AI/ML strategy: in-house or vendor capability

When the fork appears

Multiple AI use cases — credit scoring, fraud detection, NBA, AML, chatbot. Build in-house team (data scientists, ML engineers) or buy vendor models per category.

Banking-specific criteria

Regulatory traceability. Models must be explainable per cbu.uz. Vendor models — need access to internals, no black box.

Bias monitoring. Disparate impact mandatory for credit. Vendor must support bias dashboards.

Model risk management. Banks under formal model validation requirements (Basel-style). In-house easier to govern.

Data residency. PII in country — vendor must process locally or sufficient anonymisation.

Regulatory approvals. New ML model for credit may require regulator review.

Talent availability. Local data science talent constrained — in-house hard to scale.

When in-house

Strategic ambition: data-driven bank.

Capital for team build (10-50 ML people).

Strong existing data foundation.

5+ year horizon.

Regulator engagement preferred direct.

When vendor

Limited talent availability.

Standard use cases (vendor offerings mature).

Time to market priority.

Can negotiate vendor for explainability, audit access.

When hybrid (most common)

In-house for strategic models (credit decisioning, fraud).

Vendor for commodity capabilities (chatbot, OCR, generic analytics).

In-house Centre of Excellence with vendor augmentation.

Where decisions usually go wrong

In-house without realistic talent assessment. Two years later team under-staffed.

Vendor model for credit without access to internals. Regulator audit fails.

Hybrid without clear governance — modes conflict.

Related

• /en/architecture/banking-mlops-architecture/ — MLOps
• /en/insights/banking-ai-governance/ — AI governance
• /en/insights/banking-model-risk-management/ — model risk
• /en/expertise/banking-data-discipline/ — data discipline