AI Business Strategist
AIB-C01 Mock Exam 2
A harder, full-length, 85-item, blueprint-weighted independent mock exam for the AWS Certified AI Business Strategist (AIB-C01) beta exam, used as a timed readiness gate before you book.
This is the second full-length, blueprint-weighted independent mock exam for the AWS Certified AI Business Strategist (AIB-C01) beta exam. It is built from the publicly available exam guide and its task statements. It is not an official AWS practice exam, it contains no official exam questions, and it is not affiliated with, endorsed by or approved by AWS. AIB-C01 is a beta exam and the official practice exam is not available during beta. Treat this one as your timed readiness gate: it is deliberately harder than Mock Exam 1, with longer multi-constraint scenarios, more FIRST / BEST / MOST cost-effective qualifiers and more items where two options are defensible and the qualifier decides. All 85 items are new and do not repeat Mock Exam 1, the domain-page questions or the case studies.
Instructions
- Time: 170 minutes, matching the certification page’s stated duration. Note AWS’s own duration discrepancy (the exam guide says 130 minutes); sit this one timed at 170 minutes as a realistic readiness check.
- Items: 85, multiple choice and multiple response. Each item states how many answers to select.
- Selection: single-answer items use one choice; multiple-response items say Select two. Multiple-response items require all correct responses for credit — there is no partial credit.
- No guessing penalty: an unanswered item is scored incorrect, so answer every question and flag anything you want to revisit.
- Target: aim for at least 75% raw on this harder set before you book, as an independent readiness signal.
Domain distribution
| # | Domain | Weight | Items here |
|---|---|---|---|
| 1 | AI Fundamentals and Literacy | 24% | 20 |
| 2 | AI Strategy and Business Value Creation | 28% | 24 |
| 3 | AI Governance and Responsible AI Leadership | 24% | 21 |
| 4 | Business Readiness, Leadership, and AI Transformation | 24% | 20 |
| Total | 100% | 85 |
Score interpretation
The real exam is scored on a scaled 100–1,000 range with a 700 pass mark, and AWS does not publish how a raw percentage maps to that scaled score — the number of scored versus unscored beta items is also unpublished. So treat the scaled figure below as a rough indicative guide only, never as an exact conversion. A scaled 700 is not “70% correct”; use your raw percentage as the primary signal. Because this mock is harder than Mock Exam 1, expect a slightly lower raw score here than there.
| Raw score | Indicative scaled band | Reading |
|---|---|---|
| under 60% | well below the 700 line | Not ready; revisit your weakest domains, the case studies and the in-page questions. |
| 60–74% | approaching the 700 line | Close but not yet a safe margin; target your two weakest domains and re-sit. |
| 75–84% | around or above the 700 line | A reasonable margin on this harder set; review any weak domain and book with a buffer. |
| 85% and above | comfortably above the 700 line | Strong and consistent under the harder conditions; you are well prepared for the beta exam. |
Because scoring is compensatory, strength in one domain offsets weakness in another; you do not need to clear a bar in each domain. With four domains at 24–28%, though, a single weak domain is roughly a quarter of the exam, so one weak area is survivable and two is not.
Take the mock exam
Two ways to use the questions below. The interactive mode runs a timed sitting one question at a time, with a navigator, flagging and keyboard shortcuts, and ends with your raw score, an indicative scaled figure, a per-domain breakdown and a full correction. The review mode underneath lists every question with its options one per line and the answer hidden until you ask for it. Sit Mock Exam 1 first as your diagnostic, work any weak domain, then use this timed gate to confirm you are ready to book.
Interactive mode
Take the practice exam
85 questions · one at a time · 170-minute countdown · results with per-domain breakdown and full correction at the end. Your progress is saved in this browser if you leave the page.
By domain
| Domain | Correct | Score |
|---|
Correction
All questions (review mode)
Options are listed one per line. The answer and explanation stay hidden until you click Show answer. Use the interactive mode above for a timed sitting.
A logistics firm must decide, for each of three tasks, whether to use rule-based automation or AI: (a) apply a published fuel-surcharge percentage to invoices, (b) predict which shipments will be late from weather, traffic and historical delivery data, and (c) classify inbound customer photos of damaged goods. A leader proposes one generative model for all three. What is the BEST correction?
Show answer
Answer: B.
The surcharge is a fixed deterministic calculation (rules), late-shipment risk is a prediction from historical structured data (ML), and image classification needs computer vision, so each maps to a different solution class. B matches task to method. A optimises vendor count over fitness and applies a probabilistic model to a deterministic calculation. C mismatches the deterministic and predictive tasks. D cannot predict or classify images with fixed rules.
A board member argues that because a generative model produced a fluent, confident answer with citations, the answer must be reliable enough to publish without checking. What is the MOST accurate business-level correction?
Show answer
Answer: B.
Generative models can produce fluent prose and even invented citations that look authoritative but are false, so tone is not evidence of correctness. B is correct. A mistakes surface fluency for factual accuracy. C confuses reliability with a training technique. D assumes a bigger model removes the need for oversight, which it does not.
Two teams disagree. Team A says their weekly-changing pricing rules should be handled by fine-tuning a model; Team B says by retrieval-augmented generation. The rules change every week and must always be current. Which position is correct and WHY?
Show answer
Answer: B.
Weekly-changing rules that must stay current favour RAG, which reads a maintained source and updates as the source updates, without repeated retraining. B is correct. A and C misjudge fine-tuning, whose baked-in knowledge goes stale between costly retrains. D proposes stuffing context, which is expensive and still needs the current source.
