical	Initiated 9 days after login, after U1 in 61% of files; 21–26% of reports negative	Initiated at login in parallel with U1 (Appendix B, Rule 5)
Bureau	One in three logins below CIBIL 700; 40–44% of those U1-rejected	Pull at initiation; hard rules auto-decline before the desk (Appendix B, Rule 3)

G The ninth pipeline stage nobody tracks — undisbursed tranches

Zone	Partially disbursed loans	Undisbursed ₹Cr	≤30 days since last tranche	31–60 days since last tranche	61–90 days since last tranche	91–180 days since last tranche	181–365 days since last tranche	>365 days since last tranche	Idle >90 d ₹Cr
NORTH	347	14.1	3.9 104	1.7 39	0.7 21	1.8 53	2.4 53	3.5 77	7.8
SOUTH	621	31.5	10.2 199	6.5 131	4.2 62	5.6 121	3.7 76	1.3 32	10.5
WEST 1	877	42.6	14.2 242	7.2 120	3.4 79	7.0 148	4.5 114	6.4 174	17.8
WEST 2	97	3.5	1.1 29	0.5 11	0.7 10	0.4 19	0.5 13	0.3 15	1.2

Product	Loans	Undisbursed ₹Cr	Median days since last tranche
Home Construction / Extension	1,150	50.4	41
Plot Purchase + Construction	466	27.8	123
Home Purchase	259	14.8	102
Balance Transfer	79	2.6	40
LAP	76	2.4	70

2,098 partially disbursed loans hold ₹99 Cr that is sanctioned, documented and not paid out. ₹37 Cr has had no tranche for more than 90 days; 300 loans with ₹11.6 Cr are past a year. Construction files take a median 143 days from first to last tranche (p75 222). This is the stock behind the "15% of monthly disbursement is subsequent tranches" figure — it has no ageing report, no owner per file and is on nobody's morning call. It belongs on the SUD dashboard as its own tab: stage-of-construction, days since last tranche, RM named, with the same P1–P5 logic.

H Two things outside the pipeline

Finding	Data	Action
Balance-transfer-out	2,158 foreclosures, ₹235 Cr; 56% inside 24 months; 60% of early exits are DSA-sourced. A "Retention" channel exists (605 files, 72% disbursal) but nothing triggers it	Retention trigger on bureau enquiry or part-prepayment; DSA payout clawback inside 12 months
PMAY-U 2.0 interest subsidy (ISS)	The ISS screening report covers 13,597 disbursed loans: 3,300 flagged eligible, no claim numbers populated (354 of the 3,300 also carry an exclusion remark and four say "subsidy added" — reconcile before filing). 10,297 not eligible — excluded segment 4,493, APF project ineligible 1,786, no female property owner 1,631, income over limit 1,542, sanction before threshold date 540	Reconcile the 354 conflicts, then file the remaining 2,946 — up to ₹1.8 lakh per customer, credited against principal. Make female co-ownership a sourcing default for HL under ₹25L: 1,631 eligible-but-for-that. Review APF projects against PMAY norms before onboarding

14 The event log — the full year, measuredThe full-year event log reproduces every stage figure to within a day, adds the QC → U1 wait, counts 3.9 ADR entries per loan, names re-sanction reasons, and exposes revocation and a single-person OTC compliance gate.

The Loan Workflow report records every status change with timestamp and actor: 954,641 events on 67,371 loans from 1 August 2025 to 14 September 2026, 909,636 on real files after the feed and duplicate rows are removed. Elapsed times here are recomputed from timestamps (the supplied TAT column agrees with the interval on 87% of rows and is not used). Because every completed transition is observed, these are measured dwell times, not inferences from first-and-last stamps.

Milestone gaps

Milestone gap — event log, real files, Aug 25–Sep 26	Median d	p75	p90	Files
Login → QC	0.2	1.2	4.0	30,163
QC → U1	3.0	6.2	14.0	18,913
U1 → Biz UW	0.9	3.0	8.2	17,384
Biz UW → U2	4.1	9.9	20.3	13,528
U2 → sanction	0.0	0.6	1.7	13,206
Login → sanction	14.0	23.8	39.9	13,745
Sanction → docket printed	7.9	19.1	38.1	8,716
Disb auth → OTC done	0.7	3.8	12.0	8,412
OTC → OTC compliance done	0.1	0.1	0.7	8,403
OTC compliance → payment	0.0	0.0	0.1	8,405
Sanction → disbursed	12.0	27.0	48.0	8,376
Login → disbursed	31.4	51.9	77.3	8,374

The event log reproduces the loan-dump figures in Sections 2 and 3 to within a day on every step: the U2 queue (4.1 days median, 20 at p90), the sanction → docket gap (7.9 days, 38 at p90), the OTC step as the only ops tail, and the payment chain in hours. Two things it adds are the QC → U1 wait — 3 days median, 14 at p90, a queue the dump could not see — and the U1 → Biz UW day on every recommendation.

By region

Region (files to sanction)	QC → U1	U1 → Biz UW	Biz UW → U2	Login → sanction	Sanction → docket printed	Disb auth → OTC done	Sanction → disbursed	Login → disbursed	ADR entries	ADR days
AP 2077	4.0	0.4	1.1	11.0	19.7	0.1	22.2	36.0	2.0	1.7
CG 210	2.4	0.8	5.2	11.8	13.4	8.0	23.2	40.0	3.0	2.7
DNCR 1171	2.8	0.8	6.2	14.1	7.7	3.6	17.1	36.1	3.0	2.1
GUJARAT 1128	2.1	3.1	1.2	14.3	8.2	8.3	19.7	38.2	3.0	2.0
KARNATAKA 1419	3.9	0.7	3.7	14.0	5.8	0.1	7.2	26.0	3.0	2.0
MMR 1252	2.1	0.9	3.1	11.9	7.9	0.1	9.1	24.8	2.0	1.6
MP 925	3.4	0.8	5.0	14.0	8.1	0.3	12.2	31.1	3.0	2.6
RAJASTHAN 1165	1.3	1.8	2.9	11.8	6.2	1.7	10.8	26.1	3.0	2.1
TAMIL NADU 1034	4.1	1.1	6.2	17.3	5.2	0.1	7.0	26.9	3.0	1.1
TG 562	4.8	0.2	4.7	15.5	4.8	0.1	6.0	25.1	3.0	2.4
UP 665	3.2	0.3	6.1	14.0	5.0	2.1	10.8	30.1	2.0	2.0
VIDARBHA 897	3.7	0.9	6.7	16.8	11.1	2.2	17.4	41.9	2.0	1.9
WEST MAHA 1210	3.2	1.2	8.5	20.9	13.8	0.1	16.2	42.0	3.0	1.9

The register that accompanies this report (ehfl_fix_register_region_x_step) is built from this table: one row per red or amber cell, with the best region on the same step as proof the standard is reachable, and the fix, owner and target for each.

