Data Analytics, Technology, and Decision Support
What the exam tests
This domain tests analytics types, data quality, technology-enabled decision support, and the governance of models and data. Expect descriptive, diagnostic, predictive and prescriptive analytics, BI, RPA, AI, cloud, cybersecurity, privacy, and model risk.
Core rules and concepts
- Analytics progression: descriptive (what happened), diagnostic (why), predictive (what is likely), prescriptive (what to do). Each answers a different decision question.
- Data quality dimensions: accuracy, completeness, timeliness, consistency, validity, uniqueness. Poor input quality undermines every model. A data dictionary and lineage help trace definitions and transformations.
- Predictive models need a target, features, training and test data, and performance metrics. Classification uses a confusion matrix: true and false positives and negatives. Choose thresholds based on the cost of errors, not accuracy alone.
- Technology: BI dashboards support monitoring; RPA automates repetitive rules-based tasks; machine learning finds patterns; cloud changes control and vendor-risk profiles. Controls still include access, change management and data privacy.
- Model risk: document assumptions, validate, monitor drift, and keep human oversight for consequential decisions.
Worked example
A subscription business models churn. Model A identifies 100 true churners and 50 false positives, but misses 40 churners (false negatives). Model B identifies 80 true churners and 10 false positives, missing 60. A retention offer costs $20; a missed churner costs $500 in lost contribution. Model A cost = 50 x $20 + 40 x $500 = $1,000 + $20,000 = $21,000. Model B cost = 10 x $20 + 60 x $500 = $200 + $30,000 = $30,200. Model A is better on business cost even though Model B has fewer false alarms. Accuracy alone would not show this; the decision threshold and error costs drive the choice.
Traps that cost marks
- Treating correlation as causation.
- Using training accuracy as proof of future performance; overfitting is likely.
- Ignoring class imbalance in churn or fraud models.
- Automating a bad process with RPA.
- Forgetting privacy and access controls when moving data to the cloud.