Executive Summaries - Business Analytics

1. Which conclusion and recommendation best reflect the evidence?
   Answer: The observed association is promising but may reflect account-manager selection; run a randomized holdout before automating discounts broadly. Whenever you see a question where an intervention was applied selectively rather than randomly, your first instinct should be to ask: could the people who chose who received the intervention have introduced bias? That's exactly the trap here, and it's called selection bias (or confounding by indication). Account managers — not a random process — chose which high-risk customers received discounts. Experienced account managers likely picked customers they believed were saveable: perhaps those who complained, asked questions, or seemed engaged. These customers may have been more likely to renew regardless of the discount. So the higher renewal rate among discount recipients could reflect who was selected, not what the discount did. C correctly names this threat and prescribes the right remedy — a randomized holdout experiment — before scaling the policy. This is the only response that is both analytically honest and practically constructive. A is wrong because it inverts the logic. Having similar churn scores does not eliminate selection bias; the selection happened after scoring, based on unobserved factors the model didn't capture. You cannot claim causation just because the groups looked alike on one variable. B is wrong on two counts: the data don't actually tell you the model identifies responders (that's a different, untested question), and restricting to high-risk customers was already the premise — excluding low-risk customers adds nothing meaningful here. D is wrong because it overcorrects. Predictive models absolutely can inform retention decisions; the issue is about inferring causation from observational data, not a flaw with prediction models themselves. Your study tip: when an exam question involves a non-random treatment assignment, always flag the possibility of confounding before accepting any causal claim — randomization is the gold standard cure.
2. Which summary most effectively combines the analytics into a recommendation and delivery plan?
   Answer: Pilot the two-employee shift at representative branches, monitor lunch waits and morning service levels, and expand to remaining branches if wait reductions are confirmed without breaching a predefined morning-service threshold. When a question asks you to evaluate an analytics-based recommendation for executive action, think in terms of the analytics-to-decision pipeline: descriptive findings → forecast → optimization → implementation. The critical question is whether the evidence justifies immediate system-wide action or whether model assumptions introduce enough risk to warrant a more cautious rollout. Here, the optimization model rests on a key conditional assumption — morning demand stays within its historical range. That's not a guarantee; it's a premise. The most responsible recommendation acknowledges that assumption, tests it at limited scale, and defines a clear trigger for full rollout. That's exactly what D does: pilot at representative branches, monitor both lunch waits and morning service levels, and expand only after confirming the predicted benefit without degrading morning performance. It's analytically honest, operationally prudent, and still moves toward action. A is tempting because it sounds decisive and summarizes the findings accurately — but recommending immediate system-wide implementation treats a model projection as a certainty. If the morning-demand assumption breaks down, the bank has staffing problems at every branch simultaneously with no diagnostic data from a pilot. B decentralizes the decision to branch managers without structure or criteria, which fragments implementation and makes it impossible to evaluate the model's effectiveness consistently. C delays action entirely by demanding more data before any step is taken — an overcorrection that ignores the actionable evidence already available. Collecting data is good; using it as an excuse for paralysis is not. Your strategic takeaway: when a model rests on key assumptions, the right business-analytics move is usually a structured pilot, not a system-wide rollout or indefinite delay.
3. Which statement is the most decision-useful executive summary and recommendation?
   Answer: The campaign generated sales but little profit and weaker retention; pause national expansion and test narrower offers aimed at customers with stronger repeat-purchase potential. When evaluating a business campaign, you should always look beyond top-line revenue and assess the quality of growth — specifically profitability, customer retention, and long-term value. This question tests whether you can synthesize multiple signals into a sound strategic recommendation rather than fixating on a single positive metric. The situation presents three red flags simultaneously: revenue grew $8\%$ but gross profit only $1\%$, acquisition costs jumped $18\%$, repeat purchase rates declined, and buyers were predominantly one-time coupon users. This pattern suggests the campaign is buying low-quality customers at high cost with minimal retention. Answer C correctly integrates all of these signals — acknowledging the revenue gain while recognizing the weak profitability and retention outcomes — and recommends a measured next step: pause national expansion and test more targeted offers. That's exactly the kind of evidence-based, risk-aware recommendation executives need. Answer A commits the most common analytical trap: cherry-picking the favorable metric (revenue growth) while treating the warning signs as secondary concerns. Expanding a struggling campaign nationally amplifies losses, not gains. Answer B is more nuanced but still flawed — "gross profit remained positive" conflates the level of profit with its trend, and reducing the coupon amount doesn't address the deeper retention problem. Answer D overcorrects entirely; abandoning acquisition spending without testing alternatives is too drastic given that the data doesn't yet support a full pivot. Study tip: On business analytics questions, watch for answers that isolate one metric while ignoring contradictory signals. Strong recommendations synthesize all the evidence and propose a testable, proportional next step — not an all-or-nothing decision.