Time in each state — where files actually sit

State exited	Exits	Time in state, median d	p75	p90
Soft Cancelled	11,347	1.0	6.94	20.1
Rejected by First Business Underwriter	907	0.7	3.72	18.2
Rejected	5,355	0.8	6.03	17.8
Cancelled	1,037	0.0	1.97	12.2
Soft Rejected	1,206	0.9	4.04	11.8
Rejected by First Credit Underwriter	8,811	1.2	3.07	7.7
Sanctioned	121,570	0.0	0.94	5.2
Additional Data Required	94,049	0.1	1.57	4.9
Payment Authorization Done	2,251	0.0	0.15	4.1
Initiated	56,998	0.1	0.80	2.6
Offered/Recommended by First Business Underwriter	76,099	0.1	0.90	2.5
QC Approval Done	205,943	0.1	0.89	2.2
Recommended by First Credit Underwriter	47,178	0.1	0.78	1.9
Docket Printing Done	27,787	0.0	0.18	1.9
Rejected by Second Credit Underwriter	2,617	0.1	0.77	1.8
Recommended by Second Credit Underwriter	2,933	0.0	0.15	0.9
OTC Compliance Done	26,566	0.0	0.14	0.7
Re Sanction Required	9,134	0.0	0.06	0.7
OTC Done	12,550	0.0	0.08	0.2
OPS Done	14,494	0.0	0.05	0.2
Login Done	75,206	0.0	0.06	0.2
Disbursement Authorization Done	16,143	0.0	0.04	0.1
Recommended by First Payment Authorizer	11,843	0.0	0.01	0.0

Additional Data Required is the state most often exited (94,049 times) and the fourth-longest at p90 — files sit in it 4.9 days at the 90th percentile. "Rejected by First Credit Underwriter" holds files 1.2 days median and 7.7 at p90 before Biz UW confirms the rejection: a day added to every "no". The two gates that agree 99% with the step before them — Biz UW and U3 — show as states with a p90 of 2.5 and 1.8 days.

What the log settles

ADR, counted as state entries rather than queries: 86% of logged loans entered ADR, 3.9 times on average (the ADR report's 2.6 counts queries; a query answered and re-raised is two entries); time in ADR per entry 0.15 days median, 1.7 at p75, 5.6 at p90; total ADR days per loan 1.9 median, 7.1 mean, 17 at p90. Section 16's 18%-of-journey figure is conservative. Re-sanction has a reason field — 7,341 transitions on 5,842 loans; the largest documented reason is a technical-report mismatch, which is the case for initiating technical at login.

Re-sanction reason (comment field)	Transitions	Share
Technical report mismatch (property category / land type)	784	11%
ROI change	2,052	28%
Loan amount change	2,069	28%
Tenor / EMI / structure	397	5%
Expiry / validity	20	0%
Other / unstated	2,019	28%

Revocation is a stage. 2,196 "Recommended to Revoke Loan" events on 1,933 loans, 1,849 of them revoking a rejection — rejected files reopened by credit at scale, the same clock-stopping mechanism as ADR under another name. Soft cancel is administrative: 23,636 of 28,543 real-file soft cancels are "Soft Cancelled by Admin", 5,274 by one user; the cancel records a death that happened earlier without activity, which is what Section 0 shows. A stage the dump never stamped: the first OTC compliance authoriser — 22,780 events, 9,376 by one person and 4,073 by a second: a single point of failure on every disbursal, recorded in the Operations Diagnostic's controls.

The states the LOS has

21 transient states appear in the year, including every post-sanction one the dump does not stamp: Docket Printing Done, OPS Done, Second Disbursement Authorizer, OTC Done, First OTC Compliance Authorizer, OTC Compliance Done, First and Second Payment Authorizer, Recommended to Revoke Loan. Every one of the twenty-three stages in the Operations Diagnostic exists here as an event with an actor and a time. The data is captured; the standing export of this table, monthly, is the measurement layer both reports run on.

15 Digital rails against the stages they should serveThe digital rails are wired in and run backwards — used after the desk asks, not before. Sequencing, not adoption.

The digital-journey MIS lists the completion state of every automated check the LOS offers, per applicant, on 550 loans at the login stage since April. It is read here against the queries each rail exists to remove — the ADR types in Section 13, the docket and OTC query types in the Operations Diagnostic — and against the same files' own ADR history, not as a standalone adoption table.

Rail (MIS, 550 loans at login since April)	Completed	Query it would remove (Sections 13 and 8 of the Operations Diagnostic)	On the same files: with vs without
e-KYC	89.1%	Login desk KYC asks: 13,888 ADRs	Applicant KYC asks per loan: 0.16 with vs 0.53 without; ADR rounds 1.33 vs 1.57
PAN verification	20.5%	KYC re-asks at docket: 1,905	Applicant KYC asks per loan: 0.20 with vs 0.20 without; ADR rounds 1.46 vs 1.32
Crime check	13.1%	RCU / verification asks: 6,240	too few completed files to compare
Penny-drop payee verification	3.8%	Cheque-favouring / RTGS queries: 758 across docket and OTC	Bank statement asks per loan: 0.71 with vs 0.37 without; ADR rounds 2.57 vs 1.30
Account Aggregator	0.5%	Bank-statement asks: 18,202 at login, 26,054 overall	too few completed files to compare
e-Sign	0.2%	Agreement / sanction-letter / KFS signature queries: 1,148	too few completed files to compare
e-NACH	0.2%	NACH / SPDC / mandate queries: 987	too few completed files to compare

e-KYC is the one rail that is used and it shows in the queries: files with e-KYC complete get a third of the KYC asks of files without. Everything else is close to zero — and the files that do have PAN, crime-check or penny-drop completed carry more ADR rounds and more bank-statement asks, not fewer. The rails are being run remedially, after the desk has asked, not as a gate before the file reaches the desk. That is the whole point: switched on at initiation they remove the query; switched on in response to it they add a step.