A leader is told that a predictive maintenance model and a generative report-writing assistant 'both use AI, so they will fail and succeed for the same reasons'. Which distinction is MOST important for the leader to understand?
Show answer
Answer: B.
A predictive ML model and a generative assistant have different dominant failure modes and therefore different monitoring needs. B captures the distinction. A denies a consequential difference. C wrongly applies a generative context-window concept to a predictive model. D ignores that neither succeeds without monitoring and fit-for-purpose validation.
An operations lead wants to deploy an autonomous agent that can browse internal systems, call tools and take multi-step actions to resolve tickets end to end. Compared with a single-turn assistant, what is the MOST important governance implication a business leader should raise FIRST?
Show answer
Answer: B.
An agent that acts autonomously across systems can cause wider harm from a single error, so bounded authority, auditability and escalation are the first governance concern. B is correct. A reduces a control question to cost. C is dangerously wrong; autonomy increases oversight need. D fixates on the model and ignores the authority the agent is granted.
A data team reports that 40% of the customer records feeding a proposed churn model have missing or inconsistent contact history, and the fields cannot be reconstructed for six months. A sponsor insists the model launch next month regardless. What is the MOST defensible position?
Show answer
Answer: B.
Poor input data produces poor predictions regardless of model choice, so the sound options are to scope to the trustworthy segment or defer, not to launch on 40% unfit data. B is correct. A assumes the problem self-resolves. C cannot make a model overcome missing inputs. D hides the risk instead of managing it.
A knowledge team pastes an entire regulatory handbook into every prompt to 'be safe', and costs have tripled while answers now miss the specific clause asked about. Which explanation and remedy is BEST?
Show answer
Answer: B.
Stuffing everything into context inflates token spend and buries the relevant passage, degrading both cost and accuracy; targeted retrieval fixes both. B is correct. A blames the tool for a usage pattern. C contradicts the stated large handbook. D proposes a heavier, staler solution than the problem warrants.
A firm classifies AI tools as approved, blocked or under evaluation, but staff say the approved list is so short and slow to update that they keep using unlisted tools for real work. What does this reveal about managing shadow-AI risk?
Show answer
Answer: B.
Transparent classification curbs shadow AI only when the sanctioned option is practical and kept current; a starved, slow list pushes staff to unlisted tools. B is correct. A ignores the observed circumvention. C drives use onto personal devices with no control. D removes the very governance that manages risk.
A leader asks how ISO/IEC 42001 and ISO/IEC 23053 differ at a business level. Which statement is MOST accurate?
Show answer
Answer: B.
ISO/IEC 42001 is the certifiable management-system standard for responsible AI governance, and ISO/IEC 23053 is a framework for ML-based AI systems, so they serve different purposes. B is correct. A conflates two distinct standards. C confuses standards with pricing. D wrongly casts standards as risk-tier regulation.
A support assistant scores 90% aggregate accuracy but a review finds it performs at 95% for long-tenure customers and 68% for new customers, who were sparse in the historical training data. A manager wants to ship on the 90% aggregate. What is the BEST reasoning?
Show answer
Answer: B.
A strong aggregate can mask a data-representation gap that harms an important segment, so segment-level analysis and better representation come before shipping. B is correct. A hides the gap behind an average. C misattributes a data issue to context length. D treats a representation problem as a capacity problem.
A leader believes that once a model is trained it is 'finished' and independent of its training data. Which correction is MOST accurate for planning purposes?
Show answer
Answer: B.
A model carries forward the patterns and biases of its training data and assumes continuity with the past, so it degrades as the world shifts and must be monitored and refreshed. B is correct. A denies drift. C wrongly assumes autonomous improvement. D excludes predictive models, which depend heavily on their training data.
Which situation is the CLEAREST case where rule-based automation is the better choice than an AI model, even though AI is technically possible?
Show answer
Answer: B.
A fixed, published fee schedule keyed to days overdue is fully deterministic and cheaply expressed as rules, which are auditable and cannot drift. B is correct. A needs language understanding, C needs pattern learning for novelty, and D needs generative language, so all three are genuine AI cases, not rules cases.
A generative assistant gives inconsistent answers to the same question and a vendor suggests lowering the randomness setting to 'make it accurate'. A leader asks what to expect. What is the MOST accurate business-level answer?
Show answer
Answer: B.
Reducing randomness increases consistency but does not make outputs true, so grounding and review remain necessary for accuracy. B is correct. A confuses consistency with truth. C denies a real effect on variability. D confuses an inference setting with retraining.
A GenAI tool is being considered for two jobs: (a) answering staff questions from a policy library updated weekly, and (b) always producing output in a fixed legal template and formal house style that rarely changes. Which pairing of adaptation techniques is MOST appropriate?
Show answer
Answer: B.
Frequently changing knowledge suits RAG, while a stable, specialised style suits fine-tuning, so the two jobs call for different techniques. B pairs them correctly. A inverts the pairing. C ignores that neither job's core need is model size. D cannot answer policy questions without access to the current library.
A recommendation model's own outputs increasingly shape what customers click, and next quarter's training data is drawn from those narrowed clicks. Over several cycles, what is the MOST likely business consequence and the RIGHT response?
Show answer
Answer: B.
When a model's outputs shape the data it later learns from, a feedback loop can narrow behaviour and amplify bias, so monitoring and intervention are required. B is correct. A assumes a benefit the loop does not provide. C is unrelated to the data dynamic. D misapplies a generative context concept to the loop.
