AI Business Strategist
AIB-C01 Mock Exam 1
A full-length, 85-item, blueprint-weighted independent mock exam for the AWS Certified AI Business Strategist (AIB-C01) beta exam, with full explanations and an indicative scaled score.
This is a 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, so use this as your diagnostic: sit it first to find your two weakest domains before you book. All 85 items are new and do not repeat 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); you can also practise at the tighter figure or take the mock untimed the first time.
- 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 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not publish how a raw percentage maps to that scaled score — the number of scored versus unscored beta items is also unpublished.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.
| Raw score | Indicative scaled band | Reading |
|---|---|---|
| under 60% | well below the 700 line | Not ready; revisit your weakest domains and redo the in-page questions. |
| 60–74% | approaching the 700 line | Close but not yet a safe margin; target your two weakest domains. |
| 75–84% | around or above the 700 line | A reasonable margin on this mock; review any weak domain and sit Mock Exam 2 timed. |
| 85% and above | comfortably above the 700 line | Strong and consistent; 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. Work the case studies and the four domain pages first if this is your first pass, then sit Mock Exam 2 as your timed readiness gate.
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 retailer wants to automate the decision of whether a coupon is valid: it is valid only if it is before the expiry date and the order total exceeds a fixed threshold. The rule never changes. Which solution class is MOST appropriate?
Show answer
Answer: B.
The logic is fixed, deterministic and unambiguous, so rule-based automation is cheapest, fully auditable and cannot drift. B fits exactly. A applies a probabilistic generative model to a task with a single correct deterministic answer, adding cost and hallucination risk. C trains a statistical model for a rule that is already known with certainty, which is wasteful and less reliable. D adds agent autonomy and tool orchestration that the task does not require.
A business leader asks how AI, machine learning and generative AI relate to one another. Which statement is MOST accurate?
Show answer
Answer: C.
AI is the broad field, machine learning is a subset that learns patterns from data, and generative AI is a subset of ML that produces new content. C nests them correctly. A denies the nesting relationship. B inverts the hierarchy, making the narrowest category the broadest. D wrongly equates AI with ML and dismisses generative AI, which is a genuine technical category.
A fraud model was accurate when launched a year ago, but its accuracy has slipped steadily even though no code has changed. What is the MOST likely explanation a business owner should understand?
Show answer
Answer: B.
Models are trained on historical data; when the world moves away from that history, performance decays without any code change. This is model drift, and it is why AI needs ongoing monitoring. B is correct. A assumes malice with no evidence. C confuses a generative-AI context limit with a predictive-model decay problem. D contradicts the stated fact that the model launched accurate.
A team must build a chatbot that answers questions from a policy library that is updated almost weekly. Which model-adaptation technique fits BEST and why?
Show answer
Answer: B.
RAG grounds responses in an authoritative source that updates as the source updates, which suits frequently changing content. B is correct. A bakes policies into weights that go stale and are costly to refresh weekly. C is expensive, may exceed context limits, and still needs the current documents. D cannot answer accurately without access to the actual policy content.
A leader complains that a generative assistant gives vague, generic answers and wants to buy a more expensive model. What is the MOST cost-effective first step?
Show answer
Answer: B.
Vague output is most often a prompting problem; better instructions, context and examples are the cheapest lever and should be tried first. B is correct. A spends on a larger model before ruling out the free fix. C is a much larger investment for a problem that basic prompt engineering usually solves. D discards a tool before applying the obvious first remedy.
How does an AI agent MOST clearly differ from a plain generative AI assistant?
Show answer
Answer: B.
Agents are distinguished by autonomy, tool use, and orchestration of multi-step work toward a goal, whereas an assistant responds turn by turn. B is correct. A confuses model size with agentic capability. C is false; agents can and do err, which is exactly why oversight matters. D is dangerous and wrong: greater autonomy usually increases, not removes, the need for oversight.
An organisation discovers employees pasting confidential documents into free public chatbots whose terms allow training on inputs. Which response BEST mitigates this shadow-AI risk?
