Score an AI feature’s privacy impact in five minutes
Before you build an AI feature, it helps to know how much privacy risk it carries — and where. This tool asks 20 yes/no/partial questions about the feature and returns an overall privacy impact score from 0 to 100 plus a breakdown across five dimensions: data sensitivity, retention risk, third-party exposure, user control, and automated profiling. Use it to decide whether a full DPIA is needed and which safeguards to add first. It scores everything in your browser; nothing you enter is sent anywhere.
How it works
Each question maps to one of the five dimensions and contributes weighted risk points based on your answer. Questions about special-category data and large-scale profiling carry more weight because regulators treat them as higher risk. The tool sums points per dimension, normalises each to 0–100, and combines them into an overall score. The dimension breakdown tells you not just how risky the feature is, but why — so you can target the right safeguard rather than guessing.
The five dimensions and what drives each
Data sensitivity measures how personal or restricted the data that the AI feature processes is. Ordinary personal data (name, email) scores lower; special-category data under GDPR (health, biometric, political opinion, sexual orientation) scores significantly higher, because regulators impose stricter obligations for processing it.
Retention risk measures how long and how identifiably the data is held. A feature that processes data and discards it immediately scores near zero here; one that creates persistent user profiles or stores interaction logs indefinitely scores high. Retention is often where AI features accumulate risk invisibly — data that was needed for inference is kept much longer than it should be.
Third-party exposure measures how much data is shared with external AI providers, analytics platforms, or sub-processors. A feature built on a third-party foundation model API sends every inference request to that provider. The score here prompts you to confirm that each recipient has an appropriate data processing agreement and, for EU data transfers, an adequate transfer mechanism.
User control measures how much visibility and choice users have over how their data is used by the feature. No consent mechanism, no way to access data, no opt-out, and no deletion path all push this dimension up. High scores here are often the cheapest to fix — adding a settings toggle or a deletion option is technical work but rarely complex.
Automated profiling measures whether the AI feature makes or substantially informs decisions about individuals. A recommendation feature that affects what a user sees scores lower than one that determines loan eligibility, medical triage priority, or hiring decisions. GDPR Article 22 imposes specific obligations on solely automated decisions with significant effects.
When a score triggers a formal DPIA
A full Data Protection Impact Assessment is legally required under GDPR when processing is “likely to result in a high risk” to individuals’ rights. Specific indicators include: processing special-category data at scale, systematic monitoring of publicly accessible areas, and novel use of technologies to profile people for decisions with legal or similarly significant effects. A high score on this tool — particularly in the automated profiling or data sensitivity dimensions — is a reliable indicator that a formal DPIA is warranted and that your DPO should be involved before the feature ships.