A leader asks why a proposed image-based quality-inspection use case is 'harder to get right' than a proposed spreadsheet-based sales-forecasting use case, given both are 'just AI'. What is the MOST useful business-level explanation?
Show answer
Answer: B.
Structured tabular data suits well-established techniques, whereas unstructured images usually require more specialised handling and preparation, so the data type shapes effort and risk. B is correct. A denies a real distinction. C reverses the usual effort. D reduces a data-type difference to row counts.
A vendor claims their generative model 'cannot hallucinate because it only uses your documents'. The tool answers from a document store via retrieval. What is the MOST accurate business-level view a strategist should hold?
Show answer
Answer: B.
Grounding responses in retrieved documents lowers hallucination risk but does not remove it, because the model can still misread or over-reach, so grounding checks and review remain necessary. B is correct. A overstates the guarantee. C wrongly drops oversight. D ignores that retrieval, not only fine-tuning, addresses grounding.
A leadership team is deciding, for four tasks, which are TRUE cases for machine learning or generative AI rather than fixed rules. Select the TWO tasks that genuinely require AI rather than rule-based automation.
Show answer
Answer: A and B.
Predicting failure from complex sensor patterns needs ML (A), and understanding and answering from free text needs generative AI (B). C is a deterministic calculation, D is a fixed threshold, and E is arithmetic, so all three are rule-based tasks where AI would add cost and risk without benefit.
Before relying on a supplier-risk prediction model, which TWO questions about its DATA most directly determine whether its outputs can be trusted?
Show answer
Answer: A and B.
Trust in predictions rests on representativeness (A) and on data currency, quality and refresh (B); both directly govern whether the outputs hold. C, D and E are irrelevant to data fitness and do not affect whether the model's predictions can be relied upon.
An autonomous agent will be allowed to reschedule customer deliveries and issue small goodwill credits without a human in the loop. Which TWO safeguards should a business leader insist on FIRST, before go-live?
Show answer
Answer: A and B.
Autonomy over money and customer commitments demands bounded authority with escalation (A) and full auditability with drift detection (B). C is cosmetic, D is a performance tweak, and E addresses conversation length, none of which controls the risk of unsupervised autonomous action.
A strategist has one funding slot and three proposals: (1) a copilot that saves clerical time but any competitor can buy the same tool, (2) an initiative that pairs AI with the firm's proprietary claims data to price risk better than rivals, and (3) a flashy demo with no defined business outcome. Business value, feasibility and strategic alignment are the criteria. Which should win the slot?
Show answer
Answer: B.
With one slot, the deciding criterion is strategic, differentiated value, and only proposal 2 pairs AI with hard-to-copy proprietary data for durable advantage. B is correct. A delivers commodity savings any rival can match. C has no defined outcome. D dilutes a single slot across unequal proposals, funding the weak ones.
A capability is core to the firm's differentiation, no mature vendor offers it, the firm has scarce but capable ML talent, and a two-year horizon is acceptable. Applying build-buy-partner, what is the MOST appropriate decision?
Show answer
Answer: B.
When a capability is differentiating, no adequate product exists, talent is available and the timeline permits, build is the right choice to own the advantage. B is correct. A cannot buy what no vendor offers and would surrender differentiation. C shares a core advantage with a rival. D discards a strategically justified build on a blanket risk claim.
An initiative has consumed $800,000 over 18 months, still misses its target, and its sponsor argues 'we have invested too much to stop now'. New analysis shows the remaining path would cost another $600,000 with only a 20% chance of hitting the target. What is the MOST rational disposition?
Show answer
Answer: B.
The $800,000 is a sunk cost irrelevant to the forward decision; spending $600,000 for a 20% chance is a weak expected return, so terminate or rescope. B is correct. A commits the sunk-cost fallacy. C throws more money at poor odds. D scales an unproven capability to chase past spend.
A team is about to launch an AI initiative but has captured no before-state metrics for the process it will change. Leadership will later demand proof of value. What is the single MOST important action to take BEFORE go-live?
Show answer
Answer: B.
Value can only be proven against a baseline, and the one moment to capture it cleanly is before the change, so establishing baseline metrics now is essential. B is correct. A reconstructs from memory, which is unreliable. C substitutes the vendor's numbers for your own process. D forgoes the quantitative evidence leadership will demand.
A support-automation pilot cut average handle time from 9 to 6.5 minutes across 240,000 contacts per year at a loaded cost of $32 per hour. The solution costs $180,000 per year plus a $90,000 one-off in Year 1. What is the approximate Year-1 ROI?
Show answer
Answer: A.
Time saved is 2.5 minutes over 240,000 contacts = 600,000 minutes = 10,000 hours; at $32 that is $320,000 benefit. Year-1 cost = 180,000 + 90,000 = 270,000. ROI = (320,000 − 270,000) / 270,000 ≈ 19%, so A is correct. B, C and D overstate the return because they understate the Year-1 cost, which consumes most of the $320,000 benefit.
A workload has a steady, predictable, high monthly volume that runs continuously, and finance wants the lowest total cost. Which AWS AI pricing approach is MOST cost-effective at a strategic level, and why?
Show answer
Answer: B.
Steady, predictable, high utilisation is exactly when a commitment-based discount lowers total cost, because the committed capacity is fully used. B is correct. A forgoes the discount available for predictable volume. C prices per user rather than per usage of this compute workload. D pays a premium with no benefit for steady traffic.