Rails completed on the file (of e-KYC, PAN, crime, penny-drop)	Loans	ADR rounds	KYC asks	Bank-statement asks	Initiated → login d
0	45	1.69	0.60	0.38	3
1	333	1.18	0.15	0.33	2
2	154	1.55	0.16	0.48	3
3	17	2.12	0.47	0.71	5

Read the table above with the previous one: more rails completed → more rounds and a longer initiated → login. Causality runs from the query to the rail, not the other way. The remedy is sequencing, not adoption — e-KYC, PAN and Account Aggregator before submit (Appendix B, Rule 4), penny-drop on the payee at docket, e-Sign at docket, e-NACH at docket (Operations Diagnostic, Section 7).

By region

Region	Loans	e-KYC	PAN	Crime check	Penny-drop
WEST MAHA	115	91%	20%	10%	0%
TAMIL NADU	69	91%	19%	22%	4%
MMR	59	90%	37%	25%	0%
KARNATAKA	49	84%	12%	6%	6%
UP	45	82%	13%	7%	7%
DNCR	43	95%	21%	12%	5%
GUJARAT	31	81%	39%	3%	10%
RAJASTHAN	29	86%	24%	0%	0%
VIDARBHA	27	93%	4%	26%	7%
PCH	24	100%	21%	17%	8%
MP	23	78%	17%	4%	0%

Crime check is a regional habit (MMR 25%, Vidarbha 26%, Rajasthan 0%); PAN verification the same (Gujarat 39%, Vidarbha 4%). Nothing here is a policy; every difference is a branch deciding whether to click. This sample is loans that touched the digital journey; a baseline requires the same export on all logins since April.

16 The ADR ledger — who queries, how long it stands, and what it costsADR standing time is 18% of the customer's journey, up from 1% in FY22. Read one by one, half the queries are not requests for information — conversations, approvals, system fixes, fee reminders and documents already accepted at QC — parked in a ledger that stops no one's clock.

Every "additional data required" query since April 2025 on real files, joined to the loan for stage, region and outcome; FY22 onward for the long view. Standing time is Resolution Date minus Raised Date at day granularity. Department is inferred from the raiser's job title. The event log (Section 14) counts state entries rather than queries and gives 3.9 entries and 7.1 mean ADR-days per logged loan — the figures here are the conservative ones.

A The long view — ADR has become a fifth of the customer's wait

ADR ledger by financial year (raised date)	FY22	FY23	FY24	FY25	FY26	FY27*
ADRs raised	3,007	6,981	13,213	22,270	37,647	23,188
Loans with an ADR	1,778	3,661	5,848	8,639	14,488	8,966
Rounds per loan	1.69	1.91	2.26	2.58	2.60	2.59
Resolved same day	20%	39%	45%	51%	54%	43%
Standing more than 2 days	14%	27%	22%	20%	19%	25%
Standing more than 7 days	5%	11%	8%	7%	7%	8%
Mean days standing per ADR	3.7	3.5	3.0	2.5	2.5	2.7
On loans that disbursed, by year of initiation
Disbursed loans that had at least one ADR	50%	56%	70%	86%	88%	91%
Mean ADR-standing days per disbursed loan	2.1	5.3	5.7	8.7	7.6	6.1
Initiated → disbursed, median days	88	116	64	45	36	29
ADR-standing days as share of end-to-end time	1%	3%	8%	16%	17%	18%

In FY22 half of the loans that disbursed had never been queried and ADR standing time was 1% of the journey. In FY27 nine in ten are queried, 2.6 times each, and the days a file spends waiting on an answer are 18% of initiated → disbursed. The system made the journey shorter (116 → 29 days) and the query loop grew inside it. Rounds per loan stopped rising in FY25; what is now rising is how long each round stands — more than two days on a quarter of them, up from 19%.

B Month on month, FY26 to date

Month raised	ADRs	ADRs per login	Rounds per loan	Standing > 2 d	Standing > 7 d	Total ADR-days	Still open
2025-04	2,238	1.61	2.24	17%	5%	3,990	68
2025-05	2,642	1.65	2.28	18%	6%	6,245	79
2025-06	2,634	1.35	2.14	17%	7%	5,929	88
2025-07	3,186	1.61	2.29	19%	8%	8,771	82
2025-08	2,787	1.61	2.13	21%	7%	6,374	85
2025-09	3,359	1.56	2.19	18%	7%	7,048	83
2025-10	2,517	1.56	2.10	20%	9%	6,408	97
2025-11	3,317	1.08	2.12	18%	7%	7,856	52
2025-12	3,246	1.00	1.99	20%	8%	9,181	72
2026-01	3,991	1.53	2.13	21%	8%	10,991	98
2026-02	3,874	1.52	2.04	18%	7%	9,536	76
2026-03	3,856	1.66	2.10	18%	7%	10,376	97
2026-04	3,143	1.74	2.16	23%	9%	8,895	103
2026-05	4,107	2.30	2.37	22%	8%	10,556	107
2026-06	4,362	2.23	2.22	26%	9%	12,913	166
2026-07	4,731	2.31	2.25	28%	10%	13,782	162
2026-08	4,995	1.95	1.99	28%	9%	12,065	272
2026-09	1,850	1.74	1.51	14%	1%	1,736	428

Two turns in the series. From May 2026 the desk raises 2.2–2.3 queries per login against 1.5–1.6 a year earlier — the feed and duplicate load arriving at the desk. From June 2026 the share standing more than two days climbs from 18% to 28% and open queries triple. September is partial; the 225 open items are the current backlog.