4. Assuming benefits are zero when an initiative is unsuccessful, which recommendation should the executive summary make based on expected net value and feasibility?
   Answer: Select Initiative A because its capacity-adjusted expected net value is positive and exceeds the delayed expected net value of Initiative B. When evaluating competing analytics initiatives, you must integrate two factors: expected net value (probability × benefit − cost) and feasibility constraints that may alter those numbers before making a recommendation. Start with the raw expected net values. Initiative A: $0.80 \times \$700{,}000 - \$350{,}000 = \$210{,}000$. Initiative B (immediate): $0.50 \times \$1{,}100{,}000 - \$250{,}000 = \$300{,}000$. On paper, B looks better — but the passage tells you B requires eight analyst-months while only five are available. Delay is required, which reduces B's benefit by 60%, leaving $\$1{,}100{,}000 \times 0.40 = \$440{,}000$. The delayed expected net value of B becomes $0.50 \times \$440{,}000 - \$250{,}000 = -\$30{,}000$ — now negative. Initiative A, feasible immediately, retains its $\$210{,}000$ expected net value. That makes A the correct recommendation, confirming answer A. Answer B is wrong because it ignores the capacity constraint entirely and compares unadjusted figures — a critical analytical error in feasibility-aware decision-making. Answer C commits the same mistake by treating capacity as a footnote rather than a value-changing input; maximum possible benefit is irrelevant when you must account for real constraints. Answer D is wrong because expected-value analysis is specifically designed for probabilistic outcomes — uncertainty doesn't disqualify the method, it's the whole point of using it. Study tip: On business analytics questions, always check whether constraints (capacity, time, budget) modify the numbers before comparing alternatives. A recommendation that ignores feasibility is incomplete, regardless of how attractive the gross numbers appear.
5. Which recommendation is most appropriate for the executive summary?
   Answer: Lower the threshold only for cases ranked highest by expected net loss, capped at available review capacity, and measure fraud recovery and customer impact. When a model change creates more work than your team can handle, the right executive recommendation isn't "do it" or "don't do it" — it's "do it smartly within real constraints." This question tests whether you can balance model performance, operational feasibility, and customer risk simultaneously. The core problem is a capacity gap: the proposed threshold generates $18{,}000$ additional reviews, but the team can only absorb $10{,}000$. Simply lowering the threshold without addressing that gap means $8{,}000$ cases get either auto-approved (letting fraud through) or auto-declined (hurting legitimate customers). Answer C threads this needle correctly — prioritize the highest expected net loss cases up to the $10{,}000$ capacity ceiling, capturing most of the $\$400{,}000$ fraud recovery while protecting customers. Crucially, it also calls for measuring outcomes, which is sound analytics practice. Answer A fails because handing an infeasible workload to operations without a clear plan is irresponsible. "Let them figure it out" isn't executive guidance — it's delegation of an unresolved problem. Answer B overcorrects in the opposite direction: rejecting any recommendation that strains capacity ignores the real fraud losses you're leaving on the table. Feasibility constraints should shape a solution, not kill it entirely. Answer D sounds efficient but is actually harmful — auto-declining $18{,}000$ unreviewed transactions would wrongly block thousands of legitimate customers, creating regulatory and reputational risk that far outweighs the fraud benefit. Your strategy tip: whenever a business analytics question involves a model change with a resource constraint, always look for the answer that optimizes within the constraint rather than ignoring or surrendering to it.
