These 12 CBDA analyze data questions cover Domain 3 of the Guide to Business Data Analytics: preparing data, exploring it, checking models and reading technical visuals. Each original scenario asks what the analyst should do first, best or next, and the answer key cites the Guide section behind every answer.
How this CBDA analyze data questions set is built
The set targets Domain 3, Analyze Data, which carries 16% of the exam. It draws on the tasks in Guide section 2.3 and the techniques listed for this domain:
- 2.3.1 Develop Data Analysis Plan: techniques, candidate models, sources and joins, and cleaning rules. The analyst supplies business context.
- 2.3.2 Prepare Data: linking, normalizing, scaling, cleansing and weighting data. The analyst flags statistically sound choices that clash with business rules.
- 2.3.3 Explore Data: checks for integrity, validity, reliability and bias, plus gaps and outliers.
- 2.3.4 and 2.3.5: turn the business question into a mathematical one and iterate.
- Techniques: 3.9 Descriptive and Inferential Statistics, 3.11 Exploratory Data Analysis, 3.14 Optimization, 3.1 Business Simulation and 3.18 Technical Visualizations.
Questions follow the exam's scenario style, sit slightly above typical bank difficulty, and are original, not exam dumps. Three use small numbers. For the upstream steps, read our guide on sourcing and validating data.
| Sub-topic | Questions |
|---|---|
| Preparation and missing data (2.3.2) | 1, 4 |
| Exploration and outliers (2.3.3) | 2 |
| Statistics and expected value (3.9) | 3, 4, 6 |
| Technical visualizations (3.18) | 5, 7, 8, 9 |
| Assessing the approach (2.3.5) | 10 |
| Optimization and simulation (3.14, 3.1) | 11, 12 |
CBDA exam at a glance
| Fact | Detail |
|---|---|
| Format | 75 multiple-choice, scenario-based questions |
| Time | 2 hours |
| Analyze Data weight | 16% (Domain 3) |
| Exam fee | US$405, or US$250 for IIBA members |
| Passing score | Not published by IIBA |
Last verified: October 2026
Practice Questions
Question 1
Harbourline Insurance predicts claim settlement time. During preparation, the data scientist finds that 3% of claims lack a region code, all from one branch that moved to a new intake system last quarter. The data scientist proposes dropping them as statistically insignificant.
What should the analyst do FIRST?
- A. Agree to drop them, since 3% will not change model accuracy
- B. Investigate the branch's recording process with stakeholders
- C. Fill the gaps with the most frequent region in the data
- D. Ask for a different model that tolerates missing values
Question 2
Corvana Hotels forecasts room demand from booking lead time. Exploration shows a tight cluster of bookings at exactly 999 days, far above all other values. The data scientist wants to remove them as outliers.
What is the BEST action for the analyst?
- A. Keep them, because removing data always biases a forecast
- B. Replace them with the median lead time
- C. Exclude them at once, since extreme values are noise
- D. Check the booking system's rules for what 999 means
Question 3
Streamly can afford a fixed number of retention calls. The retention lead plans to call Basic customers first because they churn most. The analyst prepares this table:
| Segment | Churn probability | Annual revenue per customer |
|---|---|---|
| Basic | 30% | $120 |
| Pro | 12% | $900 |
| Team | 5% | $2,400 |
Which segment should each call target first?
- A. Team, with the highest expected revenue loss per customer
- B. Basic, with the highest probability of cancelling
- C. Pro, as a balance of probability and revenue
- D. Any segment, since expected loss is equal across them
Question 4
Mirelle Home, an online furniture retailer, has seven sample basket values: $42, $38, $45, $40, $44, $41 and $950. The $950 basket is a valid bulk order from a hotel. The data scientist plans to fill missing basket values with the sample mean.
What should the analyst recommend?
- A. The mean of about $171, as it reflects every order
- B. The mode, as the most frequent basket value
- C. The median of $42, as the bulk order distorts the mean
- D. Deleting the bulk order, then using the mean
Question 5
Brightpath Motors predicts used-car prices from over 200 variables. As variables were added, predictive power dropped, so the data scientist reduced them to principal components and plotted the first two. Sales leaders ask what the model is based on.
How should the analyst respond?
- A. Explain that components cannot be described in business terms
- B. Describe each component as a group of car traits
- C. Restore the full variable list so leaders see every input
- D. Replace the plot with a pie chart of the variables
Question 6
Tallis Grocers models weekly store sales. The two strongest predictors, advertising spend and promotional leaflets distributed, have a correlation of 0.93. The operations director wants both kept.
What should the analyst do NEXT?
- A. Keep both, since high correlation proves each drives sales
- B. Drop advertising spend, as leaflets are easier to count
- C. Add predictors so the pair carries less weight
- D. Review a correlation map to avoid double counting
Want more than 12? Create a free account to unlock more CBDA practice questions, including the full 75-question CBDA exam simulator with exam and practice modes. certificate.tips is our product.