C Who raises the ADR, and how long theirs stand

Raising department (Apr 25 →)	ADRs	Same day	Mean days standing	> 2 d	> 7 d	Total ADR-days consumed	Open
Login desk	23,464	63%	1.2	12%	3%	28,359	617
Credit	18,643	44%	3.0	24%	9%	53,244	819
Other	6,476	42%	4.2	29%	12%	25,829	311
Technical	3,221	30%	5.7	38%	17%	17,335	161
RCU	3,078	32%	4.2	34%	13%	12,348	146
Operations	3,024	66%	1.1	9%	3%	3,331	27
Sales	1,662	38%	4.6	32%	14%	7,177	85
Legal	1,267	26%	4.1	38%	14%	5,029	49

The desk raises the most queries and clears them fastest: 63% same day, a mean of 1.2 days. Credit raises fewer and consumes almost twice the days — 49,933 ADR-days against the desk's 28,336 — because its questions stand 2.7 days on average and a quarter go past two. Technical's stand nearly five days and 39% past two; RCU and Legal similar. Operations' stand a day. The cost of ADR is not where the volume is.

Department × stage at which the ADR was raised	Before login	Login → U1	U1 → sanction	Sanction → payment	After payment / none
Credit	4,020	7,453	3,932	1,344	1,894
Legal	0	8	496	470	293
Login desk	11,139	10,544	209	56	1,516
Operations	1	2	29	2,542	450
Other	854	2,421	1,321	926	954
RCU	1	0	1,901	425	751
Sales	54	95	783	311	419
Technical	11	124	2,169	171	746

Credit and technical queries arrive between U1 and sanction — on files that are already three to five days into credit — which is why their standing time lands directly on the login → sanction TAT in Section 2.

Median days standing, by month × raising department small number = ADRs raised	Login desk		Credit		Technical		RCU		Legal		Sales		Operations		Other
2025-04	0.0	549	1.0	659	3.0	16	2.0	22	2.0	49	0.0	104	0.0	107	0.0	732
2025-05	0.0	975	1.0	686	2.0	64	2.0	34	1.0	41	0.0	114	0.0	111	1.0	617
2025-06	0.0	994	1.0	791	1.0	82	1.0	36	2.0	46	1.0	86	0.5	63	1.0	536
2025-07	0.0	1035	1.0	1094	1.0	101	1.0	76	1.0	51	1.0	78	0.0	137	1.0	614
2025-08	0.0	876	1.0	925	1.0	71	1.0	131	3.0	56	0.5	80	0.0	136	1.0	512
2025-09	0.0	1254	1.0	894	1.0	119	1.0	193	2.0	64	1.0	74	0.0	151	1.0	610
2025-10	0.0	900	1.0	685	2.0	122	1.0	170	1.0	46	0.0	44	0.0	113	1.0	437
2025-11	0.0	1216	1.0	930	2.0	179	1.0	213	1.0	55	2.0	85	0.0	122	1.0	517
2025-12	0.0	985	1.0	1016	2.0	229	1.0	215	2.0	67	1.0	72	0.0	120	0.0	542
2026-01	0.0	1284	1.0	1344	2.0	286	1.0	276	2.0	79	1.0	125	0.0	188	0.0	409
2026-02	0.0	1486	1.0	1215	2.0	295	1.0	246	1.0	74	1.0	98	0.0	182	1.0	278
2026-03	0.0	1622	1.0	1236	1.0	218	1.0	209	1.0	103	2.0	118	0.0	206	1.5	144
2026-04	0.0	1269	1.0	994	2.0	164	2.0	175	2.0	84	2.0	94	0.0	212	2.0	151
2026-05	0.0	1791	1.0	1319	2.0	310	2.0	152	2.0	73	2.0	74	0.0	256	2.0	132
2026-06	1.0	2021	1.0	1379	2.0	263	1.0	178	2.0	65	1.0	91	0.0	252	2.5	113
2026-07	1.0	2187	1.0	1402	1.0	365	1.0	229	2.0	80	1.0	89	0.0	309	2.0	70
2026-08	1.0	2151	1.0	1445	1.0	337	1.0	420	1.0	193	1.0	140	0.0	250	1.0	59
2026-09	1.0	869	0.0	629	—	0	1.0	103	1.0	41	0.0	96	0.0	109	1.5	3

D By raiser

Department	Raiser (≥300 ADRs)	ADRs	Loans	Rounds per loan	Standing > 2 d	ADR-days consumed
Login desk	Rikin Tiwari	4,987	2,287	2.18	9%	4,290
Credit	Suraj Patil	4,280	2,047	2.09	8%	3,842
Login desk	Naveen	3,902	2,115	1.84	12%	4,744
Login desk	Kapil	3,685	2,328	1.58	12%	4,263
Login desk	Haripriyan B	2,523	1,555	1.62	12%	3,192
Login desk	Sagar Holkar	2,336	1,758	1.33	8%	2,004
Other	Saidavali Miriyala	2,026	1,149	1.76	7%	2,580
Login desk	Vamsikrishna Nalla	1,955	1,035	1.89	17%	3,580
Credit	Sanket Kshirsagar	1,890	1,408	1.34	15%	2,752
Credit	Naresh Baraskar	1,828	1,442	1.27	14%	2,383
Login desk	Shivam Shukla	1,611	1,072	1.50	15%	2,488
Login desk	Keshav Verma	1,359	948	1.43	16%	2,325
Login desk	Shubham Pawar	1,036	751	1.38	12%	1,347
Credit	Safik S	706	686	1.03	5%	487
Credit	Sudheer Kumar	629	500	1.26	30%	2,305
RCU	Rakesh Ganji	601	392	1.53	34%	2,378
Legal	Vinod Rathod	549	479	1.15	44%	2,333
Credit	Ajay Kumar	523	409	1.28	33%	2,609
Credit	Krishna A	523	456	1.15	22%	1,116
Credit	Jay Limbad	505	394	1.28	44%	2,949
Credit	Ravi Sharma	471	402	1.17	34%	2,021
Technical	Sagar Sharma	463	289	1.60	29%	1,490
Technical	Priyanshu	425	325	1.31	30%	1,320
RCU	Sudharson P	414	300	1.38	36%	1,433
Technical	Rahul Daina	408	313	1.30	37%	2,378

