ADMT (Automated Decision-Making Technology)

Automated decision-making technology (ADMT) refers to any system that uses computation to replace or substantially replace human judgment in a decision affecting a person, most notably decisions about lending, housing, education, employment, or healthcare.

As organizations lean on algorithms and AI models to make or heavily influence these calls, regulators around the world have moved to require transparency into how the decision was reached and to give individuals a way to understand, contest, or opt out of it.

ADMT has moved from an abstract AI-ethics concept to a defined legal term with concrete compliance obligations in a growing number of jurisdictions. Definitions vary somewhat, but the core idea is consistent:

• a decision qualifies as "automated" when computation replaces or substantially replaces the human role in reaching it, and
• heightened obligations kick in when that decision is "significant," typically meaning it affects someone's access to financial services, housing, education, employment, or healthcare

Most frameworks are careful to exclude things like routine ad targeting from that definition, since the goal is to catch consequential decisions, not everyday personalization.

Effective ADMT governance typically requires three things:

• Pre-use notice: informing individuals before their data is used in an ADMT system, including its intended purpose.
• Right to know and appeal: giving individuals meaningful information about the logic, inputs, and outputs behind a decision, along with a path to opt out or contest it.
• Recordkeeping: maintaining documentation of risk assessments, notices given, and how requests were resolved.

These obligations are showing up across multiple regulatory regimes at once. GDPR Article 22 has long addressed automated decisions with significant effects on individuals, U.S. state privacy laws (including California's CCPA) have recently added their own ADMT-specific rules, and similar provisions continue to emerge elsewhere as AI adoption accelerates.

For organizations already building AI governance programs, ADMT compliance is best treated as a subset of that broader effort. The same underlying question, what data trained or feeds a model and under what permission, applies whether the output is a marketing decision or a lending decision.