Question 7
Ferro Logistics fitted a linear regression of fuel cost on route distance. The residuals are negative at short and long distances and positive in the middle, forming an inverted U.
What does this MOST likely indicate?
- A. The relationship is not linear
- B. The model fits well, as errors average near zero
- C. The data contains duplicate records
- D. Training and test sets were mixed up
Question 8
Quillon Telecom grouped subscribers into nine usage clusters. On the silhouette plot, four clusters are wide and long; five are thin, and two start with negative scores.
What should the analyst suggest?
- A. Add clusters so each group is smaller
- B. Keep nine, since marketing asked for nine
- C. Reduce the clusters and review them with marketing
- D. Replace clustering with linear regression
Question 9
Pemberton Bank tests three card-fraud classifiers. Fraud is 0.9% of transactions. Results on the same test data:
| Model | Accuracy | Area under ROC curve |
|---|---|---|
| X | 99.1% | 0.71 |
| Y | 98.7% | 0.88 |
| Z | 99.0% | 0.80 |
Which model should go forward for fine-tuning?
- A. X, with the highest accuracy
- B. Y, with the largest area under the curve
- C. Z, as a balance of both metrics
- D. None, as accuracy is nearly equal
Question 10
Lumen Learning predicts which leads and past customers will buy a discounted course bundle. A logistic regression scores 91% on training data but 58% on test data. The data scientist suspects customers and leads differ and that missing variables, such as income, drive purchases.
What should the analyst do NEXT?
- A. Help stakeholders weigh options on cost, feasibility and impact
- B. Launch to top-scoring prospects and measure results
- C. Tune the model until test matches training
- D. Drop the test split and retrain on all records
Question 11
Arden Dairy's linear programming model plans daily yogurt and cheese output for maximum profit, using all milk and labour hours. Supervisors say the plan cannot run, because cold storage holds only two days of yogurt.
What should the analyst do FIRST?
- A. Ask for a neural network instead
- B. Present the plan, since it is optimal
- C. Lower the profit target until it fits
- D. Elicit missing constraints and update the model
Question 12
Novaro Foods plans a new meal-kit tier with no sales history. Leadership asks how adoption changes if the price rises 10% or a competitor launches. Live tests are not possible.
Which approach BEST fits?
- A. Regression on existing tiers' sales
- B. A survey on preferred prices
- C. A business simulation of what-if scenarios
- D. A dashboard of current volumes
Practise under exam conditions. Our CBDA exam simulator has a full 75-question set with an exam mode and a practice mode that explains each answer. Create a free account to try more questions across all six domains.
Answer Key and Explanations
1. Correct answer: B. The Guide notes that records skipped as statistically insignificant may still need business investigation. The analyst checks the recording process with stakeholders. Gaps from one branch and one new system suggest a process issue. Source: Guide to Business Data Analytics, 2.3.2 Prepare Data Why the others are wrong: A ignores the clustered cause; C imputes before the cause is known; D treats a process issue as a modeling one.
2. Correct answer: D. In the Guide's housing example, extreme prices were an artificial value used when prices could not be disclosed. The analyst found this by reviewing the collection method for missed business rules, which justified exclusion. Source: Guide to Business Data Analytics, 2.3.3 Explore Data Why the others are wrong: A keeps possible codes; B imputes before the meaning is known; C skips the business check.
3. Correct answer: A. Expected loss per customer is probability times value: Basic $36, Pro $108, Team $120. The Guide warns that attrition work can optimize the wrong measure, the probability of leaving, instead of the expected value of lost business. Source: Guide to Business Data Analytics, 3.9 Descriptive and Inferential Statistics Why the others are wrong: B uses probability alone; C is not the highest; D is false, the values differ.
4. Correct answer: C. The values sum to $1,200, so the mean is about $171 and the median is $42. The Guide notes that the mean is sensitive to outliers and the median is often used to fill missing data. The valid bulk order stays. Source: Guide to Business Data Analytics, 3.9 Descriptive and Inferential Statistics Why the others are wrong: A is not a typical basket; B fails because no value repeats; D deletes a valid record.
5. Correct answer: B. The Guide explains that principal components group the variables that matter most when many dimensions erode predictive power. Its car example reads one component as large-engine, high-power cars. Analysts translate components into business terms, such as segments. Source: Guide to Business Data Analytics, 3.18.3.5 Principal Component Analysis (PCA) Plot Why the others are wrong: A denies the analyst's role; C undoes the fix; D shows no components.
6. Correct answer: D. The Guide's own example: highly correlated advertising spend and distribution channels risk double counting sales. Correlation maps show pairwise dependence from -1 to 1 and, with p-values, help remove unwanted variables. Source: Guide to Business Data Analytics, 3.9 Descriptive and Inferential Statistics Why the others are wrong: A treats correlation as causation; B drops a variable without analysis; C adds noise.
