- Which analytical question is the strongest translation of management's causal question using observational data?
Answer: Did returns for comparable affected products change relative to suitable unaffected products over the same period, after accounting for product mix and seasonality?
When management asks whether a packaging change caused an increase in returns, you should immediately think: what would a rigorous causal analysis require? The gold standard is a difference-in-differences framework — comparing the change in outcomes for an affected group against a control group over the same period, while controlling for known confounders. That logic should guide your evaluation of each option.
Option D is the strongest translation because it captures all three ingredients of credible causal inference from observational data: a treated group (affected products), a control group (comparable unaffected products), a before-vs-after comparison, and explicit adjustment for confounders like product mix and seasonality. This structure isolates the packaging change as the likely explanation for any observed difference.
Option A compares the supplier's post-change rate only to its own historical average. This ignores the seasonal shipping surge, which independently inflates returns across all suppliers — making the packaging change look worse than it is without a contemporaneous control.
Option B compares the affected supplier to suppliers with less fragile products after the change, but never establishes a baseline difference before the change. Since fragility already predicts higher returns, this comparison is biased from the start — you're not controlling for what was already different.
Option C tests correlation between packaging use and returns during the surge, but correlation doesn't isolate causation. It also confounds the packaging effect with the surge itself, since both happened simultaneously across the same products.
Study tip: On causal inference questions, always ask: Does this design have a control group, a time comparison, and confounder adjustment? Any option missing one of those elements is almost certainly a distractor.
- Which analytical question best supports a data-driven evaluation of the policy?
Answer: How did the western region's sales change relative to comparable untreated regions before and after implementation, assuming their prior trends were similar?
When evaluating whether a business intervention caused a result, you need to isolate the policy's effect from other forces affecting the outcome. This is the core challenge of causal inference, and it's exactly what this question tests. The key tool is a difference-in-differences framework: compare the treated group's change to a control group's change over the same period. If both groups were trending similarly before the policy, any divergence afterward can be more credibly attributed to the intervention.
Option C is correct because it asks precisely this question — how did the western region perform relative to comparable untreated regions, before and after implementation, assuming parallel prior trends? This design controls for the national sales boost caused by the competitor's exit, since that tailwind would have lifted both treated and untreated regions equally. What remains is a cleaner estimate of the policy's actual contribution.
Option A is tempting but flawed: knowing that western sales rose by a statistically significant amount doesn't tell you whether the competitor exit explains the entire gain. You'd see the same result even if the policy did nothing.
Option B compares performance against a static internal target, which is irrelevant to whether the policy specifically drove results. The target may have been set before the competitor exited, making it an arbitrary benchmark.
Option D demands that the western region outperform every untreated region — an unrealistically strict standard that ignores natural regional variation and doesn't account for confounders.
As a study strategy, whenever a question involves a policy or intervention, ask yourself: what is the counterfactual? If the analytical question doesn't establish a valid comparison group, it cannot support causal conclusions.
- Which analytical question most appropriately translates management's request into a prescriptive analytics problem?
Answer: For each eligible customer, which offer maximizes expected incremental contribution margin while respecting inventory and campaign constraints?
When you see a question asking you to translate a management decision into an analytical problem, ask yourself: does this framing optimize a decision, or merely describe/predict? Prescriptive analytics goes beyond prediction — it recommends the best action for each decision unit (here, each customer) subject to real business constraints like margin, budget, and inventory.
Answer A captures all three pillars of a well-formed prescriptive problem: a clear optimization objective (maximize incremental contribution margin), a customer-level decision (which offer for each customer), and explicit constraints (inventory limits, campaign rules). "Incremental" is crucial — it accounts for what the customer would have purchased anyway, which is exactly the cannibalization concern management raised. This makes A the correct answer.
Answer B fails because it optimizes for purchase probability alone, ignoring margin and inventory. A customer highly likely to buy at full price doesn't need a deep discount — sending one destroys margin without adding value. This is a prediction problem dressed up as a decision.
