Meridian's finance team already runs a standard monthly data-cleaning process for its budget reports — removing duplicates and fixing obvious typos. The AI team is told to reuse that same process unmodified for preparing labeled training data for the fraud-detection model. What AI-specific data preparation need does this reuse overlook?

AI-specific data preparation needs go beyond generic cleaning: they include producing accurately labeled ground truth, such as confirmed fraud versus legitimate transactions, and transforming data into model-ready features, neither of which a standard BI-report cleaning process, such as deduplication and typo correction, is designed to do. Reusing the finance team’s process unmodified would leave the fraud-detection model without the labels and feature structure it needs to learn. The claim that nothing is overlooked incorrectly assumes generic BI cleaning is sufficient, missing the labeling and feature-engineering gap that is specifically called out as an AI-specific need. The need to remove all numeric data is a nonsensical overcorrection, because numeric data such as transaction amounts and timestamps is often essential model input, not something to remove. The need to convert the data into a printed report format confuses data preparation with reporting output format; a printed report is irrelevant to what a model consumes for training. For the exam, remember AI data preparation is a superset of generic data cleaning — it must also address labeling requirements and feature readiness, which is what distinguishes it from ordinary BI hygiene work.