Show answer
Answer: B.
Shadow AI is best managed by transparent classification plus a safe sanctioned alternative, so the approved path is also the easy path. B is correct. A drives use onto personal devices where there is no control or logging. C accepts an ongoing confidentiality exposure. D creates an unworkable bottleneck that staff will bypass, recreating the shadow problem.
At a business level, what is ISO/IEC 42001?
Show answer
Answer: B.
ISO/IEC 42001 is the certifiable management-system standard for governing AI responsibly, and AWS has stated support for it. B is correct. A confuses a standard with a pricing structure. C mistakes a governance standard for an algorithm. D describes risk-tiered regulation, not a management-system standard, and no standard simply bans AI.
A leader is told a deployed model 'runs itself and needs no attention'. What should the leader insist on instead?
Show answer
Answer: B.
Models degrade as conditions shift, so ongoing monitoring and a remediation plan are essential. B is correct. A ignores drift and is the misconception the question tests. C does not address monitoring at all. D applies a fixed cadence disconnected from the actual performance signal that should trigger action.
Why does the distinction between structured and unstructured data matter for a leader planning an AI initiative?
Show answer
Answer: B.
Data type drives which techniques apply and how much preparation is needed, so it shapes solution selection and cost. B is correct. A denies a real and consequential distinction. C makes an unsupported value claim. D is false; structured data is heavily used in AI, especially predictive ML.
A team pastes an entire 300-page manual into every request and complains the model is expensive and 'misses details'. Which explanation fits BEST?
Show answer
Answer: B.
Stuffing everything into context inflates token cost and dilutes the model's focus, and can hit context limits. Retrieving only relevant passages fixes both. B is correct. A blames the tool for a usage pattern. C proposes a heavier change than the problem warrants. D contradicts the stated 300-page size.
Which statement about training a model on historical data is MOST accurate for a business leader?
Show answer
Answer: B.
A model reflects the patterns in its training data, biases included, and assumes future conditions resemble the past. B is correct. A is the opposite of the truth; historical bias propagates. C ignores that the learned patterns persist in the model. D ignores that data quality, not just quantity, drives outcomes.
A GenAI support model scores 88% overall but only 61% for customers writing in a less common language that was under-represented in training. What is the ROOT cause and the BEST first fix?
Show answer
Answer: B.
Uneven performance across segments traces to unrepresentative training data, and the fix is better representation plus segment-level monitoring. B is correct. A treats a data-representation problem as a capacity problem. C misattributes a data issue to context length. D hides a real service gap behind an aggregate average.
When is fine-tuning the MOST appropriate model-adaptation technique?
Show answer
Answer: B.
Fine-tuning suits stable, specialised behaviour, tone or format that you want the model to internalise. B is correct. A describes a case for RAG, since fine-tuned knowledge goes stale. C is wrong because fine-tuning is generally more costly than prompting or RAG. D is backwards; fine-tuning requires representative examples.
A recommendation engine's own suggestions increasingly narrow what customers see, and that narrowed behaviour becomes next month's training data. What phenomenon is this MOST likely to cause?
Show answer
Answer: B.
When a model's outputs shape the data it later learns from, a self-reinforcing feedback loop can amplify bias and narrow behaviour. B is correct. A is unrelated to the data dynamic described. C assumes a benefit the loop does not provide. D is the opposite; feedback loops increase the need for monitoring.
A lender must (a) apply a fixed statutory eligibility rule, (b) predict default risk from structured account history, and (c) draft a personalised decision letter. Which mapping of tasks to solution types is CORRECT?
Show answer
Answer: B.
A fixed rule suits deterministic automation, structured-history prediction suits an ML model, and drafting language suits generative AI. B matches each task to its right tool. A forces one tool onto three different problem types. C mismatches every task. D cannot predict risk or draft prose with rules alone.
A colleague claims that lowering a generative model's randomness setting will 'eliminate hallucinations'. What is the accurate business-level view?