Answer C is a descriptive analytics question. It looks backward at past campaign data to find an association between discount level and revenue. It doesn't prescribe what to do for individual customers going forward, and association ≠ causal lift.
Answer D is also flawed in two ways: it ignores the counterfactual (what would happen without the offer) and focuses on basket size rather than margin. Without a baseline, you can't measure true incremental impact.
Study tip: On prescriptive analytics questions, look for three elements in the correct answer — an optimization objective, a decision variable, and constraints. If any of those is missing, the framing is likely predictive or descriptive instead.
- Which analytical question is most closely aligned with the chief financial officer's business objective?
Answer: What incremental customer contribution, net of variable costs and advertising spend, did each social channel produce relative to no exposure?
When a CFO asks whether paid social media is "working" for profitable customer acquisition, you should immediately translate that into three financial requirements: incrementality (would the customer have bought anyway?), net profitability (revenues minus all variable costs, not just ad spend), and channel-level attribution that reflects true causal impact. Any analytical question that ignores one of these dimensions is misaligned with the CFO's objective.
Answer D captures all three requirements. It asks for incremental contribution margin net of variable costs and advertising spend, compared against a no-exposure baseline. This directly answers "did the channel cause profitable new customers?" — accounting for customers who would have converted regardless and stripping out product and fulfillment costs to measure true profit.
Answer A fails because last-click platform reporting attributes every conversion that happened to click an ad, completely ignoring whether those customers were incremental. Platform-reported conversions are notoriously inflated and measure correlation, not causation.
Answer B improves on A by introducing cost efficiency, but it still treats every attributed acquisition as incremental — the passage explicitly flags this as a flaw. A channel with a low cost-per-acquisition looks great until you realize half those customers would have bought organically.
Answer C uses gross revenue divided by ad spend (a ROAS-style metric), but it ignores product and fulfillment costs, making a low-margin product look profitable. The passage specifically warns that "low-value first purchases" matter here — gross revenue is the wrong numerator.
Study tip: On business analytics questions, watch for answers that measure activity (clicks, attributed conversions) instead of incremental economic value. The CFO always cares about profit causally attributed to the investment, not platform-reported vanity metrics.
- Which analytical question best translates this expansion decision while recognizing the limitations of the existing-store data?
Answer: For each candidate city, what is the forecast distribution of incremental cash flow net of local costs, and which feasible portfolio has the highest expected value?
When evaluating a business expansion decision, the right analytical framework must do two things simultaneously: account for all relevant financial variables (not just revenue), and honestly acknowledge the limits of the data you're working with. The restaurant chain's existing data suffers from selection bias — stores were only opened in cities executives pre-approved, so you can't blindly extrapolate those relationships to fundamentally different markets.
Answer C handles both demands correctly. It asks for a forecast distribution of incremental cash flow net of local costs — meaning you're modeling full profitability (demand, rent, labor, competition, opening costs) rather than just revenue. Framing it as a distribution acknowledges uncertainty. Optimizing over a feasible portfolio ensures you're selecting the best combination of three cities together, not just the top three individually. This is the complete, rigorous formulation of the decision.
Answer A fails because it matches candidate cities to high-revenue stores while ignoring rent and opening costs — a city can look attractive on revenue while being unprofitable. Answer B is backward-looking and descriptive: identifying which current stores are profitable and correlating that with city traits is useful exploratory analysis, but it doesn't generate forward-looking forecasts for candidate cities, nor does it acknowledge selection bias. Answer D is the most dangerous distractor — it explicitly treats regression estimates from the biased existing-store data as "universally causal," which is precisely the trap the question warns against. Projecting those relationships onto structurally different candidate cities would produce unreliable forecasts.
When you see expansion or site-selection questions, watch for answers that optimize revenue alone, ignore data limitations, or mistake correlation for universal causation — C is right because it avoids all three pitfalls.
- Which analytical question is the best translation of the director's request?