Two AI initiatives clear the ROI bar. Initiative X returns more cash but depends on data the firm may lose access to next year; initiative Y returns less cash but is sustainable and aligned to a long-term strategic priority. The firm can fund only one. Which is the BEST choice and why?
Show answer
Answer: B.
Prioritisation balances value with feasibility, sustainability and strategic alignment, and X's higher return is undermined by a fragile data dependency, so Y is the sounder bet. B is correct. A optimises raw return while ignoring sustainability. C abandons two qualified initiatives for an unspecified one. D bets on a dependency that may disappear.
An application sends 2,000,000 requests per month. 60% are simple and could be served by a smaller model at $0.004 each instead of the standard $0.011 each; the remaining 40% must use the standard model. Routing the simple traffic to the smaller model reduces total inference cost by approximately how much?
Show answer
Answer: A.
Baseline = 2,000,000 × $0.011 = $22,000. With routing, 1,200,000 simple × $0.004 = $4,800 plus 800,000 standard × $0.011 = $8,800, totalling $13,600. Saving = 22,000 − 13,600 = $8,400, about 38%. A is correct. B, C and D do not match the arithmetic; the saving is the $8,400 difference on the $22,000 base.
A finance leader wants to (a) forecast next year's AI spend before committing budget and (b) investigate why last month's actual AI spend spiked. Which AWS tools fit these two needs, at a strategic level?
Show answer
Answer: A.
The Pricing Calculator produces forward estimates and Cost Explorer shows actual spend, trends and anomalies, so A pairs each need with the right tool. B swaps their roles and misuses Marketplace, which is for evaluating buy/partner options. C misassigns both to a procurement surface. D uses the estimator for a job it cannot do.
An initiative needs $120,000 upfront and is expected to return $70,000 in net cash each year for three years. At a 12% discount rate, what is the approximate three-year NPV, and does it clear a zero-NPV bar?
Show answer
Answer: A.
Discounted inflows: 70,000/1.12 ≈ 62,500; 70,000/1.2544 ≈ 55,804; 70,000/1.4049 ≈ 49,825; sum ≈ 168,129, minus the 120,000 outlay ≈ 48,129, about $48,000 and positive, so A is correct. B overstates the return. C wrongly turns a positive NPV negative. D ignores discounting and the upfront cost.
A leader wants to lock in a six-month provisioned-throughput commitment for a new GenAI feature whose adoption is entirely unproven and could be near zero for months. What is the BEST cost posture at launch, and why?
Show answer
Answer: B.
For unproven, possibly near-zero demand, consumption pricing avoids paying for idle committed capacity; commit only after utilisation proves high enough. B is correct. A risks paying for capacity that goes unused, turning a discount into a loss. C amplifies that risk. D charges per seat, unrelated to the feature's actual usage.
In a mature, slow-moving industry where an AI capability offers only marginal, easily copied gains, a rival has announced a large AI programme. What investment level is MOST appropriate for the firm?
Show answer
Answer: B.
Investment level should track the capability's potential and industry dynamics; a marginal, easily copied gain in a mature industry warrants modest, selective spend, not a spending race. B is correct. A mirrors a rival without regard to return. C over-invests in a marginal capability. D forecloses even a modest, justified investment.
A team reports that an assistant 'saved 12,000 hours this year' and asks the CFO to book the full amount as cash savings. Investigation shows most of the freed time was absorbed as slack rather than removed as cost or redeployed to revenue work. What is the MOST honest treatment?
Show answer
Answer: B.
Hours become cash only when they avoid cost or are redeployed to value; slack that stays as slack is not cash, so only the realised portion should be booked, with assumptions disclosed. B is correct. A overstates by treating all freed time as cash. C understates a genuine productivity gain. D inflates the figure with an inappropriate rate.
A firm wins early advantage from an off-the-shelf AI tool, but within a year competitors buy the same tool and the edge disappears. What does this MOST clearly teach about sustainable competitive advantage from AI?
Show answer
Answer: B.
An advantage built only on a purchasable tool evaporates once rivals buy it; durability requires pairing AI with hard-to-copy proprietary assets. B is correct. A misreads timing as the issue. C over-generalises to deny any advantage. D chases a bigger model, which rivals can also adopt.
A firm is migrating a live, revenue-critical process from one AI platform to another to cut cost. The new platform is 20% cheaper but unproven on the firm's data. What is the MOST important consideration to protect the business during the transition?
Show answer
Answer: B.
For a revenue-critical process, continuity outweighs a modest saving, so a phased transition with validation and fallback is essential before full switch. B is correct. A risks the whole process on an untested cutover. C removes the safety net prematurely. D discards the validation that catches problems before customers feel them.
A document-review pilot cut processing time from 25 to 18 minutes across 90,000 documents per year at $45 per hour. The solution costs $120,000 per year plus a $60,000 one-off in Year 1. What is the approximate Year-1 ROI?
Show answer
Answer: A.
Time saved = 7 minutes × 90,000 = 630,000 minutes = 10,500 hours; at $45 that is $472,500 benefit. Year-1 cost = 120,000 + 60,000 = 180,000. ROI = (472,500 − 180,000) / 180,000 ≈ 163%, so A is correct. B and C understate the return by underestimating the hours saved across 90,000 documents, while D overstates it by ignoring the Year-1 cost.
A pilot is tracked only by an annual revenue figure reported at year end, and mid-year no one can tell whether it is on course. Leadership is frustrated by the lack of visibility. What is the BEST fix for the NEXT initiative?