Show answer
Answer: B.
Reducing randomness makes output more deterministic but does not make it factually correct; grounding and oversight remain necessary. B is correct. A overstates the effect and confuses consistency with truth. C denies a real effect on variability. D confuses an inference setting with a training technique.
Which TWO tasks are BETTER served by rule-based automation than by AI?
Show answer
Answer: A and C.
Deterministic, unambiguous, rule-expressible tasks belong to rule-based automation. A (a fixed rate calculation) and C (explicit keyword routing with fixed destinations) are both fully specifiable as rules. B needs language understanding of unstructured text. D needs generative language ability. E needs pattern learning to catch novel behaviour, which rules cannot anticipate.
A business owner will rely on a model's predictions. Which TWO questions about the underlying DATA are MOST important to ask?
Show answer
Answer: A and B.
Data representativeness and data quality/currency directly determine whether predictions are trustworthy. A and B are the load-bearing questions. C is cosmetic. D is irrelevant to data quality. E is a financial metric unrelated to whether the model's data is fit for purpose.
An autonomous agent will issue small customer refunds based on complaint emails. Which TWO controls are MOST important before go-live?
Show answer
Answer: A and B.
Autonomy over money demands bounded authority with human escalation and full auditability. A caps risk and routes hard cases to people; B makes actions reviewable and detectable. C and D are performance details, not risk controls. E has no bearing on control of an autonomous financial action.
A CEO announces 'we will adopt generative AI this year' but names no business problem. What is the BEST first move for a strategist?
Show answer
Answer: B.
Strategy starts from outcomes and use cases, not from a technology purchase. B defines the problem before spending. A buys a tool for an undefined need. C commits to a heavy build with no target. D trains people for no defined purpose. Each of A, C and D acts before the outcome is known.
A common, non-differentiating document-extraction capability is needed live in eight weeks. Building it in-house would take six months and three new hires; mature vendor options exist. What is the BEST sourcing decision?
Show answer
Answer: B.
For a common, non-differentiating capability under time pressure, buy-or-partner beats build on cost, speed and opportunity. B is correct. A spends six months and three hires to rebuild a commodity. C forfeits the eight-week window. D ignores a stated business need.
An initiative has spent 500,000 USD, missed its accuracy target twice, and has no realistic path to the additional data it needs this year. Applying scale-pause-terminate, what should a strategist recommend?
Show answer
Answer: B.
Feasibility has failed twice and the blocking dependency cannot be resolved this year, so terminate or pause; sunk cost is irrelevant to the forward decision. B is correct. A scales an unproven capability to chase sunk cost. C spends more into a known blocker. D lets optics override evidence.
Leadership demands the ROI of an assistant deployed three months ago, but nobody measured the before-state. What is the BEST response?
Show answer
Answer: B.
The honest, defensible path is to admit the gap, reconstruct the best proxy from available history, and fix the process going forward. B is correct. A fabricates data and destroys credibility. C substitutes the vendor's numbers for your own outcomes. D abandons measurement rather than salvaging what is possible.
A pilot cut average handle time from 10 to 8 minutes across 120,000 tickets per year, at a loaded agent cost of 30 USD per hour. The AI plus oversight costs 50,000 USD per year with a 40,000 USD one-off. What is the approximate Year-1 ROI?
Show answer
Answer: A.
Time saved is 2 minutes per ticket over 120,000 tickets = 240,000 minutes = 4,000 hours; at 30 USD that is 120,000 USD benefit. Year-1 cost = 50,000 + 40,000 = 90,000 USD. ROI = (120,000 − 90,000) / 90,000 ≈ 33%, so A is correct. B, C and D overstate the return because they ignore that the 90,000 USD cost consumes most of the 120,000 USD benefit in Year 1.
A workflow has volatile monthly volume that sometimes drops to almost zero. Which AWS AI pricing structure fits BEST at a strategic level?
Show answer
Answer: A.