Answer: What call arrivals and handling times are expected by half-hour using information available at the scheduling cutoff, and what staffing meets the service target?
When translating a manager's operational question into an analytical one, you need to match the scope, granularity, and timing constraints of the original problem. The director's request has three clear requirements: forward-looking demand estimates, half-hour-level detail, and a service quality target — not just cost minimization.
Choice C captures all three. It asks for projected call arrivals and handling times at the half-hour level, uses only information available at the scheduling cutoff (respecting the four-week lead time), and explicitly ties staffing to a service target. This is a proper workforce management framing: forecast demand at the right granularity, then solve a staffing model — like an Erlang-C queue — to meet the target.
Choice A fails because it is entirely backward-looking (last quarter's averages) and collapses all intraday variation into a single number. Scheduling for average load leaves customers waiting during peak half-hours. Choice B acknowledges the need for a quarterly forecast but then explicitly ignores intraday variation — the passage tells you that variation is substantial, so any solution that dismisses it is incomplete. Minimizing payroll without a service constraint also directly contradicts the director's concern about wait times. Choice D overcorrects in the opposite direction: staffing for the all-time historical peak would virtually eliminate waiting but at enormous cost, and "historical peak" is backward-looking rather than predictive.
A good study habit here is to treat every analytical translation question as a checklist: Does the question match the time horizon (future, not past)? The granularity (half-hour, not daily)? The objective (service level, not just cost)? Any answer missing one of these dimensions is wrong.
- Which analytical question most appropriately operationalizes delivery reliability for a valid comparison over time?
Answer: Using a consistent shipment definition, what proportion of eligible deliveries arrived after the customer-promised time in comparable periods, segmented by service level?
When measuring whether a business metric is genuinely changing over time, you need to control for confounding factors — changes in measurement tools, definitions, or population composition that could create false trends. This question tests whether you can identify a valid operational definition for a KPI (delivery reliability) amid several sources of measurement noise.
Answer A is correct because it addresses all three threats to validity simultaneously: it enforces a consistent shipment definition (neutralizing the reclassification of express shipments), restricts analysis to eligible deliveries (controlling for volume mix), and compares promised-time compliance across comparable periods segmented by service level — the actual operational definition of reliability.
Answer B is the most tempting trap. Complaint counts rise mechanically when you introduce automatic notifications — more customers are prompted to complain regardless of whether service worsened. Tracking raw complaint volume after a notification system change tells you about awareness, not actual performance.
Answer C introduces survivorship bias: it only studies customers who already complained. This self-selected group cannot represent overall delivery performance. You're measuring how bad the bad experiences were, not how common they were.
Answer D partially attempts a rate (complaints ÷ shipments), which is better than a raw count, but it uses each month's current definition of service type — meaning the denominator shifts whenever the express/standard reclassification occurs. Comparing inconsistent denominators over time invalidates any trend conclusion.
Study tip: When a question involves trend analysis, immediately ask: Has anything changed in how we're counting or defining things? If yes, the valid answer will control for those changes before drawing conclusions.
- Which analytical question correctly translates the request into a break-even evaluation?
Answer: Will the feature causally generate at least 15,000 incremental subscriber-months during the year compared with not launching it, so that the resulting contribution covers the annual cost?
Break-even analysis asks a deceptively simple question: at what volume does a new investment cover its own cost — and nothing more? The key word in the CFO's question is "pay for itself," which means you're looking for the point where incremental benefit equals incremental cost. To translate that correctly, you need three elements working together: an incremental metric (what the feature adds), a contribution rate (profit per unit), and a cost threshold to clear.
The math here is straightforward: $8 per subscriber-month$120,000=15,000 incremental subscriber-months. Answer B nails all three elements — it asks whether the feature causally generates at least 15,000 incremental subscriber-months, producing enough contribution margin to cover the annual cost. That's a textbook break-even formulation.