E Who resolves, and how fast

Who records the resolution	ADRs	Share	Mean days standing	> 2 d
Another RM / BM	27,161	45%	2.2	20%
Other	16,373	27%	1.4	15%
The RM it was raised for	8,469	14%	2.7	22%
Credit	4,520	7%	3.7	28%
The raiser (self-closed)	3,962	6%	5.4	42%
Login desk	350	1%	4.0	29%

The person the query was raised for records the resolution on 14% of ADRs; another sales name on 45%; the raiser closes their own query on 6.5% — and those self-closures are the slowest, 5.4 days mean and 42% past two days, which reads as stale queries being tidied rather than answered. Resolution is recorded by whoever gets to the screen, so per-person resolver speed below is indicative.

Slowest resolvers (≥150 ADRs)	Region	ADRs	Median days	> 2 d
Avinash AV	KARNATAKA	158	2.0	46%
Yogesh Bisht	DNCR	156	2.0	45%
Vinod Yadav	MMR	156	2.0	41%
Shekhar Shinde	WEST MAHA	224	1.0	41%
Ganesh Pawar	VIDARBHA	175	1.0	39%
Ajay Kumar	MP	208	1.0	38%
Jayant Sakure	VIDARBHA	157	1.0	37%
Alok Kumrawat	MP	297	1.0	36%
Amol Gaikwad	WEST MAHA	340	1.0	34%
KranthiKumar N	AP	171	1.0	34%
Ajay Sharma	RAJASTHAN	382	1.0	32%
Shashikant Hajare	WEST MAHA	165	1.0	31%
Sunil Peddi	WEST MAHA	188	1.0	30%
Ravi Annaram	TG	176	0.0	30%
Anilkumar Bhajantri	KARNATAKA	192	0.0	28%

Fastest resolvers (≥300 ADRs)	Region	ADRs	> 2 d
Anilkumar Chauhan	GUJARAT	342	8%
Nikhil Vasiyar	GUJARAT	342	9%
Piyush Saxena	DNCR	332	10%
Vinaysagar GM	KARNATAKA	357	11%
Tushar Ushadadiya	GUJARAT	384	14%
Santhosh Raj P	TAMIL NADU	337	14%
Mahesh Dinkar	MMR	612	15%
Kundan Vinchurkar	VIDARBHA	394	19%
DA Madhukumar	KARNATAKA	359	19%
Jayesh Babshetty	MMR	652	20%

F By region and raising department

Median days standing, region × raising department	Login desk		Credit		Technical		RCU		Legal		Sales		Operations		Other
AP	0.0	1006	1.0	337	2.0	197	1.0	75	0.5	2	0.0	84	0.0	318	0.0	1151
CG	0.0	309	1.0	257	3.0	84	1.0	47	1.5	30	1.0	34	0.0	22	2.0	35
DNCR	0.0	3347	2.0	1084	1.0	457	1.0	154	0.0	7	1.0	254	0.0	160	1.0	269
GUJARAT	0.0	4875	2.0	651	3.0	79	1.5	190	1.5	67	4.5	113	0.0	329	3.0	625
KARNATAKA	0.0	146	0.0	5904	0.0	234	1.0	250	3.0	58	1.0	134	0.0	261	1.0	266
MMR	0.0	1023	1.0	2615	2.0	14	1.0	378	1.0	3	1.0	204	0.0	841	2.0	479
MP	1.0	1056	1.0	1163	1.0	401	1.0	171	1.0	110	1.0	170	1.0	34	2.0	384
RAJASTHAN	0.0	4134	1.0	852	1.0	677	1.0	390	2.0	318	2.0	183	1.0	205	1.0	681
TAMIL NADU	0.0	2296	0.0	1078	2.0	21	2.0	429	—	0	0.0	12	0.0	274	1.0	238
TG	0.0	1161	1.0	739	3.0	152	1.0	137	0.0	3	0.0	132	0.0	224	0.0	1056
UP	0.0	1418	2.0	711	2.0	319	1.0	62	—	0	0.0	61	0.0	74	1.0	440
VIDARBHA	0.0	136	1.0	2361	—	0	2.0	362	1.0	334	2.0	148	0.0	213	2.0	521
WEST MAHA	0.0	2321	3.0	853	3.0	579	1.0	427	2.0	335	3.0	126	1.0	68	2.0	325

G What is lost to ADR

Stage the loan had reached when last queried	Loans with ADR	Of which lost	Loss rate
Before login	828	141	17.0%
Login → U1	637	327	51.3%
U1 → sanction	10,677	322	3.0%
Sanction → payment	1,536	199	13.0%
After payment	9,285	10	0.1%

999 of the 22,963 loans queried since April 2025 were soft-cancelled, cancelled or soft-rejected — 4.4%. 148 of them died with a query still open; 56 had a last query that stood more than seven days. Loss is highest on files queried before login (17%) and between login and U1 (51% of the small number there), and near zero once sanctioned. Queried files are, by definition, files being worked: the files that die are overwhelmingly the ones nobody queried (Section 0).

Department that raised the last ADR on a lost loan	Lost loans	Loss rate for that department's last-ADRs
Login desk	380	5.5%
Credit	296	3.9%
Other	104	4.4%
Technical	94	7.6%
RCU	62	6.3%
Operations	24	1.0%
Sales	23	3.1%
Legal	16	2.7%

ADR rounds on the loan	Loans	Lost	Disbursed	Median ADR-days
1	7,138	4.8%	23%	0
2	6,067	4.4%	36%	1
3	3,997	4.4%	47%	3
4–5	4,106	3.7%	60%	6
6+	1,655	4.0%	70%	12

More rounds do not kill a file — loans with six or more rounds disburse at 70% against 23% for one round — they cost it time: twelve median ADR-days on the six-plus group. The ADR cost is measured in days, not in customers, and it is concentrated in credit, technical and RCU queries raised after login rather than in the desk's checklist.