Volatile, sometimes-near-zero volume favours consumption-based pricing, which you pay only when you use it. A is correct. B pays for capacity during the low months. C keeps compute running and burning money while idle. D charges per seat regardless of the workflow's variable volume.
A CMO wants an AI chatbot 'because a competitor launched one', with no target metric. What should the strategist require BEFORE funding it?
Show answer
Answer: B.
Funding should follow a defined outcome, a baseline and success metrics, not competitor mimicry. B is correct. A commits money before the case exists. C markets a project with no defined value. D competes on model size rather than on a business result.
70% of 1,000,000 monthly requests are simple. On-demand costs 0.010 USD each; a smaller model handles the simple ones at 0.003 USD. Routing the simple traffic to the smaller model reduces cost by approximately how much?
Show answer
Answer: B.
Baseline cost = 1,000,000 × 0.010 = 10,000 USD. With routing, 700,000 simple × 0.003 = 2,100 USD plus 300,000 complex × 0.010 = 3,000 USD, totalling 5,100 USD. Saving = 10,000 − 5,100 = 4,900 USD, about 49%. B is correct. A, C and D do not match the arithmetic; the saving is the 4,900 USD difference on the 10,000 USD base.
Which tool pairing does the exam reward for cost PLANNING versus cost TRACKING of AWS AI usage, at a strategic level?
Show answer
Answer: A.
The Pricing Calculator produces forward estimates; Cost Explorer shows actual spend, trends and anomalies. A pairs them correctly. B swaps their roles and adds Marketplace, which is for evaluating buy/partner options. C and D misassign one or both tools to functions they do not perform.
Over three years a 60,000 USD upfront initiative returns 110,000 USD net cash per year. At a 10% discount rate, what is the approximate 3-year NPV?
Show answer
Answer: A.
Discounted inflows: 110,000/1.1 ≈ 100,000; 110,000/1.21 ≈ 90,909; 110,000/1.331 ≈ 82,645; sum ≈ 273,554, minus the 60,000 outlay ≈ 213,554, about 213,500 USD. A is correct. B omits the upfront cost. C ignores discounting. D counts only one year of return.
A leader wants to commit to a discounted provisioned-throughput plan for a workload whose volume is unpredictable and often low. What is the MAIN risk?
Show answer
Answer: B.
A commitment is a fixed cost; if utilisation is low the committed capacity is wasted and the 'discount' costs more than consumption pricing would. B is correct. A ignores utilisation risk. C confuses a commercial commitment with model quality. D is the opposite of the stated low-volume pattern.
An industry is rapidly making an AI capability into table stakes, and a firm has not yet adopted it. What investment posture is MOST appropriate?
Show answer
Answer: B.
When a capability becomes table stakes, deliberate investment to reach parity is warranted, matched to feasibility. B is correct. A waits until the firm is behind. C dismisses competitive dynamics that define necessity. D over-invests without regard to feasibility, the opposite error.
A CFO is sceptical that '6,000 hours saved' is real money. What is the MOST honest way to present the value?
Show answer
Answer: B.
Hours saved become cash only when they avoid cost or are redeployed to value; the honest presentation states those assumptions. B is correct. A overstates by assuming all saved time is cashable. C withholds the translation the CFO legitimately needs. D inflates the figure with an inappropriate rate.
How does AI create SUSTAINABLE competitive advantage rather than a temporary one?
Show answer
Answer: B.
Durable advantage comes from pairing AI with hard-to-copy assets: proprietary data, distinctive workflows and organisational capability. B is correct. A buys a commodity anyone can buy, giving no lasting edge. C is publicity, not advantage. D is a technical choice competitors can match.
A team switching a live process to a new AI platform plans a single big-bang cutover on one date. What is the MOST important improvement for business continuity?
Show answer
Answer: B.
A phased transition with a fallback and validation protects continuity during a platform change. B is correct. A increases risk by compressing an untested cutover. C removes the safety net before the new platform is proven. D discards the validation that catches problems before they hit customers.
