A credit policy is a document. A credit scorecard is a decision engine. The Credit Decision Agent AI converts the institution's written credit policy — every rule, every threshold, every exception pathway — into an executable scorecard that runs 40 checks in parallel, produces a decision in under 90 seconds, and generates a documented rationale for every outcome. The underwriter's job begins after the scorecard, not before it.
Why manual credit policy execution fails at scale
A written credit policy contains 40 or more rules across six dimensions: borrower eligibility, income and FOIR, bureau and credit history, collateral and LTV, product-specific parameters, and fraud and compliance gates. In a manual underwriting environment, these rules are executed sequentially by a human underwriter — reading the policy, pulling the relevant data from the file, applying the rule, and moving to the next. At a pace of one rule application every 90 seconds, the manual scorecard takes over an hour per file.
More critically, manual execution is inconsistent. A senior underwriter and a junior underwriter applying the same 40-rule policy to the same file may reach different conclusions — not because the policy is ambiguous, but because human application of a rule set introduces interpretation variance. Rules that say "FOIR should not exceed 45%" are applied consistently. Rules that say "income stability should be adequate" produce variance. The Credit Decision Agent AI eliminates this variance by converting every rule into a precise, parameterised check that applies the same threshold to every file, every time.
"The credit policy manual says what the rules are. The Credit Decision AI does what the rules require. Those are different capabilities — and only the second one scales."
The 40 rules — organised by category and outcome type
Category A — Borrower Eligibility Gates
8 rules · Type: Hard knock-out where failed
01
Minimum age 23 years / Maximum age 65 years at loan maturityExtracted from Aadhaar DOB. Maximum age at maturity computed from age + tenure at application. Both must be within bounds for the product.Knock-out: age outside bounds → immediate decline · Product recommendation engine triggered for age-eligible product
KO
02
Minimum employment tenure 6 months at current employerVerified from EPFO contribution history and salary slip sequence. New employer within 6 months is a knock-out unless prior employer had 24+ months.Knock-out: current employer <6 months AND prior employer <24 months → decline
KO
03
Minimum eligible income — product-specific floorEligible income (from Bank Statement AI) must meet the product floor: ₹25,000/month for personal loans, ₹40,000/month for home loans. Computed after applying income category haircuts.Knock-out: eligible income below product floor → decline · Alternative product with lower floor triggered
KO
04
Negative list clean — internal and externalApplicant name, PAN, Aadhaar, and phone number checked against: institution's internal NPA list, CIBIL defaulter list, RBI Caution list, and CERSAI wilful defaulter list. Any match is a knock-out.Knock-out: any match on any list → immediate decline · Fraud team notified if RBI Caution or wilful defaulter match
KO
05
KYC completion — all mandatory elements verifiedAadhaar OTP confirmation, PAN NSDL verification, and CKYC registry check must all return verified status. Any unresolved KYC exception is a knock-out until the KYC exception is resolved.Knock-out: unresolved KYC exception → application held, not declined · KYC resolution pathway triggered
KO
06
Restricted employment sector / categoryCredit policy excludes certain employer categories from eligible borrower pool: casinos, illegal activities, politically sensitive sectors per the institution's policy. Employer sector verified against the restricted list.Knock-out: employer in restricted category → decline · Policy reference logged
KO
07
Geographic restriction — institution's serviceable pin codesProperty address (for secured loans) or residence address (for unsecured) must fall within the institution's approved serviceable geography. Pin code verified against approved list.Knock-out: address outside serviceable geography → decline with geographic note
KO
08
Product overlap — maximum open loan count with institutionApplicant must not exceed the institution's maximum concurrent loan limit (product-specific). An applicant with 2 active home loans applying for a third home loan is declined on overlap grounds regardless of credit quality.Knock-out: at maximum concurrent loans for product → decline
KO
Category B — Income and FOIR Rules
7 rules · Mix of knock-out and scorecard
09
FOIR at proposed loan amount ≤ 45%FOIR computed: (existing monthly EMI obligations + proposed new EMI) ÷ eligible monthly income. Must not exceed 45% for salaried, 50% for self-employed home loans. Primary income policy gate.Knock-out: FOIR >55% → decline · Refer: 45–55% → human review required · Pass: <45%
KO + Ref
10