Region	Loans with ADR	Lost	Median ADR-days per loan
TAMIL NADU	1,609	14.6%	1
TG	1,131	12.0%	2
AP	1,152	7.6%	1
RAJASTHAN	2,478	5.5%	2
DNCR	2,367	5.4%	1
UP	1,406	3.5%	2
MMR	2,472	3.2%	1
KARNATAKA	2,264	1.8%	2
VIDARBHA	1,716	1.5%	2
GUJARAT	2,364	1.4%	1
WEST MAHA	2,054	1.3%	2
MP	1,488	1.3%	3
CG	334	0.3%	2

H Read one by one — what the queries actually are

Every query line was classified by content, in priority order, into eleven buckets. Eight of them are not requests for information at all — they are conversations, approvals, system corrections, fee reminders and blank clicks that live in the ADR ledger because the ledger is the one place a file can be stopped without the stop appearing on anyone's turnaround time.

What the query actually is	ADRs	Share	Same day	Mean days standing	ADR-days	Raised by desk / credit	Raised after login	What it should have been	Fix
Placeholder — no question asked	363	0.6%	91%	0.2	69	17% / 43%	94%	Nothing — a status note, or a click to stop a clock	Delete; ADR requires a document or a question from a picklist
Conversation — "call me", "as discussed", "discuss with"	1,113	1.8%	34%	4.0	4,173	12% / 51%	92%	A phone call, and then a query stating what was agreed	Free text disabled; "as discussed" must name the document or decision
"Kindly clear / check / confirm" — no content	449	0.7%	45%	3.0	1,302	16% / 33%	91%	A specific item	Mandatory item field
Approval routed as a query	3,439	5.6%	50%	2.4	8,029	31% / 31%	82%	An approval — single-lady, insurance waiver, deviation, login-fee waiver, 2nd HL, agri-zone — from the deviation matrix, with its own SLA and approver	Approval workflow in LOS; an ADR cannot carry the word "approval"
Pending from another department (technical / legal / RCU / FI / PD)	142	0.2%	12%	10.1	1,319	0% / 21%	100%	A task on Technical, Legal or RCU — raised against the RM instead	Raised on the owning department; RM never sees it
System correction / data-entry error	2,575	4.2%	50%	2.6	6,550	7% / 46%	84%	A validation rule at data entry — name as per KYC, document in the right tab, Aadhaar uploaded, fee cleared before CIBIL	Field validation at application; no ADR
Bureau expired / re-pull — caused by delay	44	0.1%	68%	0.9	39	2% / 23%	98%	A consequence of the file taking longer than the bureau validity	Auto re-pull at 30 days; disappears when TAT does
Login fee / processing fee	6,535	10.6%	56%	1.6	10,151	66% / 22%	49%	A payment — the file should not exist in the queue until it is paid	Payment gate at application (Appendix B, Rule 4)
Document that is on the login checklist	31,596	51.3%	54%	2.1	63,669	51% / 29%	67%	The login checklist, enforced once at application; the 8,965 raised by credit, technical, RCU and legal after login should have been caught at QC	App gate at application; QC checklist mirrors the app
Property / legal / technical document	5,158	8.4%	35%	4.9	23,958	10% / 22%	94%	A property-document checklist at login for purchase cases; legal and technical asks belong to those departments	Purchase-case document gate; technical initiated at login
Other / substantive	10,154	16.5%	43%	3.5	34,603	13% / 37%	92%	Genuine credit questions — income, PD, structure	This is what ADR is for

The ledger is being used as a clock-stopper. ADR time is deducted from no one's TAT — not the desk's, not U1's, not the RM's. Four patterns in the text show it being used that way. Blank and conversational queries: 1,925 ADRs say "call me", "as discussed", "please clear", "ok" or nothing — 5,475 days of files standing still behind a phone call that had already happened. Self-closed queries: 4,026 ADRs were raised and resolved by the same person, 2,747 of them on a later day — a note to self that paused the file; 2,022 of them from credit. Credit queries raised on the eve of the decision: 2,307 of credit's 18,832 ADRs (12%) were raised the day before or the day of the U1 decision — the file is decided the next morning, and the day in between is booked to the RM. Checklist documents asked for a second time: 8,965 ADRs from credit, technical, RCU and legal ask, after login, for a document that was on the login checklist — KYC, banking, salary slip, business proof, address proof — on 6,373 loans, standing 3.5 days each and 29,402 days in all. Every one of those passed QC.

What could have been resolved without an ADR. The eight non-information buckets plus the checklist documents re-asked after QC come to 23,625 of 61,568 ADRs (38%) and 61,034 of 153,862 standing days (40%). They resolve to four mechanisms: an approval workflow with its own approver and SLA; field validation at application; a payment gate; and a QC checklist that mirrors the app, so that a document the desk accepted at login cannot be asked for again by credit. The remainder — genuine credit questions and property documents — is what the ADR mechanism exists for, and it should carry the clock: ADR standing time counts against the raiser's TAT, not the RM's, unless the RM has failed a stated due date. That single KPI change removes the incentive that produced the eight buckets.

