Credit decisioning rebuild: from manual underwriting to automated retail decisions

Context

Retail bank UZ. Consumer loan approval 5 days average. Underwriter team 30 people, capacity bottleneck. NPL ratio 4.2% (industry 2.8%). Fintech competitors give instant decisions.

CRO + Head of Lending — rebuild credit decisioning.

Diagnostic (5 weeks)

• Retail underwriting — manual, inconsistent quality across underwriters.
• Models exist but not in production decision (advisory only).
• Data sources fragmented — credit bureau call manual.
• Audit trail incomplete for regulator.
• NPL by segment: highest among underbanked first-time borrowers (12%).

Approach (10 months)

Months 1-3. Architecture + decisioning platform selection.

Months 3-5. Model rebuild. PD, LGD, EAD models for retail. ML-augmented with traditional bureau data.

Months 5-7. Decisioning platform launch. Real-time scoring + policy engine + automated decision.

Months 7-9. Migration of retail loan application to platform. Underwriter team transition to exception cases only.

Months 9-10. Optimisation, false positive tuning, edge case workflow.

Results

After 10 months:

• Retail approval time: 5 days → 12 seconds.
• Approval rate: 38% → 52% (better discrimination, not permissiveness).
• Retail NPL rate: 4.2% → 2.9% (12-month cohort).
• Underwriter team capacity freed: 70% redirected to SME / corporate.
• Customer NPS in retail loan applicants: +27.
• Regulator audit: passed first try.

What is critical

Model bias monitoring mandatory — disparate impact analysis caught early issues.

Underwriter team buy-in — they participated in edge case design.

Regulator dialogue continuous — no surprise.

Live champion/challenger — model continuously improved.

Related

• /en/solutions/banking-credit-decisioning-platform/ — credit platform
• /en/architecture/banking-mlops-architecture/ — MLOps
• /en/insights/banking-credit-risk-management/ — credit risk
• /en/expertise/banking-credit-discipline/ — credit discipline