Stockertim Dictionary · Artificial intelligence
AI impact assessment
Also called algorithmic impact assessment
In short
An AI impact assessment is a structured review, done before an AI use starts, of its purpose, the people it affects, the likely risks and the safeguards needed.
- Intermediate
- Process
- Checked
In detail
An AI impact assessment is a planned look at what could happen when an AI system is put to work. Before launch, a team writes down what the system is for, who it affects, what could go wrong, how likely and serious each harm is, and what safeguards reduce it. The result is a record that someone accountable signs off.
Most templates follow similar steps: describe the purpose and the data; identify affected people, including those not in the room; list risks such as unfair outcomes, privacy loss, errors, security weaknesses and effects on jobs; score each by likelihood and impact; set safeguards and tests; decide whether to go ahead; and set a date to review. Where personal data is involved, it often sits alongside or inside a data protection impact assessment.
The assessment matters because problems are cheaper and kinder to fix before people are harmed. It also forces alternatives into view: sometimes a simpler rule, a smaller model or a person does the job better. A good assessment is honest about uncertainty and records what the team did not know.
Assessments are proportionate. Low-risk uses get a short form; systems that shape decisions about people get a full assessment, testing evidence and senior approval, and are reassessed when they change.
Worked example
Lagoon Logistics assesses an AI tool that schedules drivers' shifts, scoring each risk from 1 to 5 for likelihood and impact.
- Unfair allocation of night shifts: 3 x 4 = 12
- Personal data in the scheduling notes: 2 x 3 = 6
- Tool outage on a busy day: 2 x 2 = 4
- After adding a fairness check and a weekly human review, the first risk falls to 2 x 4 = 8
The highest risk is reduced from 12 to 8, the decision to go ahead is recorded, and a review is set for three months.
In practice
Risk, privacy and project teams use a standard template, attach test results, record the decision and the owner, and reopen the assessment when the system, its data or its users change.
Watch out
- An impact assessment is not a one-off form; it needs revisiting when the system changes.
- Scores are a way to rank and discuss risks, not precise measurements.
Learn it properly
Official sources
- AI Risk Management Framework · National Institute of Standards and Technology
- Artificial intelligence · Information Commissioner's Office
- NIST AI RMF Playbook · National Institute of Standards and Technology
Facts last checked on 3 November 2026.
Education, not advice: the Stockertim Dictionary explains what terms mean. It never recommends buying or selling anything.