Show answer
Answer: B.
The gap is the absence of leading indicators that give early, steerable signal; adding them lets the team correct course before the lagging metric lands. B is correct. A reports the same lagging number more often, still with no early signal. C abandons measurement. D confuses a measurement gap with a model choice.
An AI initiative is technically feasible and shows strong projected value, but a pending regulation could ban its intended use in the firm's largest market within a year, and legal cannot yet say how it will land. Applying scale-pause-terminate, what is the MOST appropriate disposition now?
Show answer
Answer: B.
A valuable but regulation-exposed initiative should be paused or narrowed to a compliant scope while the legal position clarifies, preserving optionality without taking illegal risk. B is correct. A scales into possible prohibition. C kills a potentially viable initiative prematurely. D proceeds in disregard of a live legal threat.
A firm must choose between a managed foundation-model platform (Amazon Bedrock) and building a custom model on Amazon SageMaker AI for a need that a strong existing model already meets well. A senior engineer wants to build custom 'to keep the skills in-house'. What is the MOST sound strategic reasoning?
Show answer
Answer: B.
When an existing model already meets the need, the managed platform is faster and cheaper, and custom builds should be reserved for differentiated problems off-the-shelf models cannot solve. B is correct. A subordinates a business decision to skills development. C wrongly forecloses custom ML entirely. D decides on an out-of-scope UI preference.
A product team wants business users to get AI-assisted insights and build simple apps over their own data without a development project, and the commercial model is per user per month. Which AWS surface fits this at a strategic level, and what pricing structure does it imply?
Show answer
Answer: B.
Amazon Quick is the AI-powered business-assistant and BI surface for business users with a seat-based commercial model, matching the described need without a build project. B is correct. A is a custom-ML build tool on instance pricing. C is a commitment for provisioned model capacity. D is a procurement marketplace, not a business-user assistant.
A CFO accepts a hard ROI only from clearly monetisable effects and asks the strategist to separate those from softer benefits in a proposal. Which classification is CORRECT?
Show answer
Answer: B.
Tangible benefits are directly monetisable, so headcount cost avoidance and incremental revenue belong in the hard ROI, while goodwill and brand perception are real but intangible and belong in a separate category. B is correct. A and D miscategorise intangibles as cash. C discards genuine tangible value the CFO would accept.
A strategist is building the case for a customer-service AI pilot and wants LEADING indicators to steer it in the first eight weeks. Which TWO are genuine leading indicators rather than lagging outcomes?
Show answer
Answer: A and B.
Leading indicators are early, steerable signals that predict success: weekly end-to-end handling (A) and week-over-week resolution (B) can both be acted on mid-pilot. C and D are lagging annual outcomes that arrive too late to steer, and E is the vendor's benchmark, not a signal from your own pilot.
A vendor proposal is 30% cheaper than the alternatives but trains on the firm's proprietary data, provides no export of the fine-tuned model, and has a one-year auto-renewing lock-in. Which TWO risks should MOST weigh against the low price?
Show answer
Answer: A and B.
The material risks are IP and confidentiality exposure from training on proprietary data (A) and lock-in from no export and auto-renewal (B); either can outweigh a 30% saving. C, D and E are irrelevant to the strategic risk of the proposal and should not drive the decision.
A steering group is deciding when AI is NOT the right solution for a proposed use case. Which TWO conditions MOST strongly argue against using AI at all?
Show answer
Answer: A and B.
AI is the wrong tool when a fixed rule already solves the task cheaply (A) or when the data it needs is absent or unfit and cannot be obtained in time (B). C, D and E are exactly the conditions where machine learning or generative AI add value, so they argue for AI, not against it.
A hiring-screening model raises overall efficiency but an audit finds it systematically down-ranks candidates from one region that was under-represented in the historical hiring data. Leadership wants to keep the efficiency gain. Which responsible AI dimension is MOST at stake and what is the BEST action?
Show answer
Answer: B.
Systematically disadvantaging a group is a fairness problem that must be investigated and mitigated with oversight, and an efficiency gain does not justify it. B is correct. A trades a discriminatory outcome for speed. C and D name performance attributes that have nothing to do with the disparate impact identified.
A business must choose between a slightly more accurate black-box model and a marginally less accurate but explainable model for automated consumer-credit decisions in a regulated market. Two managers argue accuracy should always win. What is the BEST position?
Show answer
Answer: B.
In a regulated, individually consequential decision, explainability is a requirement that a marginal accuracy gain cannot override. B is correct. A optimises accuracy while ignoring the duty to explain. C decides on cost alone. D discards a legitimate, well-governed use rather than choosing the appropriate model.
A project team plans to design privacy protections, bias testing and human-review checkpoints into the initiative from the planning stage, before any build. A cost-conscious sponsor wants to defer all of that to a pre-launch compliance sweep to save time now. What is the BEST argument for the team's approach?
Show answer
Answer: B.
Governance by design integrates controls from the start, avoiding costly retrofits and last-minute launch-blocking findings, so it is cheaper and safer than a late sweep. B is correct. A ignores the retrofit cost and risk. C treats accuracy as a substitute for fairness and privacy work. D dismisses a core safeguard.
A leader assumes that adopting a fully managed AWS AI service transfers responsibility for the appropriateness and outcomes of the AI to AWS. Under the shared responsibility model for AI workloads, what remains the CUSTOMER's accountability?
Show answer
Answer: B.
