USE-CASE REVIEW · Checklist
AI Use-Case Risk Triage
Decide which AI use case needs a person to look at it first. Eight questions give a review priority, the reasons behind it and the open questions you still have to answer.
- Runs in your browser
- No sign-in
- Result before any contact step
- Priority
- High · review before launch · 17 of 26 points
- Floor
- consequential purpose with an AI that decides → at least High
- Reviewers
- business owner · legal · privacy · HR · risk
A hiring screener that decides: 17 of 26 points, and a floor holds it at High whatever the average says.
Most organizations have more AI ideas than reviewers. This tool helps you order them. You describe one use case in eight multiple-choice questions: what the AI is for, who it touches, what data it reads, what it does, what a mistake costs, whether a mistake can be undone, who checks the output and whether it acts outside your business.
The result is a review priority with the rule behind every part of it. Where you pick Not sure, the tool assumes the worst and lists the question, so a gap in your knowledge never makes a use case look calmer than it is.
Your inputs · worked out in your browser
0 of 9 answered · empty
Waiting for inputs
Review priority
Answer all 8 questions and the restricted-practice check to see the review priority, the answers behind it and what to do next. 9 left.
What you get
- A review priority and the first action it points to
- An illustrative review score out of 26, with the answers that earned each point
- Floors that a good average cannot lower, flags with a next step and open questions
- Suggested reviewers, an action plan and an evidence-request list you can export
If you accept analytics cookies, we record only which tool was opened, that a result was shown and which export or service link was used. Inputs, answers and results are never part of that record.
Method version 2026-10-14. Review aid from self-reported answers. It is not legal advice, a regulatory classification, a certification or an audit, and it cannot see or test any system.
How it works
Each answer carries points. A riskier answer carries more. The eight questions add up to a top score of 26. The score is a way to order use cases against each other, so it is shown as an illustrative review score and never as a chance that harm will happen.
The score maps to a starting priority: 0 to 6 points is Low, 7 to 13 is Medium and 14 to 26 is High. Stop is never reached through points. Only a floor can set it.
An average can hide a serious fact. A use case that looks gentle on six questions and severe on two can land in the middle. Floors prevent that. A floor is a minimum priority that applies whatever the other answers are. The final priority is the higher of the score band and the highest floor that fired.
Not sure is never treated as a pass. It scores the top points of its question, it counts toward every floor the answer could trigger, and it appears in the list of open questions. The result tells you how many answers it had to assume. Find out, change the answer and the priority moves.
Each flag names the answer behind it and one next step. The suggested reviewers follow printed rules too: personal data points to privacy, an AI that acts outside the business points to security, and a purpose in a closely watched area points to legal.
The restricted-practice boxes list four kinds of practice that some laws limit or ban. They are plain-language prompts, not legal definitions. This tool is not a legal classification and not an EU AI Act determination. Only a lawyer can say which rules apply to a real system.
Formula and rules
Illustrative review score = the sum of the eight answer points (top score 26) Review priority = the higher of the priority from the score and the highest floor that applies Purpose: internal work 0, marketing 1, customer service 1, other 2, hiring and work 4, credit and insurance 4, education 4, health 4, housing and essential services 4, legal and public services 4, safety 4 Who is affected: no individuals 0, staff 1, customers or the public 2, children or vulnerable groups 3 Data used: public 0, internal business 1, personal 2, sensitive personal 3 Role of the AI: drafts 0, recommends 1, decides 3, acts alone 4 Impact if wrong: minor 0, moderate 2, significant 4 Reversibility: easy 0, with effort 1, hard 3 Human oversight: named reviewer 0, spot checks 1, none 3 External actions: no 0, yes 2 Not sure scores the top points of its question and becomes an open question 0 to 6 points: Low: standard checks 7 to 13 points: Medium: review soon 14 to 26 points: High: review before launch Stop: legal review first. Points never set it; only the first floor below does Floors (a minimum priority that points cannot lower): A restricted practice is ticked: the priority is Stop. Restricted practices not checked: the priority is at least High. Consequential purpose with an AI that decides or acts: the priority is at least High. Significant impact that is hard to reverse: the priority is at least High. Vulnerable groups with sensitive personal data: the priority is at least High. Reviewers: the business owner always; legal, privacy, security, HR and risk are added by the answers
Assumptions
- The points, bands and floors are rules chosen for this tool to put use cases in order. They are illustrative and do not come from a law or a standard.