6. Which recommendation best communicates uncertainty while remaining actionable?
   Answer: Begin a reversible trial and make the contract permanent only if weekly orders remain above a predefined demand trigger before the commitment date. When business analytics questions involve high forecast uncertainty, your job isn't just to pick a financial outcome — it's to recommend a decision process that manages risk while preserving upside. Think of this as the difference between a static choice and a dynamic, information-gathering strategy. Option D is the strongest recommendation because it directly addresses both the uncertainty problem and the need for action. Rather than betting everything on the base forecast or walking away entirely, it structures the decision as a staged commitment: run a reversible trial, collect real demand signals through weekly orders, and only lock in the permanent contract once those signals cross a predefined trigger. This converts a risky all-or-nothing decision into a sequential one where new information reduces uncertainty before the irreversible cost is incurred. That's precisely what unusually high forecast error calls for. Option A is tempting because $600{,}000 > 200{,}000$ looks favorable in expected-value terms, but it ignores the explicitly stated high forecast error. Approving a permanent commitment when your forecast reliability is compromised is analytically unsound. Option B overcorrects — rejecting a positive-expected-value opportunity simply because uncertainty is high isn't a rigorous recommendation; it forfeits real upside without exploring ways to manage the downside. Option C is the classic analyst cop-out. Presenting scenarios "without a recommendation" sounds neutral, but it abdicated your analytical responsibility. Executives expect analysts to synthesize information into actionable guidance, not just reformat it. A useful rule of thumb: when a question combines high uncertainty with a reversible information source, the right answer almost always involves staged decision-making — gather data, then commit.
7. Which recommendation should appear in the executive summary?
   Answer: Do not roll out the current offer because expected contribution per visitor declined; test a smaller or more targeted subsidy instead. When evaluating a business experiment, statistical significance tells you whether an effect is real — but it says nothing about whether that effect is profitable. The key metric for executive decisions is expected contribution per visitor, which combines both conversion rate and margin per order. Here's the math: before the offer, expected contribution per visitor = $0.10 \times \$30 = \$3.00$. After the offer, it's $0.106 \times \$27 = \$2.86$. The offer destroys $\$0.14$ of value per visitor. That's why B is correct — rolling out a subsidy that reduces expected contribution is a net negative, and the right path is to test a smaller or better-targeted subsidy that might recover conversion gains without fully eroding the margin. A commits the most common trap in A/B testing: confusing statistical significance with business significance. A result can be reliably real and still be financially harmful. The conversion lift is genuine — it's just not worth the cost. C is flawed because it assumes volume cures margin compression, which it doesn't. Scaling a negative-margin-per-visitor offer means losing more money, not less. Unless there's a strong customer lifetime value argument (not mentioned here), volume growth makes things worse, not better. D is wrong because opposing movements in two metrics aren't contradictory — they're exactly what a proper business evaluation synthesizes. You can and must evaluate them together using expected contribution per visitor. Study tip: On business analytics questions involving experiments, always ask: What happens to profit per unit of traffic? That single calculation usually resolves apparent conflicts between conversion and margin metrics.
8. Which executive recommendation best addresses the aggregate and segment-level findings?
   Answer: Roll out to small-business users and continue testing an enterprise-specific version before exposing the high-value segment broadly. When you see a question mixing aggregate trends with segment-level data, your instinct should immediately shift to Simpson's Paradox territory and segment value weighting. The core skill here is recognizing that an overall metric improvement can mask damage to your most valuable customer group — and that smart rollout decisions account for both. Option C is correct because it threads the needle between the two findings. Small-business activation improved meaningfully ($62\% \rightarrow 68\%$, a 6-point gain) among the majority of participants, giving you sufficient evidence to roll out confidently there. But enterprise activation fell ($78\% \rightarrow 69\%$, a 9-point drop) in a segment that produces substantially more lifetime value — and critically, the test wasn't sized to determine whether that decline was real or noise. Rolling out to small-business users captures proven gains while protecting high-value enterprise customers from a potentially harmful process until a properly powered, enterprise-specific test clarifies the picture. A is tempting because overall activation did rise, but it ignores that the aggregate improvement is largely driven by the larger small-business segment masking a serious enterprise decline. "Monitor after launch" is too risky when enterprise LTV is substantially higher. B overcorrects — rejecting the process entirely throws away a real, meaningful gain for small-business users based on inconclusive enterprise results. D sounds rigorous but is operationally paralyzing. Withholding a proven benefit from small-business users while waiting for enterprise data is unnecessarily costly. Your strategic takeaway: segment before you aggregate. On exam questions involving multiple subgroups, always ask whether overall trends are hiding divergent outcomes in high-value or high-stakes segments.
