Algorithmic decision-making (ADM)
A procedure in which a computer program makes, partially or fully on the basis of data, a decision about people, for example on a loan, a job or a benefit.
In plain terms
We speak of algorithmic decision-making when a program, not a person, weighs the options: it receives your data and outputs a result. This can be yes or no, a score, or a ranking. The program can be a simple set of rules ("if income is less than X, reject") or an AI model trained on past cases.
It is like a tireless administrator who processes thousands of applications in seconds, always using the same criteria. It has no bad days and does not play favorites, but it does not look up from the paperwork to hear why your case is special.
Its advantages are speed, low cost, and consistency. Its disadvantage is that if the rule or training data is flawed or biased, the error repeats in every case, and it is often not visible from the outside on what basis the decision was made.
A key distinction is whether the machine only makes a suggestion that a human approves, or makes decisions without human involvement. The latter is regulated more strictly by law.
How it works
- Data collection: the system gathers available data about you, such as your income, payment history, resume content, or past purchases.
- Rule or model: the basis of the decision is either a rule system written by humans or a machine learning model that learned from patterns in past cases what usually leads to a "good" or "bad" outcome.
- Scoring: the program calculates a value from your data, such as a risk score or a probability.
- Decision: the score is measured against a threshold, and from this comes approval, rejection, ranking, or pricing. This can happen fully automatically, or with a human still taking a look.
- Oversight: in the best case, the decision is justified, you can appeal it, and the system is regularly checked to ensure it does not disadvantage certain groups.
An everyday example
You apply for an online personal loan, and two minutes later the response arrives: rejected. No human could have reviewed your application in that time; a scoring program decided based on the provided data and information held about you. In such cases, you have the right to ask what underpinned the decision and request human review.
Who is affected
A common misconception
Limits and risks
- Bias in training data can lead to discrimination, even if the system does not directly use gender or age.
- Many systems are 'black boxes': even the operator cannot always say precisely why that particular decision was made.
- Human approval easily becomes a formality, because the caseworker tends to accept the machine's suggestion without thinking.
- Legal protection is narrower than many think: the GDPR generally prohibits decisions based solely on automated processing that have a significant effect on you, but there are exceptions (contract conclusion, legal authorization, explicit consent). In such cases, you must request human review, and if a person meaningfully participates in the process, this rule does not apply.
Similar, but not the same
- Decision support system – It only provides information or suggestions; the human weighs and makes the decision. In algorithmic decision-making, the output is essentially determined by the program. The line blurs if the human only formally approves the machine's suggestion.
- Profiling – Profiling is the evaluation and prediction of your personal traits, which may inform a decision, but is not a decision itself.
- AI-agent – An agent independently executes tasks in multiple steps, whereas algorithmic decision-making typically answers a single, well-defined question.
Related concepts
automated credit assessment, algorithmic recruitment, predictive policing, dynamic pricing
algorithmic bias, Profiling, Explainable AI, human oversight, recommendation system
A brief history
Credit scoring mechanized banking decisions in the second half of the 20th century. The EU's data protection regulation (GDPR), adopted in 2016 and applicable since 2018, regulates automated decision-making in a separate article, while the AI Act adopted in 2024 classifies systems used in sensitive areas as high-risk.
Where things stand
In late 2026, it is commonplace in banking, insurance, recruitment, online platforms, and increasingly in public administration. The requirements of the EU AI Act are becoming mandatory gradually, in several stages. The debate focuses mainly on transparency and meaningful human oversight.
In the news
- UK GDPR still applies to AI after the Data (Use and Access) Act, NeuralTrust saysOct 22, 2026 · Jog
- US AI hiring laws and discrimination liability in vendor contractsOct 12, 2026 · Work & HR
- EDPB publishes €824.99 million GDPR fine against Uber; Noyb cites €2 billion in finesOct 12, 2026 · Jog
- US state and federal AI rules collide: Connecticut, Colorado, FTCAug 17, 2026 · Jog
- Colorado proposes automated decision-making rules; ABA calls for federal AI regulationAug 14, 2026 · Money
Also known as
automated decision-making · Algorithm-based decision-making · machine decision-making · ADM · automated decision-making · algorithmic decision-making · automated decision system · ADS
This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.