Income stability score ≥ 60 (from Bank Statement AI)The income stability score from the Bank Statement Analyst AI must meet the product minimum. Accounts with high income variance, frequent gaps, or downward trends receive lower stability scores.Score contributes to weighted scorecard · Score <40 → refer · Score <25 → decline
Score
11
Maximum loan amount ≤ computed eligibilityRequested loan amount must not exceed the maximum eligibility computed from FOIR, income, and LTV constraints. Requests above maximum eligibility are declined or counter-offered at maximum eligible amount.Knock-out: requested amount >max eligibility → counter-offer at maximum, not decline
KO/Counter
12
Income source diversity — single-source dependency riskFor large loans, applicants with 100% income concentration in a single employer are assessed for employer stability. Public sector / MNC employers score positively; startups with <3 years in business score negatively.Score: employer stability tier adds ±5 points to weighted score
Score
13
Declared income vs bank statement corroborationDeclared income on application form must be within 15% of bank statement computed eligible income. Significant divergence (upward declaration) is a fraud and data quality flag.Divergence >15%: soft flag · Divergence >30%: hard flag and refer · Divergence >50%: decline
Score
14
Savings rate ≥ 15% of eligible incomeMonthly average surplus as a percentage of eligible income should be at least 15% — demonstrating that the borrower is not fully consuming their income and has capacity to absorb the additional EMI without financial stress.Score: savings rate 15%+ adds 5 points · <10% subtracts 5 points
Score
15
Income trend direction — 12-month trajectoryIncome trajectory over the 12-month statement period: improving, stable, or declining. Improving income (10%+ growth) adds to the scorecard; declining income subtracts. A 20%+ income decline in the last 3 months triggers a refer regardless of score.Score: improving +5 · stable 0 · declining −5 to −10 · sharp decline → refer
Score + Ref
Category C — Bureau and Credit History Rules
8 rules · KO for severe defaults, scored for history quality
16
CIBIL score ≥ 650 (hard floor)Any CIBIL score below 650 is a knock-out for all standard products. Scores 650–699 enter the scorecard with a negative weight. Scores 750+ are a positive scorecard contributor.Knock-out: CIBIL <650 → decline · Score <700 → negative weight · Score >750 → positive weight
KO + Score
17
Zero DPD events in last 12 monthsAny DPD event (on any account) in the last 12 months is a negative scorecard factor. Any DPD event of 60+ days in the last 12 months is a refer trigger. Any DPD of 90+ days in the last 12 months is a knock-out.Knock-out: DPD 90+ in 12 months → decline · Refer: DPD 60+ · Score: DPD 30 = −8 points
KO + Ref + Score
18
No written-off or settled accounts in last 3 yearsAny account showing "written off" or "settled" status in the last 3 years is a knock-out. Older write-offs (3–7 years) are a negative scorecard factor. Beyond 7 years: no penalty.Knock-out: write-off/settlement <3 years · Score: 3–7 years = −10 points
KO + Score
19
Enquiry count ≤ 4 in last 6 monthsMore than 4 credit bureau enquiries in the last 6 months suggests aggressive credit seeking. 5–6 enquiries: soft negative. 7+: refer. The pattern of enquiries (all for one product vs multiple types) is also assessed.Score: 4 enquiries = −5 points · 7+ = refer trigger
Score + Ref
Categories D, E, F — Collateral, Product, and Fraud Gates
17 additional rules · LTV, property, fraud signals, compliance
20–26
Collateral and LTV rules (7 rules)Maximum LTV by product (75% home loan, 65% LAP). Property type eligible (residential over commercial). Property age within limits. Valuation recency (within 12 months). CERSAI charge check. Legal title clear. Construction stage for under-construction properties.Rules 20–22 are KOs · Rules 23–26 are scored
Mix
27–33
Product-specific parameter rules (7 rules)Minimum loan amount (₹5L). Maximum loan amount (₹3Cr for retail home loan). Tenure within policy range. Floating vs fixed rate eligibility. Co-applicant requirement for amounts above threshold. Top-up eligibility (existing borrower only). Balance transfer eligibility.Knock-outs for min/max amounts · Scored for other parameters
Mix
34–40
Fraud and compliance gates (7 rules)Document authenticity score ≥70 (from Document Verification AI). Synthetic identity score <45. Device fingerprint clean. Application data consistency check. AML screening clear. Watchlist screening clear. Regulatory compliance check (RBI Fair Practices Code, DPDP Act data consent).Rules 34–36 are KOs on hard failure · Rules 37–40 generate refers on flags
KO + Ref
The parallel execution: all 40 rules simultaneously, not sequentially
In a manual underwriting process, rule 17 cannot be checked until rule 16 has been retrieved and rule 1 has confirmed the applicant is eligible. The Credit Decision Agent AI runs all 40 rules simultaneously — each rule receives its required inputs from the pre-computed data assembled by the Loan Origination AI, Bank Statement AI, Document Verification AI, and Bureau Pull — and all 40 rules complete within the same 90-second window.