9. Which statement should lead the executive summary?
   Answer: The chatbot's success is uncertain because the favorable satisfaction result conflicts with operational deterioration and may contain response bias; revise and retest it. When evaluating business analytics results, your first instinct should be to check whether all available evidence tells a consistent story — and if not, to ask why before drawing conclusions. This question tests your ability to integrate conflicting data signals and recognize measurement threats like response bias. The passage presents a contradiction: satisfaction scores rose slightly ($4.0 \rightarrow 4.2$), yet first-contact resolution dropped sharply ($72\% \rightarrow 63\%$) and repeat contacts surged ($14\% \rightarrow 22\%$). Those operational metrics suggest customers are not getting their problems solved. Critically, survey response rates fell most among customers who contacted support multiple times — exactly the dissatisfied group most likely to rate the chatbot poorly. That self-selection means the satisfaction score is almost certainly inflated by response bias. Answer B correctly names both problems — the conflict between satisfaction and operational data, and the bias threatening the survey result — and recommends a cautious, evidence-based path: revise and retest rather than expand or kill the program prematurely. Answer A commits the classic trap of cherry-picking the favorable metric while treating the operational decline as someone else's problem. It ignores the bias issue entirely. Answer C overcorrects the opposite direction — it dismisses the satisfaction data without accounting for the possibility that satisfaction captures something real, and it jumps straight to removal without further investigation. Answer D sounds balanced but is actually passive; framing deteriorating operational metrics as merely "balanced" against a biased satisfaction score misstates the risk and delays a needed diagnostic response. The strategic takeaway: when data signals conflict, your executive summary should name the conflict and its cause, not paper over it. On analytics exams, answers that acknowledge measurement problems (bias, confounds, sample issues) almost always outrank answers that treat a single clean metric as conclusive.
10. How should the executive summary present the finding and recommendation?
    Answer: Highlight the association but state that nonrandom missingness may bias its magnitude; validate the carrier data before making a termination decision. When a dataset has missing values that aren't randomly distributed — what statisticians call non-random missingness or missing not at random (MNAR) — any estimate derived from the remaining data can be systematically biased. That's the core concept this question tests. Whenever you see missing data concentrated in a specific subgroup (here, one carrier serving a distinct region with a different customer profile), your first instinct should be: this missingness pattern could distort the finding. Option A is correct because it does exactly what responsible analysis requires. It honestly presents the $12\%$ association to executives while flagging that the $28\%$ missing timestamps come disproportionately from one carrier — the very carrier under review. That creates a potential selection bias: the observed effect may be overstated, understated, or even directionally wrong depending on what the missing records would show. Recommending validation before termination is analytically sound and protects the company from a costly, premature decision. Option B overcorrects. Missing data doesn't automatically invalidate analysis — it requires careful handling and transparent communication. Withholding the finding deprives executives of relevant information. Option C makes the opposite error: treating silence as confirmation. The fact that missing records "don't contradict" the finding is meaningless — absent data can't contradict anything. Presenting the estimate as reliable and recommending termination ignores the bias risk entirely. Option D applies mean imputation, which is generally a weak strategy and particularly problematic here because the missingness is non-random; replacing values with the overall mean would suppress the very variance that matters most. Your study tip: on business analytics questions involving missing data, always ask why the data is missing and who it's missing from — the pattern, not just the proportion, determines how much you can trust your estimates.

Start this quiz