Decision Models

A decision model is a structured representation of how decisions are made. It lists which input data are needed, which rules apply, and what outcome ultimately results. That sounds abstract, but it is very concrete: a company that issues loans follows a fixed pattern for every request — check income, query credit score, compare thresholds, output a result. This pattern is a decision model. It can exist on paper, in a spreadsheet, in software, or as part of an AI system.

Making decisions visible and controllable

Without such a model, a decision is a black box. No one can trace why a particular outcome resulted. In many areas, this is a serious problem. Banks are legally required to be able to explain why a loan was denied. Insurers must justify why a policy is more expensive. Without an explicit model, this is barely possible.

Decision models solve this problem by extracting the logic from an expert’s head or from inside an algorithm and putting it into a readable form. This makes it possible to check, adapt, and share it with others. This is especially important when many people have to make the same kind of decision every day — for example, caseworkers at a government agency. A shared model ensures that everyone decides the same way, rather than each person following their own discretion.

Structure and representation of a decision model

The most widespread standard is called DMN — Decision Model and Notation. It originates from software development and defines how to write down decision logic in a uniform way. The central tool here is the decision table: a table in which each row describes a combination of conditions, and the last column names the corresponding outcome. For example: age under 25 and no work experience results in “application denied”; age over 30 and three years of experience results in “application approved”.

More complex decisions are broken down into several sub-decisions that build on one another. The result of one sub-decision becomes the input for the next. This keeps even multi-layered logic manageable. It’s important to distinguish this from a process model: a process model describes the sequence of steps — who does what and when. A decision model only describes which outcome follows from which data. The two representations complement each other but are not the same thing.

In AI development, decision models play an increasing role in explainability. Once a model has been trained and makes decisions, one can try, after the fact, to approximate its behavior with a decision model — that is, to map it using simplified, readable rules. The result is never perfect, but it gives developers and regulators a reference point.

Decision models in products and current debates

Decision models are embedded in a wide variety of everyday systems. Credit-scoring apps, automated insurance quotes, applicant tracking systems, and risk checks in online retail all operate according to explicitly or implicitly defined decision models. Streaming services also use variants of this to decide when a reminder is sent or when a subscription may be automatically cancelled.

In the debate over the EU AI Act — the European law regulating AI systems — decision models appear as a tool for transparency and accountability. Systems that automatically make decisions about people are supposed to be documented and traceable. Decision models are one of the answers to this requirement. Anyone wanting to operate an AI system in a regulation-compliant way can hardly avoid clean documentation of the decision logic.

A common misconception is to equate decision models with AI. In fact, they exist completely independently of one another. A decision model can be a pure collection of rules, without any machine learning component whatsoever. And conversely, many AI systems make decisions without any explicit model behind them. The term describes a method for structuring logic — not a specific technology.