The text, as written

Bucket	Verbatim examples from the ledger
Placeholder — no question asked	· Please clear ADR · Login Done · Clear ADR · RESOLVE
Conversation — "call me", "as discussed", "discuss with"	· As discussed 1 financial guarantor to be add in loan deal · Wt case is rejected \| No discussion done \| · As discussed SOA required of active loans. · customer not received call.
"Kindly clear / check / confirm" — no content	· kindly clear technical approval · Recheck · please clear mandatory rule check · Please confirm offer loan loan amount.
Approval routed as a query	· 1) Admin Fees Pending In the system \| 2) Revise Insurance Approval Req · login fees approval pending \| applicant ekyc consumer app pending \| co app ekyc pending · Dear Team \| Technical to be approved before Sanction · 1.applicant osv document required \| 2.system approval pending
Pending from another department (technical / legal / RCU / FI / PD)	· Technical Report Required \| · Technical Pending · legal technical pending · Technical pending
System correction / data-entry error	· ROSHANI AADHAR CARD NOT UPLOADED IN SYSTEM \| RM RESIDENCE VISIT PENDING \| BANK STATEMENT PASSWORD REQUIRED \| · Pls correct gift deed upload in system · LOGIN FEES PENDING \| Roshan AND Haridini AADHAR CARD NOT UPLOADED IN SYSTEM \| \| · RCU HOLD:- Req. Applicant Harihar- business proof with complete business address with popular landmark upload in system
Bureau expired / re-pull — caused by delay	· Generated fresh CIBIL and CRIF reports \| Take latest pay slip and banking \| · Seller KYC and Seller CIBIL Require · latest PD to be done and upto date banking analysis to be provided \| re-generate fresh CIBIL & check any new loans are t · Generate fresh CIBIL and comment about this \| Taking latest Banking & salary slip \|
Login fee / processing fee	· \| 1 kindley login fee required \| 2 CO APP Balavant INCOME PROOF REQUIRED \| 3 CO APP Balavant CONSUMER APP DAWONLOD REQ · please complete login fees pending · IMD Receipt Pending \| App 6 Month Banking Pending \| Co App Check In Residence Pending \| · 1)login fee \| 2)app not download consumer app
Document that is on the login checklist	· 1.APPLICANT BUSINESS PROOF REQUIRED \| · Santosh Aadhar KYC Pending \| Atar Singh Aadhar KYC Pending · Not available for PD. Applicant and co-applicant additional income not added. · 1 APP AND CO APP CURRENT ADDRES PROOF REQUIRED \|
Property / legal / technical document	· Dear Team \| \| Kindly provide SOA of Running loan HL or EMI paid proof . \| \| \| Thanks and Regards \| · 1 Vetting Report Req \| 2 MODT RR copy Req · require complete property demarcation. · Under-construction property can't allow in LAP product
Other / substantive	· customer not available for tele PD. (contact number is out of coverage) \| · clear log in fee \| upload appropriate income proof · PD Report pending Because this case is Karanja Maharashtra Location We are initiate the case of Nearest branch of Kara · add guarantor

Appendix B — the seven LOS rules

The system changes that the findings resolve to. Each is sized from the data in the sections it cites; Section 15 shows that most of the underlying rails already exist in the LOS.

#	Rule in the loan origination system	Evidence	Effect
1	Duplicate block at initiation on PAN + mobile: same-day duplicates refused; a customer rejected in the last 90 days routed to a cooling-period exception, not a new file	4,861 duplicate files (10% of the real book), 57% created the same day, 92% by the same RM; 1,416 re-initiated within 90 days of a rejection (Section 0)	Honest denominators; the cooling period enforced
2	Lead-feed and bulk-initiated files stay in a lead stage until first touch (document upload or logged call)	32,825 initiations, under 130 disbursals (Section 6); the auto-cancel already fires on them (Section 14)	The LOS carries only real files
3	Bureau pull and hard-rule engine at initiation — auto-decline on DPD, overdue and write-off codes; FOIR computed before the desk	32% of rejects are bureau facts; one in three logins is below CIBIL 700 and 40–44% of those are rejected at U1 (Sections 12D, 13F)	About 30% fewer files reach the desk and U1
4	Document gate before the file is queued to the desk — e-KYC, PAN, consumer-app completion, login fee, bank statement (Account Aggregator or upload), co-applicant KYC, one income proof	Top eight ADR types are ~90,000 of 175,000 query lines and are answered within a day 71–78% of the time (Section 13A); e-KYC completion cuts KYC asks by two-thirds (Section 15)	ADR rounds 2.6 → about 1; desk TAT to same-day
5	Technical initiated at login, in parallel with U1; property documents mandatory at login for purchase cases	Property rejects are 82% after U2 at 19 days; 21–26% of technical reports negative (Sections 12D, 13F)	Late rejects cut by a third; a week off sanction TAT
6	Auto-cancel idle files — Initiated 15 days without application; ADR-before-sanction 30 days without response	1,755 initiations idle over 30 days; ADR-before-sanction median 38 days since login (Section 9)	Stock reflects reality
7	Structured ADR and reject codes — checklist picker replaces free text; "OTHER" requires a sub-code	22% of rejects coded OTHER; ADR is free text (Section 13)	Reasons become reportable

Method and caveats

Window: loans initiated 1 Apr 2025 – 31 Aug 2026. Earlier cohorts excluded (business-underwriter stage introduced Sep 2021; sourcing mix and stage definitions changed materially through FY25). September 2026 excluded as incomplete.
Customer identity: Sections 0 and 12E collapse files to customers on PAN, else Aadhaar, else mobile, from the All-Customers export (PAN valid on 98% of real files, Aadhaar 99%, mobile 100%). Of 47,816 real files, 29 have no main-applicant row in the export and are excluded, leaving 47,787. Multiple files on one key are not all errors — top-ups and second properties are legitimate — which is why the remedy is a same-day duplicate block and a post-rejection cooling period, not deletion. The loan dump's own name field holds first names only for much of the South and Rajasthan book and is not usable as a key; on the identity keys the real book is 47,787 files, 42,926 customers and 4,861 duplicate files (10%).
Cohort basis: conversion is tracked by initiated month (funnel) and login month (TAT), each loan counted once at its first stage date. This is a true cohort view, unlike the MTD flow ratios in the daily extracts.
Maturity: post-sanction leakage uses sanctions before 1 Jul 2026 only; the last two quarters' disbursal rates are still rising.
Lead feed: a DSA is classed as a feed when it has ≥50 initiations and <25% login (nine partners); a direct initiator is classed as a bulk lead source when it has ≥200 files not in its own RM name and <25% login (Admin System, Vivek Gajula). Ordinary RM-to-RM file transfers are not excluded. Both rules are transparent and can be tightened.
Rupees: loan_amount in the dump is the sanctioned amount — it matches the Loan Detail Report to within 0.1% on the same loans. Actual paid-out on loans first disbursed Apr 2025–Aug 2026 is ₹1,201 Cr (₹832 Cr FY26, ₹369 Cr Apr–Aug FY27); cumulative since inception ₹2,119 Cr. Rupee sizing throughout is stated at sanction value and overstates paid-out by about 7%.
Actors: underwriter and RM figures use the name recorded on the first stage action; shared logins or name changes would distort individual rows, not totals.
ADR classification: query lines are typed by keyword (co-applicant, bank statement, Aadhaar/PAN, salary, ITR/GST, selfie, consumer app, fee, check-in, PD, property, loan statement); a line can match more than one type. Resolution speed is Resolution Date minus Raised Date at day granularity — the ADR report carries no time of day.
Workflow log: 1–10 Sep 2026 only; dwell times are time in the previous state and are lower bounds because unfinished files are not observed. Digital MIS: 550 loans at login stage since April, 531 still at login on 14 Sep — no outcome correlation is possible; the with/without comparison is on ADR content only.
Not in the loan dump: customer-level reasons for cancellation, the salaried / self-employed / APF fields (present in the QC Approval and Loan Detail reports, used in Section 12D — they must be added to the dump query), CIBIL, property type, pricing. Each finding above names the where and the who; the why needs the ADR remark and reject-reason fields, which this dump does not carry.

