▸ Case walkthrough
Case workbench · Case 2
Hamood Al Shukaili
36 · private-sector sales at a trading LLC in Al Khuwair. Declared salary OMR 870. Requests RO 7,500 personal loan.
Applicant panel
| Customer | Hamood Al Shukaili · C100203 |
| Civil | ••••0203 |
| Archetype | Private sector Omani · SME |
| Employer | Independent Trading LLC |
| Employer stability score | 0.41 |
| Area · Governorate | Al Khuwair · Muscat |
| Tenure with bank | 34 months |
| Products held | 2 |
| Relationship value | OMR 980 |
Requested facility
| Product | Personal loan |
| Amount | OMR 7,500 |
| Tenure | 48 months |
| Purpose | Consolidation + liquidity |
| Collateral | None (salary assignment) |
| Instalment at indicative rate | OMR 182 / month |
| Indicative rate | 7.95% |
Decision
DECLINE
Observed cash flow method declines. Three card cash advances in the last 90 days indicate behavioural stress despite the salary meeting policy ratios.
Counter-offer: OMR 2,500 at same rate
Authority level: credit officer
Reason code: C-OBSV-STRESS-3
Affordability · three methods in parallel
1 · Declared method
What most banks do
| Gross income | OMR 870 |
| Policy deductions | OMR 70 |
| Net disposable | OMR 800 |
| Existing obligations | OMR 0 |
| Proposed instalment | OMR 182 |
| DBR | 22.8% |
| Policy ceiling | 50% |
| Approvable amount | OMR 7,500 ✓ |
APPROVE
2 · Verified method
Better · still incomplete
| Observed salary credit | OMR 868 avg |
| Internal obligations (BM systems) | OMR 0 |
| External obligations (Mala'a bureau) | OMR 0 |
| Policy deductions | OMR 70 |
| Net disposable | OMR 798 |
| Proposed instalment | OMR 182 |
| DBR | 22.8% |
| Approvable amount | OMR 7,500 ✓ |
APPROVE
3 · Observed cash flow
The differentiator
| 12m salary credit observed | OMR 868 avg · high variability |
| Timing · arrives 2nd-11th of month | Volatility ×1.3 |
| Card cash advances · 90d | 3 observed |
| Standing order failures | 1 (Feb) |
| End-of-month balance trend · 5m | declining |
| Rent / housing (observed) | OMR 220 |
| Buffer requirement (vol. scaled) | OMR 196 |
| Net disposable (observed) | OMR 382 |
| Approvable (stress markers) | OMR 2,500 |
DECLINE at OMR 7,500 · APPROVE at OMR 2,500
Policy gates · visible checklist
| # | Gate | Rule | Source | Result |
|---|---|---|---|---|
| G1 | Age · min 21, max 65 salaried | Applicant age 36 | Illustrative | PASS |
| G2 | Employment tenure · ≥ 6 months | 34 months at employer | Illustrative | PASS |
| G3 | Mala'a bureau check | Enquiry mandatory · clean record | Confirmed | PASS |
| G4 | IFRS 9 stage · must be stage 1 | Current stage 1 | Assumed | PASS |
| G5 | Debt-burden ratio · ≤ 50% salaried | Declared method: 22.8% | Assumed | PASS |
| G6 | Minimum net after obligations | Observed: OMR 382 vs OMR 200 floor | Illustrative | PASS |
| G7 | Conduct · vulnerability markers | 3 stress markers observed | Assumed | FAIL · suppress |
The rules engine evaluates in order. G1-G6 pass; G7 triggers the conduct suppression, which is captured for audit as a human-reviewable override point.
Risk assessment
| Probability of default (12m) | 4.8% |
| — driver 1: behavioural score | 0.42 |
| — driver 2: Mala'a position | 0.18 |
| — driver 3: internal conduct | 0.28 |
| Loss given default | 82% |
| Exposure at default | OMR 7,500 |
| Expected loss (12m) | OMR 295 |
| Risk grade | B+ |
Pricing & IFRS 9
| Risk-based rate recommendation | 7.95% |
| Floor / ceiling (policy) | 6.00% / 11.50% |
| Funding cost (from Deposit Pricing engine) | 4.15% |
| Operating cost allocation | 1.20% |
| Net interest margin on facility | 2.60% |
| RWA (regulatory) | OMR 11,250 (150%) |
| Return on RWA | 7.1% |
| Stage at origination | Stage 1 |
| 12-month ECL booked | OMR 295 |
| First-year contribution | OMR 285 · after ECL |
36-month cash flow observation
Monthly income & outflow · observed
Salary volatility, cash-advance episodes highlighted
Policy studio
Every credit rule in one place
Versioned, with source and owner. Live impact testing: change a parameter and see how many historical applications flip.
