Credit Risk Is No Longer Just a Risk Problem — It’s an Architecture Problem

Building a sound Decision Model and Notation (DMN) model to support a core customer service or operational process is an enormous responsibility. These models encapsulate critical business logic that directly influences decisions, customer outcomes, and overall organizational performance.

This is not due to lack of effort—on either side.

Banks struggle with complexity, legacy systems, and interpretation challenges. Regulators face heterogeneous implementations, inconsistent outcomes, and limited visibility into how rules are actually applied.

The result?

• Delayed alignment;
• Divergent practices across institutions;
• Frustration—for both supervisors and banks.

The Core Issue: Interpretation → Translation → Implementation

Today’s regulatory process typically follows this path:

At every step:

• ambiguity can creep in;
• interpretations diverge;
• implementation slows down.

By the time everything is live:

• the regulatory context may already have evolved;
• inconsistencies remain across institutions.

A Missed Opportunity: Supervising Outcomes versus Supervising Logic

Supervision today is still largely focused on outcomes (e.g. Stage 2 volumes, NPE ratios), documentation (policies, procedures), and on ex-post reviews. But the real leverage point sits elsewhere: the decision logic that drives those outcomes.

Questions like:

• What exactly triggers a Stage 2 classification?
• How is “significant increase in credit risk” operationalized?
• When is forbearance granted—and under which conditions?
• How are early warning signals combined and weighted?

These are decision problems, not just policy statements. Even with detailed guidance, regulators face structural limitations:

• Policies are written in natural language → open to interpretation.
• Banks implement in code → opaque and difficult to compare.
• Audits rely on sampling and documentation → not scalable.

This creates a persistent gap. Regulators define intent, but have limited visibility into how that intent is executed in practice.

The Shift: From Text-Based Guidance to Executable Decision Models

This is where decision modeling—specifically Decision Model and Notation (DMN)—can fundamentally change the dynamic. Instead of only publishing written guidelines, regulators could define standardized decision models alongside regulatory texts.

These models would:

• Translate regulatory intent into explicit decision logic;
• Be human-readable for supervisors and banks;
• Be machine-executable for implementation.

In other words: the same artifact becomes policy, documentation, and execution blueprint.

How could this look like in practice? Well, imagine for instance a regulatory “digital companion” to existing guidelines, containing:

• standardized models, that clearly define inputs, structure the decision logic for implementation, and remove ambiguity in interpretation for key areas such as stage allocation (Stage 1 / 2 / 3), early warning indicators, forbearance classification or UTP identification.
• a supervisory digital hub, i.e. a shared platform where regulators publish versioned decision models, where banks can import, test, and adapt them, and where changes are tracked and compared. This would enable faster alignment when guidelines evolve, transparent comparison across banks, and reduce interpretation risk.
• simulation and impact analysis: before implementation, both regulators and banks could simulate portfolio impacts, test edge cases and assess proportionality. Shifting the conversation from “Did you interpret the rule correctly?” to “Does your implementation produce the intended outcomes?”.

The Real Shift: From Static Compliance to Dynamic Alignment

What this enables is a fundamentally different model:

Old world:
Write → Interpret → Implement → Audit

New world:
Model → Share → Execute → Monitor → Adapt

This is not about removing flexibility. It’s about structuring it, making it transparent and aligning it faster.