AWS secures the cloud, but the customer stays accountable for use-case appropriateness, output fitness, oversight and process compliance even on a managed service. B is correct. A, C and D are AWS's security-of-the-cloud duties that never transfer to the customer and are not the accountability in question.
A generative claims assistant occasionally invents policy clauses that do not exist, and a leader wants the MOST direct safeguard against these fabricated statements without abandoning the assistant. Which is BEST at a strategic level?
Show answer
Answer: B.
Fabricated clauses are hallucinations, and contextual grounding checks verify output against a source and flag unsupported claims, with human review for high-impact cases. B is correct. A may hold more text but does not verify grounding. C addresses speed, not truth. D is a commercial model unrelated to hallucination control.
An organisation applies a risk-classification framework across its AI portfolio. A proposed tool makes automated, individually consequential decisions about customers but a team wants it in the lowest tier 'to move fast'. What is the BEST treatment and why?
Show answer
Answer: B.
Automated, individually consequential decisions are high impact and warrant the strongest, proportionate control set. B is correct. A under-controls a high-impact use case for speed. C skips the classification the framework requires. D equates a consequential external decision with a trivial internal tool.
A team argues that because a vendor offers an IP indemnity, generated marketing content can be published without factual or brand review. What is the flaw in this reasoning?
Show answer
Answer: B.
An indemnity addresses some legal exposure but not factual accuracy, brand fit or reputational harm, so review is still required. B is correct. A and C overstate what an indemnity does. D over-corrects by banning a legitimate use rather than adding a review step.
A firm deploys an AI process that makes automated decisions affecting individuals in a jurisdiction with risk-tiered AI regulation, and assumes compliance can wait until a regulator raises a concern. What is the BEST corrective stance?
Show answer
Answer: B.
Obligations attach to the risk tier of the use case and apply from the outset, so compliance must be addressed from planning through the lifecycle. B is correct. A waits for enforcement. C addresses obligations too late, after individuals may be affected. D wrongly outsources accountability that stays with the deploying business.
An organisation has an AI ethics charter, an annual training slide and a mailing list, but no named owner of AI risk, no decision rights, and no path to review an unusual high-impact case quickly. How is this BEST characterised, and what is the priority fix?
Show answer
Answer: B.
Artefacts without ownership, decision rights or an operating path are governance in name only, and the fix is to create real accountability and an exception process. B is correct. A mistakes documents for a working structure. C and D misclassify a governance-structure failure as a technical or commercial issue.
A leader is told to 'apply a risk management framework' to the AI portfolio and asks how the NIST AI Risk Management Framework helps at a business level. Which description is MOST accurate?
Show answer
Answer: B.
The NIST AI RMF structures risk work into Govern, Map, Measure and Manage across the lifecycle, helping leaders prioritise and act. B is correct. A confuses a framework with pricing. C mistakes it for a model design. D contradicts the framework, which reinforces rather than removes oversight.
A deployed assistant was set up with only infrastructure uptime monitoring. Months later it is giving outdated and occasionally incorrect answers, but dashboards show 100% uptime. What governance gap does this expose, and what is the FIRST fix?
Show answer
Answer: B.
Uptime says nothing about answer quality, so the gap is the absence of output-quality and drift monitoring, and the first fix is to add it with a remediation path. B is correct. A misreads uptime as a compute-sizing issue. C invents a technical cause. D over-reacts; the fix is monitoring, not abstention.
A central AI review board must approve every AI request, meets monthly, and low-risk requests wait weeks; teams have begun using unapproved tools to avoid the delay. What is the BEST governance redesign?
Show answer
Answer: B.
Governance that reviews everything at one slow gate breeds shadow AI; a risk-tiered process that fast-tracks low-risk work and scrutinises high-risk work fixes the bottleneck while keeping control. B is correct. A intensifies the bottleneck. C forfeits value. D removes the controls that manage real risk.
A firm wants to deploy an AI tool whose greatest value comes from a use that its own responsible AI policy on transparency would restrict, and a leader proposes quietly disabling the transparency notice to capture the value. What is the BEST resolution of this conflict?
Show answer
Answer: B.
When business goals conflict with a responsible AI principle, the principle is a constraint to design within, so the team should find a compliant path rather than covertly disabling transparency. B is correct. A sacrifices a core principle for value and invites harm and liability. C over-reacts by abandoning the tool. D defers action until harm surfaces.
A leader lists AWS's core responsible AI dimensions as only fairness, explainability and privacy, and treats the rest as optional extras. How should this be corrected at a business level?
Show answer
Answer: B.
AWS publishes eight core responsible AI dimensions, and the exam guide's shorter list is a subset, so dimensions like safety, controllability and governance are integral rather than optional. B is correct. A understates the count. C denies the published set. D confuses principles with pricing.
A leader must explain the difference between controllability and veracity as responsible AI dimensions to a non-technical board. Which explanation is MOST accurate?
Show answer
Answer: B.
Controllability is about steering and overseeing behaviour, while veracity and robustness concern truthful, reliable output, so they are distinct. B is correct. A conflates two dimensions. C misattributes both to unrelated attributes. D swaps their meanings.
A model that passed fairness testing at launch begins producing more errors for a demographic group after nine months as the customer base shifts. A manager calls it a one-time bug to patch and close. What is the MOST accurate view and the RIGHT response?
Show answer
Answer: B.
Fairness at launch does not guarantee fairness later; as populations shift, bias can drift, so ongoing monitoring and remediation are required rather than a one-off patch. B is correct. A treats a recurring lifecycle risk as closed. C misclassifies it as a cost issue. D ignores that launch fairness was genuine and drift is managed by monitoring.