Parameter register
| Parameter | Current | Owner | Effective | Source | Last fired |
|---|---|---|---|---|---|
| DBR ceiling · personal · salaried | 50% | Credit policy | 2024-04-01 | Assumed | today |
| DBR ceiling · housing included | 60% | Credit policy | 2024-04-01 | Assumed | today |
| Max personal loan tenure | 120 months | Product | 2023-01-01 | Assumed | today |
| Max housing loan tenure | 300 months | Product | 2023-01-01 | Assumed | today |
| Interest rate ceiling · personal | 11.50% | Pricing | 2025-07-01 | Assumed | today |
| Salary assignment required | Yes · salaried | Credit policy | 2023-01-01 | Assumed | today |
| Max LTV · first home | 80% | Credit policy | 2024-01-01 | Assumed | today |
| Max LTV · subsequent property | 70% | Credit policy | 2024-01-01 | Assumed | today |
| Min net income after obligations | OMR 200/mo | Conduct | 2025-01-01 | Illustrative | today |
| Min employment tenure | 6 months | Credit policy | 2023-01-01 | Illustrative | today |
| Min age / max at maturity | 21 / 65 (70 pension) | Credit policy | 2023-01-01 | Illustrative | today |
| Mala'a bureau enquiry | Mandatory | Regulatory | 2024-07-01 | Confirmed | today |
| Expat tenure cap · visa validity | Yes | Credit policy | 2023-01-01 | Illustrative | today |
| Dormant · promo rate waiver for Asalah pre-2023 | — | Pricing | 2022-06-01 | Assumed | 24m dormant |
Live impact testing · move DBR ceiling
| Historical applications affected | 12,000 total |
| Approvals that change outcome | — |
| Volume impact | — |
| Expected loss impact | — |
| Most affected segment | — |
Rule conflict detection
⚠ Potential conflict: "Min net income after obligations OMR 200" currently requires manual override for ~1,100 approvable expatriate blue-collar applications per quarter. The expatriate archetype's income minima intersect this ceiling.
Dormant rule: "Promo rate waiver for Asalah pre-2023" has not fired in 24 months. Consider retirement — rules that add friction without managing risk degrade decision speed.
Approval workflow · maker-checker
| # | Change | Owner | Status | Historical outcome change |
|---|---|---|---|---|
| PC-2026-071 | DBR ceiling · housing to 65% | K. Al Rawahi (CCO) | Pending CCO sign-off | +2,840 approvals · +OMR 48m vol · +4 bps EL |
| PC-2026-072 | Mala'a: tiered bureau requirement by amount | Risk | Pending legal | same volume · faster TAT |
| PC-2026-073 | Retire dormant promo rule | Pricing | Ready to deploy | Zero outcome change · simplifies 2 workflows |
Decision analytics
Where approvals and declines come from
Approval rate decomposition and the panel estimating how many declines are policy-driven rather than risk-driven — where the approvable-population argument lives.
Approval rate · last 90d
74.3%
+2.1 pp YoY
Decline rate
18.2%
of which 42% policy-driven
Refer rate
7.5%
to credit officer or committee
Turnaround · median
3h 42m
+18m affordability verification
Decline reasons · ranked
Approvable population: ~42% of DBR-ceiling and tenure-based declines are classified as policy-driven rather than risk-driven. If the observed cash-flow method were applied, 1,842 of these cases would approve at essentially unchanged modelled loss (~82 bps).
Override tracking · analyst departures from system
| Override direction | Count 90d | Subsequent 90d default rate | Signal |
|---|---|---|---|
| System DECLINE → human APPROVE | 142 | 3.8% | adds value |
| System APPROVE → human DECLINE | 38 | n/a | conservative |
| System APPROVE → human CONDITION | 410 | 0.9% | appropriate |
Human APPROVE on system DECLINE is value-adding (3.8% default vs 8-11% model prediction). Human DECLINE on system APPROVE is where governance looks.