- Your answers describe the use as you plan to run it, not as you hope it will run.
- The seven consequential purposes (hiring, credit, education, health, housing, legal and public services, safety) are a plain list chosen for this tool. A law may name other areas or word them differently.
- A restricted-practice box you leave clear means you checked and none applies. If you have not checked, tick the box that says so and the tool treats it as an open question.
- Not sure always means the worst case for that question.
Worked example
Illustrative worked example · example valuesSample
Example 1, a drafting assistant for staff: it writes first drafts of internal notes from internal documents, and a named colleague reads every draft. The answers score internal work 0, staff 1, internal business data 1, drafts 0, minor impact 0, easy to undo 0, named reviewer 0 and no external actions 0. That is 2 of 26 points and no floor applies, so the priority is Low: standard checks, with no flags.
Example 2, a hiring screener that decides: it ranks applicants and rejects the lowest without a person reading each file, and a recruiter spot-checks later. The answers score hiring and work 4, customers or the public 2, personal data 2, decides 3, significant impact 4, undoable with effort 1, spot checks 1 and no external actions 0. That is 17 of 26 points, which is High on points alone, and the floor for a consequential purpose with an AI that decides also applies. The priority is High: review before launch. Suggested reviewers are the business owner, legal, privacy, HR and risk.
Go back to example 1 and tick emotion recognition at work or school. The score stays at 2 of 26, yet the priority becomes Stop: legal review first. Good answers elsewhere cannot average a floor away.
Evidence to gather
Records that make the result easier to check and act on:
- An intended-use statement: what the AI is for, what it must not be used for and who owns it.
- A data inventory: what the AI reads and produces, where it comes from, who can see it and how long it is kept.
- A human review procedure: who checks outputs, how often, what they may override and where overrides are recorded.
- A rollback and correction plan: how a wrong result is found and undone, and who is responsible.
- Test results on realistic cases, including cases where the AI gets it wrong.
- A written view from a lawyer where a restricted practice, a closely watched purpose or vulnerable groups are involved.
- A dated copy of the triage, so a later review can see what was known at the time.
What it does not do
- The tool orders your answers. It does not look at any real system, test it or read any document.
- The points, bands and floors are rules chosen for this tool. They are not taken from a law and they are not a probability of harm.
- A restricted practice counts only when you tick it. A clear box means you say none applies, not that the tool has confirmed it.
- A purpose area may fall into categories some regulators treat as high-risk. Only a lawyer can say whether your use does.
Questions
Does this tell me whether my AI use is allowed?
No. It orders use cases for review using only your answers. Whether a law applies to a real system depends on facts and rules that only a lawyer can check. The result is a triage, not a legal classification.
Is the priority an EU AI Act risk level?
No. That law has its own legal tests and this tool does not apply them. A purpose area here may fall into categories some regulators treat as high-risk, but it is not an EU AI Act determination, and only a lawyer can say whether your use falls under it.
Is the score a probability that something goes wrong?
No. It is a count of points from your answers, used to put use cases in order. A higher number means look sooner. It does not estimate how likely harm is, and it is not calibrated to any data.
Why does Not sure raise the priority?
A gap in what you know is a gap in the review. If the tool treated it as a good answer, you could reach a calm result by not looking. So it assumes the worst, counts the answer against every floor it could trigger and lists the question for you to settle.
Why have floors when there are points?
Points average out. One serious fact, such as a ticked restricted practice or an AI that decides on hiring, can disappear in an average. A floor sets a minimum priority that no other answer can lower.
Are my answers stored or sent anywhere?
No. Everything is worked out in your browser and nothing is saved when you leave the page. If you accept analytics, the site records only that the tool was opened and that a result was shown, never the answers or the result.