AAIA study plan: pass in 6 weeks

Six weeks at 6–8 hours a week, assuming you hold CISA or a qualifying designation — you must, to certify.

The plan deliberately inverts most auditors’ instincts. AI Auditing Tools and Techniques is only 21% of the exam and you already have that discipline. AI Operations is 46% and is mostly new. Study time follows the weighting, not the comfort.

Week 1: AI governance and risk (33%), part one

• AI models, considerations and requirements. What a model is, the difference between training and inference, why behaviour comes from data. You cannot audit what you cannot describe.
• AI governance and programme management. Inventories, ownership, approval paths, oversight structures.

Week 2: governance and risk, part two

• AI risk management. Risk sources specific to AI, thresholds expressed as measurements rather than pass/fail, treatment decisions, and why reassessment is mandatory when drift exists.
• Privacy and data governance programmes. Training data as personal data, minimisation, retention, and the awkward fact that deleting a record does not remove its influence from a trained model.
• Leading practices, ethics, regulations and standards. Not clause recitation — the consistent shape: inventory, impact assessment, human oversight, documentation, transparency, auditability.

Weeks 3, 4 and 5: AI operations (46%)

Three weeks for nearly half the exam.

Week 3 — data and lifecycle:

• Data management specific to AI: provenance, integrity, quality, labelling, segregation, retention.
• AI solution development methodologies and lifecycle: how these systems are actually built, from problem framing through data preparation, training, validation, deployment and monitoring.

Week 4 — change and supervision:

• Change management specific to AI. The three ways behaviour changes: retraining, a provider updating a hosted model, and drift. None appears in a traditional change record.
• Supervision of AI solutions — outputs, impacts and decisions. Who reviews what, how often, and with what authority to intervene.

Week 5 — testing, threats, incidents:

• Testing techniques for AI solutions. Validation, evaluation metrics, benchmark sets, adversarial testing, fairness testing. Understand what each demonstrates and what it does not.
• Threats and vulnerabilities specific to AI: poisoning, extraction, inversion, evasion, prompt injection — and which lifecycle stage each targets.
• Incident response specific to AI: what an AI incident is, how containment works when behaviour comes from weights, and what rollback means.

Week 6: audit techniques (21%) and revision

The smallest domain, and largely a mapping exercise onto method you have.

• Audit planning and design for an AI engagement: scoping a system whose boundaries include its data and its provider.
• Testing and sampling methodologies applied to AI — sampling outputs rather than transactions.
• Evidence collection techniques. The central question of the credential: what constitutes sufficient, appropriate evidence for a model? Validation results, evaluation metrics, provenance records, monitoring history, approval records.
• Audit data quality and data analytics.
• AI audit outputs and reports — writing findings about a probabilistic system without overstating certainty.

Then a full practice exam, mistakes sorted by domain, weakest area rebuilt, second practice exam.

Where the hours go

The habit that earns marks

For every operations topic you study, finish by answering one question: what evidence would I request to verify this?

Studying data provenance? The evidence is a documented data lineage and source agreements. Studying validation? Evaluation results against defined thresholds, dated and approved. Studying change management? A record of model versions with revalidation evidence for each.

That single habit converts operations knowledge into audit answers, and it is precisely what the exam is testing.

If time runs short, cut week 6, not week 4. You can improvise audit method under pressure; you cannot improvise an understanding of how model behaviour changes.