A claims-summarisation pilot cut reading time from 20 to 15 minutes across 200,000 claims per year at 40 USD per hour, costing 108,000 USD per year plus a 50,000 USD one-off. What is the approximate Year-1 ROI?
Show answer
Answer: A.
Time saved = 5 minutes × 200,000 = 1,000,000 minutes = 16,667 hours; at 40 USD ≈ 666,667 USD benefit. Year-1 cost = 108,000 + 50,000 = 158,000 USD. ROI = (666,667 − 158,000) / 158,000 ≈ 322%, so A is correct. B overstates the return, while C and D badly understate it by underestimating the hours saved across 200,000 claims.
A team reports only a year-end lagging metric for a pilot and is surprised it 'failed'. What was the MISTAKE?
Show answer
Answer: B.
Relying only on lagging metrics leaves no chance to steer; leading indicators exist to catch problems early. B is correct. A misattributes the failure to AI itself. C invents a vendor issue not stated. D assumes more spend would have helped a measurement failure.
An initiative is promising but blocked because legal has not yet cleared the required data access. Under scale-pause-terminate, which disposition is MOST appropriate?
Show answer
Answer: B.
A promising initiative blocked only by a resolvable dependency should be paused, not killed, and resumed once cleared. B is correct. A discards a viable idea over a temporary blocker. C and D both proceed without the legal clearance, exposing the organisation to compliance risk.
When choosing between Amazon Bedrock's managed foundation-model platform and building custom ML on Amazon SageMaker AI, which reasoning is MOST sound at a strategic level?
Show answer
Answer: B.
Managed platforms suit common needs quickly and cheaply; custom ML is justified only when a differentiated problem cannot be met otherwise. B is the balanced strategic rule. A defaults to the costliest path. C denies that custom ML is ever warranted. D decides on an out-of-scope UI preference.
A steady, predictable, high-volume workload runs continuously all year. Which pricing approach is MOST cost-effective at a strategic level?
Show answer
Answer: B.
Steady, predictable, high volume is exactly when a commitment-based discount pays off, because utilisation will be high. B is correct. A forgoes the discount available for predictable volume. C prices per user, unrelated to this compute-volume workload. D pays a premium with no benefit for steady traffic.
Which TWO are TANGIBLE benefits suitable for a hard ROI calculation, as opposed to intangible ones?
Show answer
Answer: A and B.
Tangible benefits are directly monetisable: A (labour cost reduction) and B (incremental revenue) both convert cleanly to cash. C, D and E are genuine but intangible benefits that are hard to quantify directly and belong in a separate, clearly labelled category, not the hard ROI figure.
Which TWO are LEADING indicators you would watch to steer an AI support pilot early, rather than lagging outcomes?
Show answer
Answer: A and B.
Leading indicators are early, actionable signals that predict success: A (adoption/volume) and B (weekly deflection/resolution) can be steered mid-pilot. C and D are lagging annual outcomes you cannot act on in time. E is the vendor's benchmark, not a signal from your own pilot.
A vendor proposal is the cheapest option but trains on your data and offers no exit clause. Which TWO considerations should MOST weigh on the decision?
Show answer
Answer: A and B.
The two material risks are data/IP exposure from training on your data and lock-in from no exit clause; both can outweigh the low price. A and B capture them. C, D and E are irrelevant to the strategic risks of the proposal.
When is AI NOT the appropriate solution? Select the TWO strongest cases.
Show answer
Answer: A and B.
AI is the wrong choice when a fixed rule already solves the task (A) or when the data it needs does not exist and cannot be obtained in time (B). C, D and E are precisely the situations where AI, machine learning or generative AI add value, so they are not cases against AI.
A loan-decisioning model denies an application and the applicant asks why, but the business cannot produce a reason. Which responsible AI dimension is MOST at risk?
Show answer
Answer: A.
Being unable to give a reason for a consequential decision is a failure of explainability, which regulated lending in particular demands. A is correct. B, C and D are performance or engineering attributes, not responsible AI dimensions, and none addresses the applicant's need to understand the decision.
