Analyze Data
Analysis that supports a claim
This domain covers exploratory analysis and the inferential and predictive techniques a business analyst must be able to interpret, even when a data scientist runs them. Know descriptive statistics (central tendency, spread, distribution shape), segmentation, correlation, hypothesis testing (null and alternative, p-value, confidence interval, Type I/II error), effect size, and the predictive family - regression, classification, clustering - plus validation through train/test splits and cross-validation.
Traps. Reading a p-value as "the probability the result is true" (it is the probability of data this extreme if the null holds). Treating statistical significance as business importance - with large samples, trivial effects turn significant; always report effect size. Confusing correlation with causation, or ignoring a confounder such as seasonality. Multiple comparisons inflating false positives. Data leakage, where information unavailable at prediction time sneaks into the model, and overfitting, where training accuracy is high and holdout accuracy collapses.
Rule of thumb - always beat a baseline. Any model or finding must be compared with the naive alternative: the current process, last period's value, or the majority-class rate.
Worked example. A churn model reports 92% accuracy. The base churn rate is 8%, so predicting "nobody churns" also scores 92%. You check precision and recall on the churn class and lift over baseline; only then does the model demonstrate value. Visualize distributions and segment results before trusting a summary statistic - averages hide the subgroups where the decision actually happens.