Parallel rule execution — Application LA-2025-8841 · 90-second total decision time
Rules 01–08 · Eligibility
12s
Rules 09–15 · Income/FOIR
18s
Rules 16–19 · Bureau
23s
Rules 20–26 · Collateral
34s
Rules 27–33 · Product
8s
Rules 34–40 · Fraud/Compl.
41s
All rules passed
KO triggered
Refer triggered
Not applicable
Total elapsed: 41 seconds (bound by slowest API · AML check)
The decision output: what the scorecard produces
Credit Decision Output — Application LA-2025-8841 · Priya Ramachandran Sharma
Home Loan ₹28L · 20yr · Decision computed: 41 seconds · Nov 14, 2025 · 09:14:13
Rules 01–08: All eligibility gates passed · Age 36 (within 23–65 range) · Employer tenure 18 months · Income above floor · Negative list clean · KYC verifiedKO: PASS
Rule 09: FOIR at proposed EMI — 38.4% · Below 45% ceiling · Primary income gate: PASSEDKO: PASS
Rules 10–15: Income scorecard — stability 91, savings rate 21%, bank corroboration exact, income trend +37% over 12 months · Net income score: +14 points+14 pts
Rule 16: CIBIL 736 · Score band B+ · Trend improving (+18 pts in 12 months) · Net bureau score: +8 points+8 pts
Rule 17: 0 DPD events in 24 months · 1 × 30-DPD 28 months ago (outside 24-month window, noted) · DPD score: clean0 pts
Rules 18–19: No write-offs or settlements · 3 enquiries in 6 months (within 4-limit) · Net bureau history: +3 points+3 pts
Rules 20–26: LTV 68.2% (within 75% ceiling) · Property: residential, age 8 years, legal title clear · CERSAI clear · Valuation Oct 2025 (current) · Collateral score: +12 points+12 pts
Rules 27–33: Loan amount ₹28L within range (₹5L–₹3Cr) · Tenure 20yr within policy · No co-applicant required at this amount · Product score: 0 adjustment0 pts
Rules 34–40: Document authenticity 96.8 · Synthetic identity score 8 (low risk) · Device clean · AML clear · Watchlist clear · Compliance: all clear+5 pts
Weighted scorecard total
87 / 100
Pass threshold: 65 · Refer zone: 55–64
Decision
AUTO-APPROVE
0 KO failures · 0 refer triggers · Score 87
● 40 rules executed · Decision time: 41 seconds · No KO failures · No refer triggers
● Sanction letter generation queued · Underwriter notified for sign-off · Application LA-2025-8841
40Rules executed in parallel — KO gates, scorecard factors, and refer triggers all run simultaneously
41sTotal decision time — bound by the slowest external API call (AML check), not by rule complexity
87Scorecard total for this application — against pass threshold of 65 and refer zone of 55–64
ZeroInterpretation variance — the same 40 rules, applied to the same file, produce the same score every time
A scorecard is only as consistent as its execution
The institution's credit policy is only worth what it produces in outcomes. A policy manual read and applied differently by 20 underwriters across 400 files per day is not a credit policy — it is 20 different credit policies operating under the same name, producing outcomes that reflect individual underwriter judgment more than institutional risk appetite. The Credit Decision Agent AI ensures that the policy produces what the policy committee intended: the same decision criteria applied consistently to every application, with every rule documented in the decision record, making the scorecard auditable, supervisable, and improvable based on actual outcome data rather than underwriter opinion.