The framework
Five principles that keep people in charge.
FRAME rests on five principles. Together they keep human accountability, stakeholder voice, evidence quality, openness, and good judgment about when to use AI at the heart of every evaluation.
01
PRINCIPLE ONE
Accountability Architecture
Establish clear lines of human accountability before deploying any AI system. Document who is responsible for which decisions.
In practice
- Decision authority: who can authorise or discontinue AI use.
- Validation responsibilities: who reviews outputs before they inform findings.
- Escalation procedures: what happens when problems emerge.
- Documentation standards: how decisions and overrides are recorded.
Clear role differentiation is what makes effective human–AI collaboration possible.
02
PRINCIPLE two
Stakeholder Engagement
Ensure the meaningful participation of affected communities in decisions about AI use. This goes beyond consultation to genuine power-sharing.
In practice
- Involve stakeholders in deciding when and how AI is appropriate, not just informing them after the fact.
- Address the “paradox of power”: those with the power to address injustice are often least able to perceive it.
- Design AI use so it elevates local voices rather than overriding them.
Stakeholder voice belongs in evaluation design, not just at the consultation stage.
03
PRINCIPLE THREE
Epistemic Integrity
Maintain rigorous standards for evidence quality. AI can generate plausible but inaccurate content, so conclusions must rest on verified evidence, not AI artefacts.
In practice
- Treat every AI-generated claim as unverified until traced to source data.
- Watch for hallucinations: errors that are internally coherent and linguistically polished are the hardest to catch.
- Remember that early language models have invented citations that looked academically credible.
Every AI-generated claim must be traceable to source data.
04
PRINCIPLE four
Transparency
Document and disclose AI use at all stages so stakeholders can understand how AI has influenced the process and the findings.
In practice
- Disclose which tasks AI performed and how its outputs were validated.
- Avoid “shadow AI”: undeclared AI use erodes the trust that evaluation legitimacy depends on.
- Make disclosure a standard part of reporting, not an afterthought.
Undeclared AI use undermines the trust relationships evaluation depends on.
05
PRINCIPLE FIVE
Proportionality
Match AI deployment to genuine need. The complexity and risk of AI governance should be proportionate to the benefits sought.
In practice
- Avoid using AI for its own sake.
- Account for hidden costs: time saved on a task may be consumed by the verification it requires.
- Use AI to enhance output without displacing the resources that quality depends on.
Match the effort of governance to the real benefit AI delivers.
Principles in motion
Five principles, applied across seven phases
The principles are not abstract ideals — FRAME maps them onto every phase of a real evaluation, from design through to how findings are used.
Ready to apply FRAME to your next evaluation?
Download the FRAME checklist, or enrol in the AI for MEL course to put the framework into practice.