Algorithmic Impact Assessment
When an algorithm decides, accountability does not disappear.
Canada AIA
ELI Model Rules
It shifts to the organisation that built or deployed the system. We assess automated decision systems for fairness, transparency, and accountability, using a dual-framework approach grounded in Canada’s Algorithmic Impact Assessment and the ELI Model Rules.
Algorithmic Impact Assessment
Assessment of fairness, transparency, accountability, and decision-making risks.
Canada AIA
The Challenge
Automated decisions carry consequences that outlast the algorithm.
Organisations increasingly rely on automated systems to make or inform decisions that affect individuals: credit assessments, hiring screens, benefits eligibility, risk scoring. When those systems produce unfair, opaque, or unaccountable outcomes, the organisation is exposed legally, reputationally, and in the eyes of regulators.
An Algorithmic Impact Assessment is the structured process for understanding those risks before they become incidents. We evaluate your automated decision systems using a dual-framework approach Canada’s AIA and the ELI Model Rules and a proprietary assessment instrument that is questionnaire-led and evidence-driven.
The output is a structured, defensible set of findings with the accountability chain, mitigation controls, and transparency documentation to go with them.
“When an algorithm decides, accountability does not disappear.”
The principle that makes algorithmic impact assessment necessary not optional.
Framework 1
Canada’s Algorithmic Impact Assessment
Developed by Canada’s Treasury Board Secretariat for automated decision systems in public services. Widely used internationally as a structured reference for assessing the impact, fairness, and accountability of algorithmic systems.
Framework 2
ELI Model Rules on Algorithmic Decision-Making
The European Law Institute’s formal European model rules for automated decision-making systems. Provides the European legal reference layer that complements the Canadian operational framework.
What We Assess
Six dimensions of algorithmic accountability.
Our assessment covers the dimensions that matter when an algorithm makes or influences decisions affecting individuals or organisations from decision impact to mitigation controls.
Decision Impact & Reversibility
What decisions does this system make or inform? Who is affected and how significantly? Can decisions be challenged, reviewed, or reversed? We assess the scope and consequence of algorithmic decisions — and whether adequate correction mechanisms exist.
Affected Stakeholders & Rights
Who are the individuals and groups affected by the system’s decisions? What rights do they have in relation to those decisions to be informed, to contest, to seek human review? We map stakeholder exposure and evaluate whether rights protections are operationally real.
Fairness & Non-Discrimination
Does the system produce outcomes that are fair across protected characteristics age, gender, race, ethnicity, disability? We assess for proxy discrimination, disparate impact, and the training data patterns that may encode unfairness into decision logic.
Transparency & Explainability
Can the system’s decision logic be explained in terms that affected individuals and oversight bodies can meaningfully understand? We assess whether the level of explainability is adequate relative to the impact of the decisions being made and whether disclosure obligations are met.
Human Oversight & Redress
Is there meaningful human oversight of automated decision-making not just a nominal review step? We assess whether human oversight is structurally effective, whether redress mechanisms are accessible, and whether they actually function when challenged.
Mitigation Controls & Accountability Chain
What controls are in place to mitigate identified risks and who is accountable for them? We assess the completeness and clarity of the accountability chain from the system’s developers to its operators to the organisation ultimately responsible for its use.
Dual-Framework Approach
Why two frameworks?
Canada’s AIA provides structured operational guidance for assessing automated decision systems developed through practical government application and widely adopted as an international reference instrument.
The ELI Model Rules provide the European legal reference layer addressing the rights and obligations that apply when algorithmic systems make or influence decisions affecting individuals in a European legal context.
Together, they produce an assessment that is both operationally rigorous and legally grounded relevant across jurisdictions.
Evidence-driven – findings grounded in documentation, system analysis, and stakeholder interviews.
Defensible – assessment output is structured for regulatory scrutiny, not internal comfort.
Actionable – findings come with prioritised mitigation controls and an accountability chain, not just a risk catalogue.
What You Gain
Accountability, before the incident.
An Algorithmic Impact Assessment is not crisis management. It is the structured work that prevents automated decision systems from becoming a crisis in the first place.
Reliability
Confidence that the data powering your AI and decisions is accurate when it matters most.
Traceability
Clear data lineage and audit trails that support regulatory disclosure and model validation.
Control
Governance structures and access controls that prevent integrity failures before they propagate.
Defensibility
Findings anchored in SOC 2 externally credible and verifiable beyond internal self-assessment.
AI Foundation
A data control environment that supports trustworthy AI models that perform because the data they depend on is sound.
Complete the Picture
Explore our other governance services.
AI Act Readiness is the strategic foundation. Data integrity, algorithmic accountability, and production-grade AI agents complete the governance lifecycle.
Ready to Begin
Let's assess your algorithmic decision systems.
An Algorithmic Impact Assessment is most valuable before an issue surfaces publicly. The earlier it is done, the more options the organisation has to act on what it finds.