A single retail assistant both answers product questions and can process returns up to a value limit. A team proposes governing both capabilities under one blanket low-risk policy for simplicity. What is the BEST governance approach?
Show answer
Answer: B.
Governance should be proportionate to each capability's risk, and an action that moves money differs sharply from an informational answer. B applies controls where the risk is. A under-controls the financial action. C over-controls the harmless capability. D leaves the highest-risk capability uncontrolled.
A governance committee is deciding when human oversight of an AI decision must be MANDATORY rather than optional. Which TWO conditions make it mandatory?
Show answer
Answer: A and B.
Oversight is mandatory when decisions are high-consequence or irreversible (A) and when regulation demands accountability (B). C is a reason automation is attractive, not a reason to remove oversight. D is a cost fact and E a popularity fact, neither of which bears on the need for human oversight.
Before signing with a generative-content vendor, which TWO questions MOST directly manage intellectual-property and confidentiality risk?
Show answer
Answer: A and B.
IP and confidentiality risk turn on whether your data is used for training and what happens to it (A) and on the indemnity, provenance and originality of the output (B). C, D and E are cosmetic or vanity facts that do not touch the legal exposure a leader must assess.
A content-moderation model over-blocks lawful posts from one community while letting genuinely harmful posts from another slip through. Which TWO responsible AI dimensions are MOST directly implicated?
Show answer
Answer: A and B.
Uneven treatment across communities is a fairness failure (A) and letting harmful content through is a safety failure (B). C, D and E are performance, cost or usability attributes, not responsible AI dimensions, and none captures the disparate treatment or the harm getting through.
A leader wants a functioning AI governance structure rather than a paper charter. Which TWO elements are MOST essential to make it work?
Show answer
Answer: A and B.
Functioning governance needs named accountability and decision rights (A) and cross-functional representation with a working exception path (B). C is a technical choice unrelated to governance. D is a bottleneck that breeds shadow AI. E over-restricts rather than governing use proportionately.
A readiness assessment scores leadership alignment 8, technical infrastructure 8, governance 7, cultural preparedness 7 and data quality 2, with fragmented ownership and heavy silos. Leadership wants to scale AI across five functions next quarter. What should be prioritised FIRST, and why?
Show answer
Answer: B.
Readiness behaves like a weakest-link function, so a data-quality score of 2 is the binding constraint that must be raised before scaling. B is correct. A scales five functions onto a broken data foundation. C invests where the firm is already strong. D cannot make a model overcome fragmented, poor-quality data.
A company has run four disconnected pilots in four departments, each with its own tools and no shared strategy, and now calls itself 'enterprise-scale in AI'. How should a strategist characterise its maturity and what is the RIGHT next step?
Show answer
Answer: B.
Disconnected pilots without shared strategy or foundations are experimentation, not enterprise scaling, so the next step is a strategy and repeatable pattern. B is correct. A mislabels the maturity stage. C multiplies disconnected experiments. D is plainly premature.
A leader wants to close capability gaps for AI transformation and asks which AWS framework helps identify gaps across people, process, technology and governance dimensions of the organisation. Which is the BEST fit?
Show answer
Answer: B.
AWS CAF and its perspectives are the tool for finding organisational capability gaps when planning AI transformation. B is correct. A is a cost-estimation tool. C is model documentation. D is a contractual availability commitment; none of these frames organisational capability gaps.
A firm at an early, experimentation maturity stage must choose its FIRST major transformation investment between (a) an advanced enterprise MLOps platform at $500,000 and (b) a $150,000 programme to raise data quality and AI literacy and appoint data owners. Which is the better first investment and why?
Show answer
Answer: B.
At the experimentation stage the binding constraints are usually data quality, literacy and ownership, and advanced MLOps tooling delivers little until those foundations exist. B is correct. A buys advanced tooling ahead of the maturity to use it. C forgoes a needed investment. D over-spends without regard to sequencing.
A successful single-site pilot cannot progress to enterprise deployment despite strong results. The model is accurate and well-liked, but there is no cross-business data readiness, no owner for enterprise rollout and no production operating model. What is the MOST likely root cause?
Show answer
Answer: B.
A pilot that succeeds but cannot scale usually lacks the organisational foundations, enterprise data readiness, ownership and an operating model, that scale requires. B is correct. A contradicts the stated accuracy. C and D address budget and publicity, not the readiness gap that blocks scale.
A CEO plans to announce that an AI assistant will let a team 'do more with fewer people' and then roll it out to that team the same week, with no communication about how roles will change. Adoption in a prior rollout collapsed under similar conditions. What is the BEST leadership approach?
Show answer
Answer: B.
Pairing a headcount message with an unexplained rollout poisons adoption, so leaders should communicate early, address fears honestly and frame the role shift toward higher-value work. B is correct. A ignores the human system that drives adoption. C lets fear fester. D makes a promise that is not credible.
A programme lead references the AWS Cloud Adoption Framework's transformation phases and asks a strategist to confirm the correct order for planning an AI transformation. Which order is correct?
Show answer
Answer: B.
AWS CAF runs its four phases in the order Envision, Align, Launch, Scale. B is correct. The exam guide's looser example wording says 'envision, experiment, launch, scale', but CAF's own second phase is Align. A, C and D scramble the order.
A pilot cut route-planning time by 28% at one of 60 depots. Leadership wants it live at all 60 within a quarter, but the other depots record data in incompatible local spreadsheets with no shared definitions. What is the BEST decision?