Decision consistency monitor
Pairs of near-identical applications flagged where the system reached different outcomes. Surfaced monthly to Chief Credit for review.
| Pair | Delta | Outcome diff |
|---|---|---|
| App-2601 / App-2874 | Same employer, DBR ±0.4pp | APPROVE / DECLINE |
| App-2914 / App-3002 | Same income, tenure diff 1m | APPROVE / REFER |
| App-3104 / App-3188 | Same product, nationality diff | APPROVE / CONDITION |
Fairness monitor
Approval rates by governorate, gender, age band, nationality and employer category — controlling for income and employer. The monitor detects statistical disparity, not discrimination. Any finding requires human investigation.
| Dimension | Group | Approval rate | Uncontrolled Δ | Controlled Δ | Status |
|---|---|---|---|---|---|
| Governorate | Dhofar | 72.1% | −2.2 pp | −0.4 pp | below threshold |
| Governorate | Ash Sharqiyah | 73.8% | −0.5 pp | +0.1 pp | below threshold |
| Gender | Female | 75.1% | +0.8 pp | +0.2 pp | below threshold |
| Nationality | Expatriate (all) | 68.4% | −5.9 pp | −2.1 pp | above threshold · investigate |
| Age | 55-65 | 71.2% | −3.1 pp | −0.8 pp | below threshold |
Finding: expatriate decline rate 2.1 pp above threshold after controlling for income and employer. Likely attributable to visa-validity tenure cap. Investigation ticket raised. No automatic policy change.
Portfolio monitor
Composition · vintage · migration
Portfolio composition by risk grade, product, segment and vintage. Vintage analysis exposes loosening of standards month by month.
NPL ratio
3.4%
stable 90d
Provision coverage
170%
+8pp YoY very conservative
Capital adequacy
18.7%
−0.3pp drifted from 20%
Vintage loss curves · by origination quarter
Cumulative losses at month-on-book
Loosening of standards shows up as worse curve
Vintages Q1-Q3 2024 within model band. Q4 2024 cohort elevated at MOB 12-15 — oil-sector contractor concentration (Galfar-type). Q1-Q2 2025 cohorts back within band after segment-tightening.
IFRS 9 stage distribution
| Stage 1 · performing | 94.1% · OMR 7.42bn |
| Stage 2 · SICR | 4.5% · OMR 352m |
| Stage 3 · credit-impaired | 1.4% · OMR 108m |
| Migration · S1→S2 last month | OMR 42m |
| Migration · S2→S1 (cure) | OMR 28m |
| Migration · S2→S3 last month | OMR 11m |
Concentration · by employer
| Government ministries | OMR 2.1bn · 26.7% |
| Royal Oman Police / Army | OMR 980m · 12.4% |
| Petroleum Development Oman | OMR 540m · 6.8% |
| Omantel / Ooredoo | OMR 380m · 4.8% |
| Galfar + construction contractors | OMR 184m · 2.3% watch |
| Teacher / health sector | OMR 420m · 5.3% |
Recovery & collateral
| Area | Secured exposure | Valuation vintage | Avg LTV | Loss given default at base | LGD if valuations drop 20% |
|---|---|---|---|---|---|
| Muscat · Al Khuwair & Qurum | OMR 648m | 2025 revaluation | 64% | 24% | 32% |
| Muscat · Sultan Haitham City | OMR 184m | 2026 at origination | 78% | 28% | 41% |
| Al Batinah · Sohar | OMR 142m | 2024 | 62% | 26% | 34% |
| Dhofar · Salalah | OMR 68m | 2023 | 58% | 22% | 28% |
IFRS 9 impact
Growth and provision · same screen
Because every approval books expected credit loss on day one, the credit decision and the IFRS 9 consequence belong on the same screen.
