Certified AI Business Analysis Professional

What you will be able to do

• Decide whether a business need is a good candidate for AI, conventional automation or a process change
• Write requirements for probabilistic behavior: accuracy thresholds, confidence handling and fallbacks
• Assess the data a proposed AI solution depends on before committing to it
• Model human-in-the-loop work and exception paths
• Define acceptance criteria and evaluation sets that a model must pass
• Identify bias, privacy and explainability requirements early

Modules

1. 1The analyst's role on AI initiatives

   Where business analysis changes when part of the solution learns from data, and where the classic techniques still apply unchanged.

   • Rule-based, statistical and generative systems compared
   • Framing problems as predictions, classifications or generations
   • The AI suitability test: volume, variability, data and tolerance for error
   • Stakeholders unique to AI work: data owners, model risk, ethics review

   Exercise. Take three requests from your own backlog and classify each as AI, conventional automation or process change, giving the reason in one sentence.

2. 2Eliciting requirements for probabilistic systems

   Users are used to software that is right or wrong. AI is right most of the time. Requirements have to say how often, for whom and what happens otherwise.

   • Accuracy, precision and recall in business language
   • Confidence thresholds and routing to people
   • Cost of a false positive against a false negative
   • Non-functional requirements: latency, drift, retraining frequency

   Exercise. For a document classification scenario, write the requirement that states the acceptable error rate and what happens to low-confidence items.

3. 3Data requirements and readiness

   Most AI projects that fail do so on data. The analyst documents what data exists, its lineage, its quality and whether it may legally be used.

   • Data inventories and source-to-feature mapping
   • Quality dimensions: completeness, accuracy, timeliness, consistency
   • Labeling needs and who will do the labeling
   • Consent, retention and permitted use

   Exercise. Build a one-page data readiness assessment for a proposed model, rating each source on the four quality dimensions.

4. 4Modeling human and machine work together

   Process models that show where the model decides, where a person reviews and how exceptions flow back into training.

   • Human-in-the-loop and human-on-the-loop patterns
   • Swimlanes for model, reviewer and system
   • Feedback loops and correction capture
   • Decision tables alongside model outputs

   Exercise. Draw the future-state process for invoice handling where extraction is automated and a clerk reviews items below 90% confidence.

5. 5Acceptance, evaluation and validation

   Acceptance criteria for AI are statistical. The analyst defines the evaluation set, the measures and the pass mark before the model is built.

   • Building a representative evaluation set
   • Baseline comparison: today's manual error rate
   • Pilot design and shadow running
   • Sign-off that business owners can understand

   Exercise. Write acceptance criteria for a customer-email triage model, including the size and composition of the test set.

6. 6Responsible AI requirements

   Fairness, transparency, privacy and accountability expressed as requirements that can be traced and tested, not as statements of intent.

   • Identifying affected groups and fairness measures
   • Explainability needs by audience
   • Privacy impact and data minimization
   • Audit trail and decision records

   Exercise. Add three testable responsible-AI requirements to an existing requirements document and trace each to a test.