A claims team wants to auto-approve 80% of claims, but testing shows the model's error rate is 4% for the majority group and 12% for a minority group. What is the BEST course of action?
Show answer
Answer: B.
A three-fold error gap between groups is a fairness issue that must be investigated and mitigated, with oversight, before deployment. B is correct. A hides the disparity behind an average. C conceals the very metric that reveals the harm. D over-reacts by abandoning a fixable initiative rather than remediating it.
A team plans to add bias testing and a human-review step 'as a compliance check the week before launch'. What is the problem with this timing?
Show answer
Answer: B.
Bolting governance on at the end is the anti-pattern; governance by design builds controls in from planning, avoiding costly late rework. B is correct. A ignores the retrofit cost and risk. C wrongly treats accuracy as a substitute for fairness testing. D dismisses a core safeguard.
Under the AWS shared responsibility model for AI workloads, which duty remains the CUSTOMER's even when using a fully managed AI service?
Show answer
Answer: B.
AWS secures the cloud; the customer remains accountable for use-case appropriateness, output fitness and oversight of their business process. B is the enduring customer duty. A, C and D are AWS's responsibilities for security of the cloud and never transfer to the customer for a managed service.
A generative customer-service assistant occasionally states company policies that do not actually exist. Which safeguard MOST directly addresses this?
Show answer
Answer: B.
Fabricated policies are hallucinations; contextual grounding checks verify output against a source and catch unsupported claims. B is correct. A may hold more text but does not verify factual grounding. C affects speed, not accuracy. D is a commercial model with no bearing on hallucination control.
A business classifies AI use cases into risk tiers and attaches a control set to each. A new customer-facing tool that makes automated decisions affecting individuals is proposed. Which treatment fits BEST?
Show answer
Answer: B.
Automated decisions affecting individuals are high impact and warrant the strongest control set. B applies proportionate controls. A under-controls a high-impact use case. C skips the classification the framework requires. D equates a consequential external decision with a trivial internal tool.
A team argues that because the AI vendor provides an intellectual-property indemnity, the company can publish generated marketing content without any review. What is the flaw?
Show answer
Answer: B.
An indemnity addresses some legal exposure but not factual accuracy, brand fit or reputational harm, so review remains necessary. 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 business process uses AI to make automated decisions affecting individuals in a jurisdiction with risk-tiered AI regulation. When should compliance be considered?
Show answer
Answer: B.
Compliance should be considered from planning onward, matched to the use case's risk tier. B is correct. A waits for enforcement. C addresses obligations too late, after individuals may be affected. D wrongly outsources accountability that remains with the deploying business.
An organisation has an 'AI ethics charter' document, but no one owns AI risk, there are no decision rights, and there is no path to approve an unusual case. How is this BEST characterised?
Show answer
Answer: B.
A charter with no owner, no decision rights and no operating path is governance in name only. B names it correctly. A mistakes a document for a working structure. C and D misclassify a governance-structure failure as a technical or commercial issue.
Where do the NIST AI Risk Management Framework functions fit when a leader is asked to 'apply a risk classification framework'?
Show answer
Answer: B.
The NIST AI RMF organises risk work into Govern, Map, Measure and Manage functions across the lifecycle. 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 pilot chatbot passed testing and was deployed with only infrastructure uptime monitoring. Three months later it is giving outdated answers. What governance gap does this reveal?
Show answer
Answer: B.
Uptime monitoring says nothing about answer quality; the gap is the absence of output-quality and drift monitoring in production. B is correct. A misreads uptime as a hardware sizing issue. C invents a technical cause. D over-reacts; the fix is monitoring, not abstention.
An organisation routes every AI request, however trivial, to a central review board that meets monthly, and teams have started using unapproved tools to avoid the wait. What is the BEST diagnosis and fix?
Show answer
Answer: B.