12m ECL
OMR 142m
on Stage 1 book
Lifetime ECL
OMR 182m
S2+S3
Impairment charge · H1
−20% YoY
cost of risk falling
Forward-looking macroeconomic scenarios
| Scenario | Weight | Oil | GDP non-oil | Unemp. | ECL | Δ vs base |
|---|---|---|---|---|---|---|
| Base | 0.50 | USD 68 | +3.1% | 0 | OMR 142m | — |
| Upside | 0.25 | USD 82 | +4.6% | −0.8pp | OMR 128m | −10% |
| Downside | 0.25 | USD 50 | +1.2% | +1.5pp | OMR 162m | +14% |
| Stress · oil at 45 | — | USD 45 | −1.8% | +2.4pp | OMR 184m | +30% |
ECL sensitivity to oil price
Dominant Omani macro driver
Regulatory impairment reserve
| IFRS 9 provision (ECL) | OMR 182m |
| Prudential requirement (coverage) | OMR 310m |
| Shortfall | OMR 128m |
| Appropriation from net profit | OMR 128m to equity reserve |
| Coverage ratio | 170% |
| If coverage moved toward 130% | releases OMR 72m RWA |
Explicit as a question requiring CBO engagement — not a model recommendation.
Islamic (Meethaq) ECL treatment
Islamic portfolio handled separately with the SPPI classification question (solely payments of principal and interest). Mudaraba and Wakala structures do not pass SPPI and are measured at fair value through profit or loss (FVTPL). Diminishing Musharakah and Murabaha are SPPI-compliant and follow IFRS 9 impairment logic.
Interface note: SPPI classification follows Sharia Supervisory Board terms. The ECL computation operates within those terms and does not opine on the Sharia question.
Audit & explainability
Every decision reconstructible
Any decision reproducible exactly, with the data, model version, policy version and parameter set as they stood at decision time. Must be built from the start — cannot be retrofitted.
Per-decision explanation · three registers
Customer-facing · plain language
"Thank you for your application. Based on our review of your account activity over the last 12 months, we are able to approve OMR 2,500 of your request. Three indicators suggested that a higher amount could create pressure on your monthly commitments. If you would like to discuss, please contact your branch."
Analyst view · structured
| Observed stress markers | +0.42 to decline |
| Salary volatility | +0.18 |
| Clean Mala'a | −0.14 |
| Employer stability (low) | +0.11 |
| Tenure with bank (34m) | −0.08 |
| Net score | +0.49 · decline |
Technical · computation trace
model_version: SCORE-v4.2.1
policy_version: POL-v2.14
dataset_version: v2026.08.14
seed: 73019
features: 47
inference: 142ms
policy_gates: 7 evaluated, 6 pass, 1 fail (G7)
decision_path: AFFORD_OK → POLICY_OK → CONDUCT_FAIL
audit_hash: a9f3e2c4b8d1
policy_version: POL-v2.14
dataset_version: v2026.08.14
seed: 73019
features: 47
inference: 142ms
policy_gates: 7 evaluated, 6 pass, 1 fail (G7)
decision_path: AFFORD_OK → POLICY_OK → CONDUCT_FAIL
audit_hash: a9f3e2c4b8d1
Model register
| Model | Version | Purpose | Trained | Pop | Validation | Approval | Review |
|---|---|---|---|---|---|---|---|
| SCORE-behavioural | v4.2.1 | Behavioural scoring | 2026-05 | 184k | AUC 0.84 | approved | 2026-07 |
| ECL-engine | v4.1.0 | PD / LGD / ECL | 2026-04 | 250k | MAE 7.8bps | approved | 2026-04 |
| AFFORD-observed | v3.0.2 | Observed cash flow affordability | 2026-06 | 12k | backtest 8.4% MAE | pilot | — |
| ELAST (Deposit Pricing) | v3.2.0 | Rate elasticity | 2026-05 | 82k | R² 0.71 | pilot | — |
| TRIG (Trigger Lending) | v2.1.0 | Event detection | 2026-07 | 250k | Precision 0.82 | pilot | — |
Full reconstruction · auditor view
Any decision reproducible exactly 18 months later. Re-run with the data, model and policy versions as of decision time → same answer. This is not a retrofit — the audit hash is created at origination.
