An impact assessment is a structured check of who may be affected by a proposed ML system, what benefits and costs they may experience, and what ethical issues deployment may create. Begin once the project hypothesis is clear: document corporate-responsibility and ethical requirements, challenge the project assumptions, and create privacy and data impact assessment (PDIA) and algorithmic impact assessment (AIA) documents. section 2.6 section 6.2
Make the assessment a project checkpoint, not a one-time form:
- Involve stakeholders, including users and people affected by the system; examine costs alongside benefits for those affected. section 6.2
- Consider the full project and system lifecycles. Decide how impacts will be measured as the system evolves and whether its governance mechanisms are adequate. section 6.2
- Check applicable data and domain-specific law, while treating legal compliance as a floor rather than a complete ethical assessment. section 2.6
Revisit the assessment after exploratory data analysis and further work on business requirements. The stronger understanding gained at that point supports a better review of ethical, privacy, and security issues. Available evaluation tools vary in project-stage fit and ethical scope, so choose deliberately rather than assuming any one tool is complete. section 6.2
