IIBA-CBDA Sample Questions & Answers

This recommendation is effective because it perfectly translates a statistical finding into a concrete business strategy. It follows the SMART criteria: Specific (increase bundled policies), Measurable (15%), Actionable (offer a discount), Relevant (to the insight about churn), and Time-bound (next fiscal year). This creates a clear path from data insight to business action and measurable outcome.

Before creating a strategy or selecting tools, it is crucial to understand the current state. An analytics maturity assessment evaluates the organization's existing people, processes, and technology. This provides a clear baseline, identifies inconsistencies and gaps (like data silos), and informs the development of a realistic and targeted strategic roadmap. Jumping to a tool selection or governance model without this understanding often leads to failure.

With 20% of values missing, deleting the rows (listwise deletion) would discard a significant portion of the dataset and potentially introduce bias. Replacing missing values with a measure of central tendency like the mean (for symmetric distributions) or median (for skewed distributions) is a standard and reasonable first approach. This technique, called imputation, preserves the sample size without drastically distorting the feature's overall distribution. More advanced techniques exist, but mean/median imputation is a common and appropriate starting point.

This statement is false. A high R-squared value simply means the model explains a large proportion of the variance in the training data. However, it can be misleading. A model can be overfit, meaning it has learned the noise in the training data too well. Such a model will have a high R-squared on the data it was trained on but will fail to generalize and perform poorly on new, unseen data. Other diagnostics, such as checking residual plots and using an adjusted R-squared, are necessary to assess model reliability.

CATWOE is specifically designed to analyze a problem situation from multiple stakeholder perspectives. It would compel the analyst to identify the Customers (shoppers), Actors (inventory managers, buyers), the Transformation process (raw materials to sold goods), the Weltanschauung (the worldview that inventory optimization is critical), the Owners (company executives), and Environmental constraints (fast-fashion trends, supply chain issues). This comprehensive view is ideal for framing a complex business problem like inventory optimization.

This is a classic example of the principle 'correlation does not imply causation'. It is highly likely that a third, confounding variable—such as sunny, hot weather—is causing an increase in both ice cream sales and sunglasses sales independently. Acting on the assumption that one causes the other (causation) could lead to an ineffective marketing strategy. The correct interpretation is that the two are related, but one does not necessarily cause the other.

This scenario is a critical lesson in business data analytics: quantitative data (the 'what', 0.5% usage) is incomplete without qualitative context (the 'who' and 'why', valuable enterprise clients). A purely data-driven decision based on the usage metric would have been disastrous for the business. The product manager's domain knowledge provided the essential context that the raw numbers lacked. Effective decision-making requires synthesizing both types of information.

While quality, format, and privacy are important, the most fundamental problem is that the data is not available for analysis in a centralized or persistent way. The data exists but is ephemeral and siloed. The team's first major task is to establish a data acquisition pipeline (e.g., using IoT gateways) to stream this sensor data to a central repository (like a data lake or time-series database) where it can be stored, processed, and used for modeling.

The large gap between training and validation accuracy is a classic sign of overfitting, where the model has learned the training data's noise rather than its underlying patterns. For decision trees, overfitting manifests as an overly complex tree. Pruning is the standard technique to combat this. By limiting the tree's depth or requiring a minimum number of samples to create a new split (leaf), the analyst simplifies the model, forcing it to learn more general patterns and improving its performance on unseen data.

Understanding stakeholder expectations around time and value is paramount. If the stakeholder needs a quick answer to make an immediate decision, the three-month project (B) is unsuitable, regardless of its potential. Conversely, if they are planning long-term strategy and can wait for a more robust answer, the quick analysis (A) might be insufficient. Clarifying the acceptable time-to-value and their comfort with the differing levels of insight from each approach is the critical first step in selecting the right path.