Case: Hamood Al Shukaili · App-20260812-C100203
Decision: DECLINE at OMR 7,500 · counter-offer OMR 2,500
Audit hash: a9f3e2c4b8d1 · timestamp: 2026-08-12 14:21:08 OST
Reproduce: /audit/reproduce?hash=a9f3e2c4b8d1 → re-runs model v4.2.1 against snapshot
Snapshot age: 2 days · retention 7 years
Decision: DECLINE at OMR 7,500 · counter-offer OMR 2,500
Audit hash: a9f3e2c4b8d1 · timestamp: 2026-08-12 14:21:08 OST
Reproduce: /audit/reproduce?hash=a9f3e2c4b8d1 → re-runs model v4.2.1 against snapshot
Snapshot age: 2 days · retention 7 years
Early warning
Watchlist · lead time measured
Behavioural markers that an existing-customer relationship makes visible, each with its lead time to delinquency measured from the dataset.
Watchlist · ranked by stage migration probability
| # | Risk cluster | Exposure | Facilities | S2 probability | Lead marker | Lead time | Action |
|---|---|---|---|---|---|---|---|
| 1 | Galfar employees · salary late 2 of 6m | OMR 48m | 184 | 21% | Late salary + overdraft excess | 2.4m | RM touch · suppression on new credit |
| 2 | Rising card cash-advance usage | OMR 32m | 412 | 14% | 3 cash adv. in 90d | 3.1m | watch |
| 3 | Standing order failures | OMR 18m | 94 | 24% | Multiple SO fails | 1.9m | immediate RM touch |
| 4 | Overdraft excess days ↑ | OMR 24m | 278 | 17% | Excess days 5+ for 2m | 2.8m | counsel |
| 5 | Dormancy onset | OMR 42m | 318 | 11% | Txn frequency ↓ 60% | 4.2m | watch |
| 6 | Salary diversion to another bank | OMR 12m | 52 | 31% | Transfer to competitor | 1.4m | retention team |
| 7 | End-of-month balance trending down | OMR 68m | 820 | 9% | 5-month decline | 5.1m | monitor |
Employer concentration shock · simulated
Scenario: a large contracting employer stops paying salaries for two months. Shows the exposure affected, the stage migration, the provision, and the early-warning lead time the behavioural markers would have given.
Exposure affected
OMR 48m
184 personal loans
Stage migration S1→S2
OMR 36m
75% of affected book
Provision increase
OMR 2.8m
immediate · lifetime ECL
Lead-time EWS would have given
2.4 months
late/short salary marker
Mitigating actions available in lead time
OMR 18m
pre-emptive restructure
Residual provision after mitigation
OMR 1.4m
−50%
What-if simulator · two levels
Single-case & portfolio
Single-case wins an analyst. Portfolio wins a CFO.
Portfolio controls
On · recommended
Approval rate · modelled
74.3%
at base
Incremental volume
OMR 0
vs current policy baseline
Expected loss (bps)
82 bps
at base
Net contribution
OMR 0
after funding + ECL + capital
Approval rate & expected loss · side by side
Segments that gain / lose access · named
Gain access
- Teachers · existing-customer, income OMR 700-1,400 (+28%)
- Pensioners · ROP / Royal Army (+22%)
- Mid-income Omani women · joint affordability (+19%)
- Government mid-career · 1,800+ income (+14%)
Lose access
- Private-sector salaried · elevated stress markers (−12%)
- Self-employed · thin file & cash-flow volatile (−8%)
- Expats · short visa validity & new tenure (−5%)
Conversational analyst
Explains · reconstructs · refuses
Will not decide a human judgement case. Will reconstruct any past decision. Will tell you when it cannot answer.
AI
Credit Decisioning analyst
Claude · Policy, Afford, ECL, EWS engines wired
When the chatbot refuses
- Human judgement cases (Case 7 pattern) → declines to decide, surfaces policy vs conduct conflict
- Regulatory compliance claims → never asserts a position; names the rule
- Sharia opinions → never issues; references the Sharia Supervisory Board
- Figures the dataset cannot support → declines rather than fabricate
- Actions breaching configured guardrails → refuses, names the guardrail
Counterfactual explanation
Any decline comes with a counterfactual: the minimum change that would flip the outcome.
"Decline would flip to APPROVE at OMR 6,200 (instead of OMR 8,000)" or "APPROVE if existing card balance were cleared before this application"