A is tempting because 15,000 is the right number, but it counts distinct subscribers at any moment, ignoring duration. A subscriber retained for one day counts the same as one retained for twelve months — that's not how contribution margin accumulates over time. The unit of value is subscriber-*months*, not subscriber-*headcount*.
C is a classic "total vs. incremental" trap. Including margin subscribers would have generated anyway overstates the feature's contribution and makes break-even look easier than it truly is. Break-even analysis demands incrementality.
D uses revenue rather than contribution margin, ignoring variable costs. Since the problem explicitly states that $8 already reflects contribution after variable costs, comparing gross revenue to the fixed cost is an apples-to-oranges error.
Study tip: Whenever a break-even question involves a rate per unit, immediately compute Break-even units=Contribution per UnitFixed Cost, then check whether each answer choice measures the right unit incrementally. - Which analytical question best translates the chief marketing officer's business question while addressing the change in customer mix?
Answer: For comparable acquisition cohorts, what proportion made a second purchase within 90 days, and how did that proportion vary by channel?
When translating a business question into an analytical question, your goal is to preserve the intent of the original question while controlling for any confounding factors. Here, the CMO wants to know if loyalty is genuinely weakening — but the influx of first-time customers from a new channel artificially deflates the repeat-purchase rate, since new customers haven't had time to return. This is a customer-mix problem, and the right analytical question must isolate loyalty behavior from acquisition volume.
Answer B does this correctly by defining comparable acquisition cohorts (groups of customers acquired in similar periods) and measuring the proportion making a second purchase within a fixed window — 90 days. This controls for customer mix, accounts for channel differences, and produces a like-for-like comparison of loyalty behavior over time. That's exactly what the CMO needs.
Answer A looks at raw monthly repeat-purchase counts, which are heavily influenced by how many customers were acquired — not whether loyalty changed. It doesn't control for mix at all.
Answer C restricts the sample to customers who already made at least two purchases, which introduces selection bias: you've excluded the very customers who didn't return, making loyalty look artificially strong.
Answer D measures repeat purchasing among "active" customers in a single recent quarter. This snapshot doesn't separate cohorts, doesn't control for channel, and doesn't track how loyalty has changed over time.
The key strategy: whenever a business metric shifts alongside a change in customer composition, the right analytical question must hold the mix constant — usually through cohort analysis or segmentation — before drawing conclusions about behavior. - Which analytical question provides the most defensible translation for the experiment?
Answer: Among assigned eligible customer accounts, does redesign assignment increase purchase incidence while keeping the refund rate within a prespecified acceptable limit?
When translating a business question into an analytical one for an experiment, you need to match three things: the right unit of analysis, the right outcome variables, and the right scope of inference. The executives asked about "completed purchases" and "refund problems" — so your analytical question must address both metrics and stay faithful to how the experiment was actually designed.
Option B does exactly this. Randomization happened at the customer-account level, so the analysis must also happen at that level — not sessions, devices, or completed purchases. B asks about "assigned eligible customer accounts," preserving the integrity of the randomization. It covers purchase incidence (the primary outcome) and anchors refund rate to a prespecified limit (a guardrail metric), which is precisely how responsible A/B tests handle secondary concerns without inflating false-positive risk through post-hoc thresholding.
Option A fails because it filters to sessions that reached the payment page and excludes non-compliant customers — this introduces selection bias, essentially analyzing a self-selected subgroup rather than the full randomized population (a violation of intent-to-treat principles).
Option C changes the outcome entirely. Average order value was never part of the executive question, and analyzing only completed purchases conditions on a post-treatment outcome, which is another form of selection bias.
Option D uses devices as the unit of analysis. Because customers can use multiple devices, device-level analysis breaks the randomization scheme and can double-count or misattribute treatment exposure. "No statistically significant change" also conflates absence of evidence with evidence of absence.
Study tip: In experiment design questions, always trace the chain — unit of randomization → unit of analysis → outcomes specified by the business question. Any mismatch is a red flag.