7. Correct answer: A. The Guide cites an inverted U in residuals as a sign of a non-linear relationship. Unequal spread (heteroskedasticity) can point to subgroups. Source: Guide to Business Data Analytics, 3.18.3.6 Residual Plots Why the others are wrong: B ignores the structure; C is not visible in residuals; D is a validation issue.
8. Correct answer: C. Silhouette scores run from -1 to 1, and clear clusters have wide, long silhouettes. In the Guide's example, ten segments held only four or five dominant ones, so the analyst might reduce the number of clusters. Business direction shapes clustering choices. Source: Guide to Business Data Analytics, 3.18.3.8 Silhouette Plot Why the others are wrong: A weakens clusters further; B puts a request above evidence; D drops segmentation.
9. Correct answer: B. With 0.9% fraud, a model that flags nothing scores 99.1% accuracy, the same as X. The Guide says traditional measures fall short in fraud detection; more area under the ROC curve means a better classifier to fine-tune. Source: Guide to Business Data Analytics, 3.18.3.7 Receiver Operating Characteristics (ROC) Curve Why the others are wrong: A matches a trivial model; C has lower AUC than Y; D ignores the separating metric.
10. Correct answer: A. The Guide's course-correction example shows high training and low test scores with suspected confounding variables. Options are more data, a randomized test, or models suited to confounders. The analyst leads the cost-benefit, feasibility and impact review. Source: Guide to Business Data Analytics, 2.3.5 Assess the Analytics and System Approach Taken Why the others are wrong: B acts on an unreliable model; C chases a score; D hides the problem.
11. Correct answer: D. Analysts find constraints through elicitation and business rules. The Guide lists a key limitation: optimization needs accurate constraints, and wrong formulations waste time and resources. Source: Guide to Business Data Analytics, 3.14 Optimization Why the others are wrong: A ignores the formulation; B presents an unworkable plan; C changes the goal, not the constraint.
12. Correct answer: C. The Guide recommends business simulation for sparse data, such as a new launch, for what-if cases like a 10% price rise, and when live experiments are not feasible. Event-based simulation injects an event, such as a competitor launch. Source: Guide to Business Data Analytics, 3.1 Business Simulation Why the others are wrong: A lacks relevant history; B gives stated preference; D describes only the present.
How to score this set
- 11–12 correct: Move on to full timed practice.
- 8–10 correct: Reread the sections behind your misses, often 3.18 visuals and 3.9 expected value.
- Below 8: Reread 2.3.2 to 2.3.5 and the EDA scheme in 3.11, then retry in a week.
Next, read decision modeling and decision analysis or try our CBDA source data questions. For a full run, the CBDA exam simulator has 75 questions in exam and practice modes; sign up free to use it.
What candidates say about Analyze Data on the CBDA
Candidates describe the exam as harder than most question banks. The certificate.tips author of our CBDA certification overview flagged about 20 of 75 questions and found more statistical reasoning than definitions. Mufaddal Haidermota, certified in December 2022, recommends per-domain mock questions, full-length tests, mind maps and cheat sheets. The Techcanvass story of Ivan, a developer, credits mind maps and 2–3 hours of daily study. Trainer LN Mishra of Adaptive US, a practitioner, advises daily questions, timed mocks and reviewing wrong answers.
FAQ
How many Analyze Data questions are on the CBDA exam?
IIBA weights Analyze Data at 16%, which is about 12 of 75 questions. IIBA does not publish exact per-domain counts, so treat this as an estimate.
Do I need to calculate statistics by hand on the CBDA?
Expect light reasoning with numbers, not long formulas. You should compare a mean and a median, compute an expected value, and read metrics such as area under the ROC curve.
Which technical visualizations should I know for Domain 3?
Know the Guide's 3.18 set: autocorrelation, Box-Cox, contour, correlation map, PCA, residual, ROC and silhouette plots, and what a bad pattern looks like in each.
Who decides how to treat outliers?
The data scientist decides the technical treatment, such as exclusion or imputation. The analyst adds business context, for example whether an extreme value is a placeholder code.
Sources
- Certification in Business Data Analytics (CBDA) — IIBA, checked October 2026
- Certification fees — IIBA, checked October 2026
- CBDA Certification Handbook — IIBA, n.d.
- What is EDA? — NIST/SEMATECH e-Handbook of Statistical Methods, n.d.
- How can I detect non-constant variation across the data? — NIST/SEMATECH e-Handbook of Statistical Methods, n.d.
- Business Data Analytics Certification — Mufaddal Haidermota, Medium, January 2023
- CBDA certification journey: Ivan — Techcanvass, 2024
- Mastering the CBDA exam — LN Mishra, Adaptive US, October 2024