Appendix A — the brief this answers

Objective. Using the full loan-level dump, identify where EHFL loses volume, time and money between initiation and disbursal, quantify each leak, and name the owner and the 90-day action — at McKinsey diagnostic standard: cohort-based, sized in rupees, ranked by value.

Scope. Loans initiated Apr 2025 onward (comparable process). Dimensions: zone → region → branch, channel (direct / DSA / feed), product, RM, underwriter (U1, Biz UW, U2). Stages: initiated, application, login, U1, Biz UW, U2, sanction / rejection, ops, payment authorisation, disbursal.

Questions. (1) Where does conversion break, and is it getting better or worse by cohort? (2) Which stage owns the TAT increase? (3) How consistent is credit decisioning across people and regions? (4) How much sanctioned value dies before disbursal, and where? (5) Which sourcing channels, partners, RMs and products deserve more or less capital? (6) What stock is sitting idle today?

Output. RAG tables by region/person with a specific action per row; a ranked 90-day plan with owner, metric, target and ₹ worth; method and caveats stated. No prose summaries of tables.

Glossary

Account Aggregator: RBI framework for consented digital bank-statement fetch
ADR: Additional data required — a query raised on a file, mostly by the login desk, that the RM must answer before the file moves
ADR round: One query raised on a file — a full stop-answer-resume cycle; rounds per loan is how many separate queries the loan received over its life
APF: Approved project financial — a builder project pre-approved for lending
BCM: Branch credit manager
Biz UW: Business underwriter — a sales-side recommendation step between U1 and U2 (B1 in the QC report)
case-days: Files multiplied by days above standard — the volume-weighted cost of a delay
CERSAI: Central registry of securitisation asset reconstruction and security interest — the statutory register of mortgages
cohort: All files that entered a stage in the same period, followed to today — as opposed to a flow ratio, which divides this month's outputs by this month's inputs regardless of when each file entered
CPA: Credit processing associate — the 16-person central login desk
CPU: Central processing unit — the HO underwriting team for salaried files
DPD: Days past due
DSA: Direct selling agent — a third-party sourcing partner
E2E: End to end — initiated to disbursed on the same file
EC: Encumbrance certificate — the sub-registrar record proving the mortgage is registered
flow ratio: This period's outputs divided by this period's inputs — quick, but files from earlier periods leak in, so it can exceed 100%
FOIR: Fixed obligation to income ratio — EMIs as a share of income
ISS: Interest subsidy scheme under PMAY-U 2.0
ISS eligible: Flagged eligible in the ISS screening export; 354 of the 3,300 also carry an exclusion remark and need reconciling before filing
KFS: Key facts statement — the regulatory loan summary the customer signs
lead feed: A digital lead aggregator whose leads are created as loan files automatically
LOS: Loan origination system
LTV: Loan to value — loan amount as a share of property value
median: The middle value — half the files are faster, half slower; unaffected by extreme cases
MODT: Memorandum of deposit of title deeds — the registered mortgage in EHFL's favour
NACH: National automated clearing house — the EMI auto-debit mandate
NPA: Non-performing asset — 90+ days past due
OTC: Over the counter — documents to be collected at the branch before disbursement cheque handover
overturn: A file U1 recommended that U2 rejected
p75: 75th percentile — three files in four are at or below this value; the value the slowest quarter exceeds
p90: 90th percentile — nine files in ten are at or below this; the long tail
PDD: Post-disbursement documents — originals (registered deed, mortgage deed, EC, vetting report) due after the money is paid
penny-drop: A one-rupee transfer to verify a bank account and its holder
PMAY: Pradhan Mantri Awas Yojana — the central housing subsidy scheme
QC: Quality check — a desk review between login and U1
RAG: Red / amber / green — the colour on a cell is the rating; thresholds are stated under each table
re-sanction: A sanction letter re-issued with changed terms before disbursal
RM: Relationship manager — the salesperson who owns the file
ROI: Rate of interest
rounds: Separate ADR queries on a loan — each a stop-answer-resume cycle; the count, not the time, which is measured as standing days
rounds per loan: Number of separate ADR queries a loan received, summed over its life — three rounds means the file was stopped and restarted three times
SENP: Self-employed non-professional
soft cancelled: A file closed without a decision — by the RM, or automatically
SUD: Sanctioned undisbursed — sanctioned loans not yet paid out
TAT: Turnaround time — elapsed days (or hours) between two stage timestamps on the same file
TOM: Territory operations manager
U1: First credit underwriter — branch credit manager (BCM) in most regions, or the central CPU for salaried files
U2: Second credit underwriter — regional credit manager
U3: Sanctioning authority — zonal credit head or CRO; sanction is auto-generated on U2 recommendation, U3 signs rejections