PeopleNotResumesBook a Compliance Score
← Back to white papers

What Is an Automated Employment Decision Tool (AEDT)? Definition, Examples, and What Counts

Summary: An automated employment decision tool (AEDT) is any computational process, built from machine learning, statistics, data analytics, or AI, that produces a simplified output such as a score, ranking, or recommendation, and that substantially assists or replaces human discretion in a hiring or promotion decision. The term comes from New York City's Local Law 144. Whether a specific tool qualifies depends on two things: how it works, and how much weight its output carries in the decision.

Key takeaways

  • An AEDT is defined by function, not by branding. A tool does not need to be marketed as AI to qualify.
  • Two tests decide it: the tool produces a simplified output (score, class, or recommendation), and that output substantially assists or replaces a discretionary decision.
  • Resume rankers, scored assessments, and video interview scorers usually qualify. Manual spreadsheets and simple keyword filters usually do not.
  • Most hiring stacks contain at least one AEDT the company had not classified as one.
  • Getting the classification right is the first and most consequential step in any compliance program.

Why the definition matters

Every obligation under Local Law 144, the bias audit, the public disclosure, the candidate notice, only applies if a tool is an AEDT. Classification is therefore the hinge on which the whole compliance question turns. Misclassify a tool as out of scope and you carry silent risk. Over-classify and you spend audit budget you did not need to spend. Getting this right is where a compliance program either starts on solid ground or quietly goes wrong.

The formal definition

Under Local Law 144, an AEDT is a computational process derived from machine learning, statistical modeling, data analytics, or artificial intelligence that issues a simplified output, including a score, classification, or recommendation, and that is used to substantially assist or replace discretionary decision making for employment decisions that impact people.

Break that into two parts and it becomes much easier to apply.

Part one: what the tool produces

The tool has to generate a simplified output. A score, a rank, a pass or fail classification, a match percentage, or a shortlist recommendation all qualify. If the tool simply stores information or displays raw data without reducing it to a judgment, it is less likely to meet this part of the test.

Part two: how the output is used

The output has to substantially assist or replace discretionary decision making. This is the part companies misread most often. A tool still substantially assists a decision even when a human signs off, if the human is relying on the tool's output to decide who advances. The law anticipated the rubber-stamp scenario, where a recruiter approves whatever the model surfaces. That is still substantial assistance.

Examples of what usually counts

  • Resume screeners that rank or score applicants. These produce a simplified output and directly shape who advances.
  • Assessment platforms that score candidates against a trained model. Personality, cognitive, or skills assessments that generate a fit score typically qualify.
  • Video interview tools that evaluate responses. If the platform scores answers, expressions, or language, it is producing a simplified output used in the decision.
  • Matching or sourcing engines that generate a shortlist. A recommendation of the top candidates is a simplified output.
  • Chatbots that screen out applicants based on responses. If the conversation gates who proceeds, it can qualify.

Examples of what usually does not count

  • Manual spreadsheets a recruiter updates by hand. No computational process derived from machine learning or analytics is generating the output.
  • Basic keyword filters that only show or hide records. A simple must-have filter that does not score or rank is generally outside the definition, though this gets closer to the line as filters grow more sophisticated.
  • Scheduling, document management, and communication tools. Software that supports the process without evaluating candidates is not making a decision.

The gray zone sits between a simple filter and a scored ranking. When a filter starts weighting factors and producing an ordered list, it is behaving like an AEDT even if the vendor never uses that word.

Is your applicant tracking system an AEDT?

An applicant tracking system, or ATS, is not automatically an AEDT. Many are primarily systems of record that store applications and move them through stages. The question is whether any feature inside the ATS scores, ranks, or recommends candidates. Modern platforms increasingly bundle matching, ranking, or knockout scoring, and any of those features can bring the ATS into scope even if the core system would not.

This is why an inventory should look at features, not products. A single platform can contain both in-scope and out-of-scope functionality, and turning off or documenting the scoring feature is sometimes the cleanest path.

How to classify your tools

Start with a full inventory of everything in your hiring stack, including tools individual teams adopted without central approval. For each one, ask the two questions in order. Does it produce a simplified output such as a score, rank, or recommendation? If yes, is that output substantially assisting or replacing a human decision about who advances? A yes to both means the tool is in scope and needs a bias audit, disclosure, and notice.

Document your reasoning for each tool, including the ones you classify as out of scope. If a regulator or a candidate ever questions a decision, a written rationale is far more defensible than a verbal judgment call.

Frequently asked questions

Does a tool need to use AI to be an AEDT? No. The definition includes statistical modeling and data analytics, not just machine learning or AI. Branding is irrelevant. Function is what counts.

If a human makes the final call, is the tool still an AEDT? Often yes. If the human relies on the tool's score or ranking to decide, the tool is substantially assisting the decision even with a person in the loop.

Is a simple keyword filter an AEDT? Usually not, if it only shows or hides records without scoring or ranking. As filters grow more sophisticated and start weighting or ordering candidates, they move toward the definition.

Can one platform be partly in scope? Yes. Evaluate features rather than whole products. A system of record with an optional matching score can be in scope only for that feature.

What happens if we classify a tool incorrectly? Missing an in-scope tool leaves you exposed to daily penalties and possible discrimination claims. A documented classification process is the best protection.

Classification is the step that quietly determines whether a compliance program succeeds. If you want a defensible inventory that names every in-scope tool and the reasoning behind each call, that is the first thing a good compliance check produces.