Show answer
Answer: B.
The pilot proved feasibility in one place, but inconsistent cross-depot data must be standardised before a phased, gated rollout. B is correct. A scales onto incompatible data. C over-reacts to a solvable problem. D relies on unrepeatable manual effort across 60 sites.
A firm scaling AI across many teams keeps rebuilding the same governance templates, evaluation approaches and reusable components from scratch in each project, making every new use case slow and costly. What organisational mechanism MOST directly addresses this?
Show answer
Answer: B.
Repeatedly rebuilding the same assets is exactly what an AI center of excellence solves by pooling expertise, standards and reusable components. B is correct. A adds an approval gate, not shared assets. C is a purchasing decision. D removes the coordination that enables reuse.
As AI takes over routine contact-centre queries, a leader must frame the change to the agents. Which framing BEST fits the intended human-role transition and is most likely to sustain adoption?
Show answer
Answer: B.
The intended transition moves people toward oversight and higher-value work that leverages human strengths while AI handles routine volume. B is correct. A is harmful and not the design intent. C denies a real change. D pits people against AI on the dimension AI is best at.
A cross-functional AI team includes business, technical, legal and compliance members, but 'everyone is jointly responsible' for outcomes and no single person can make or own a decision. Delivery keeps stalling. What is the problem and the fix?
Show answer
Answer: B.
When everyone owns an outcome no one is accountable and decisions stall, so the fix is clear owners and decision rights within the cross-functional structure. B is correct. A romanticises diffuse ownership. C adds people without fixing accountability. D discards a needed structure instead of clarifying it.
A strategist must recommend how to BEGIN scaling AI across a large, cautious enterprise where a previous big-bang rollout failed. Which approach is BEST?
Show answer
Answer: B.
Scaling methodologies start with short-term wins that prove value and inform a repeatable pattern expanded in gated waves, which also avoids repeating the big-bang failure. B is correct. A repeats the failed approach. C never starts. D forfeits shared standards and reuse.
A firm's AI maturity is at early experimentation, but its three-year strategy depends on enterprise-scale AI. A director wants to jump straight to an enterprise-wide deployment to save time. What is the BEST guidance on progression?
Show answer
Answer: B.
Maturity advances in stages, and a progression pathway that closes readiness gaps is needed to move from experimentation toward enterprise scale; skipping stages usually fails. B is correct. A leaps past the capabilities scale requires. C discards a valid strategy. D halts progress rather than building toward it.
A team wants to lift a pilot build directly into enterprise production 'because it works', with no changes to governance, monitoring, security or operations. What is the MAIN risk, and what should happen first?
Show answer
Answer: B.
A pilot proves feasibility but rarely carries production-grade governance, monitoring, security and operations, so lifting it unchanged into enterprise use is risky and those foundations must come first. B is correct. A ignores the experimental-to-production gap. C and D describe non-issues rather than the real readiness risk.
A firm's biggest barrier to AI adoption is not technology but a risk-averse culture where staff fear that using AI or admitting a failed experiment will hurt their careers. Which leadership intervention MOST directly addresses this?
Show answer
Answer: B.
Risk aversion and fear of failure ease when leaders create safe experimentation, model the behaviour and celebrate learning, which directly targets the cultural barrier. B is correct. A deepens fear and produces gaming. C ignores the human barrier. D lets fear grow in the vacuum.
A change leader must overcome entrenched risk aversion and resistance during an AI rollout. Which TWO interventions are MOST appropriate?
Show answer
Answer: A and B.
Risk aversion and resistance ease when leaders create safe experimentation (A) and model the behaviour while celebrating wins and lessons (B). C deepens resistance and produces gaming. D lets fear grow. E ignores the cultural barrier that is the actual constraint.
A firm wants to build AI literacy and momentum across its workforce quickly and sustainably. Which TWO mechanisms MOST directly serve that goal?
Show answer
Answer: A and B.
Literacy and momentum grow through hands-on training and hackathons (A) and through POC programmes and responsible-AI training (B). C is a technical purchase, not a literacy mechanism. D removes the means to build literacy. E confines learning to one team instead of spreading it.
A customer-facing pilot is proposed for promotion to production. Which TWO conditions should hold BEFORE it is allowed to scale?
Show answer
Answer: A and B.
Before scaling a customer-facing pilot, production-grade governance and oversight (A) and measured value against a baseline (B) must be established. C is publicity. D is an unrelated purchase. E is a narrow sign-off that does not address readiness or proven value.
A firm is scaling an AI-driven process from one team to the whole enterprise. Which TWO practices MOST protect business continuity and performance during the scale-out?
Show answer
Answer: A and B.
Continuity requires a fallback path (A) and ongoing evaluation of cost, data readiness and performance across scaling (B). C removes the safety net prematurely. D wrongly assumes pilot results transfer unchanged at scale. E stops the monitoring that continuity depends on.
Adoption of a capable, well-supported internal assistant has stalled at 14% of licensed users after three months, and analysis shows the technology is not the constraint; the blockers are habit, fear of role impact and lack of visible leadership use. Which TWO interventions are MOST likely to move adoption?
Show answer
Answer: A and B.
When technology is not the constraint, manager modelling with clear expectations (A) and a champion network plus honest handling of role-impact fears (B) drive adoption. C repeats an information fix that is not the gap. D abandons the transformation. E deepens fear and produces gaming rather than genuine adoption.
Last updated Sep 18, 2026