Governance that blocks everything drives shadow AI; a risk-tiered process that fast-tracks low-risk work while scrutinising high-risk cases is the fix. B is correct. A intensifies the bottleneck. C over-reacts and forfeits value. D removes the controls that manage genuine risk.
A leader must choose between a highly accurate black-box model and a slightly less accurate explainable model for consumer credit decisions. What is the BEST approach?
Show answer
Answer: B.
In regulated, individually consequential decisions, explainability is a requirement that a small accuracy edge does not outweigh. B is correct. A optimises accuracy while ignoring the regulatory need to explain. C decides on cost alone. D discards a legitimate, well-governed use.
How many core responsible AI dimensions does AWS publish, and how does the exam guide's own shorter list relate to them?
Show answer
Answer: B.
AWS publishes eight core responsible AI dimensions, and the exam guide's shorter list is a subset of that set. B is correct. A understates the count. C wrongly claims the guide's list is unrelated rather than a subset. D denies the published set exists.
Which statement BEST distinguishes controllability from veracity as responsible AI dimensions?
Show answer
Answer: B.
Controllability is about steering and overseeing behaviour, while veracity and robustness concern truthful, reliable output. B draws the distinction correctly. A conflates two distinct dimensions. C misattributes both to unrelated attributes. D swaps their meanings.
A model that was fair at launch begins producing more errors for one demographic group after six months as the customer base shifts. What does this illustrate, and how should it be managed?
Show answer
Answer: B.
Fairness at launch does not guarantee fairness later; as populations shift, bias can drift, so ongoing monitoring and remediation are needed. B is correct. A treats an ongoing lifecycle risk as a one-off. C misclassifies it as a cost issue. D ignores that the launch fairness was genuine and the drift is managed by monitoring.
A retail assistant both answers product questions and can issue refunds up to a value. How should governance treat these two capabilities?
Show answer
Answer: B.
Governance should be proportionate to each capability's risk: an informational answer differs sharply from an action that moves money. B applies controls where the risk is. A ignores the difference in consequence. C over-controls the harmless capability. D under-controls the financial action.
Which TWO conditions make human oversight of an AI decision MANDATORY rather than optional?
Show answer
Answer: A and B.
Oversight becomes 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 skip oversight. D is a cost fact. E is a popularity fact; neither bears on the need for human oversight.
A vendor pitches a generative tool for producing marketing copy. Which TWO questions MOST directly manage intellectual-property risk before signing?
Show answer
Answer: A and B.
IP risk turns on whether your data is used for training (A) and on the indemnity and provenance of the output (B). C, D and E are cosmetic or organisational facts that do not touch the intellectual-property exposure the leader must assess.
A content-moderation model over-blocks posts from one community and under-blocks harmful posts from another. 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, not a paper charter. Which TWO elements are MOST essential?
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.
An organisation has strong executive backing, good infrastructure and mature governance, but its customer data is siloed across three systems with no shared identifier. What should it prioritise BEFORE scaling AI?
Show answer
Answer: B.
Readiness is limited by the weakest dimension; here data is the constraint, so unifying the silos and fixing ownership comes first. B is correct. A cannot compensate for fragmented data. C scales on a broken foundation, multiplying the problem. D adds sponsorship that is already strong.
A company with two isolated pilots and no strategy describes itself as 'scaling AI'. How should a strategist respond?
Show answer
Answer: B.
Isolated pilots are experimentation, not enterprise scaling; a strategy and readiness foundations must precede scale. B is correct. A mislabels the maturity stage. C multiplies disconnected experiments without direction. D is plainly premature.
Which set of dimensions should an AI readiness assessment cover?
Show answer
Answer: B.
Readiness spans leadership, data, culture, infrastructure and governance; a gap in any one can stall a programme. B is complete. A, C and D each reduce readiness to a single factor and ignore the others that most often cause failure.
A firm at the piloting stage must choose between a 400,000 USD advanced MLOps platform and a 120,000 USD data-quality-and-literacy programme. Which is the better FIRST investment and why?
Show answer
Answer: B.
At the piloting stage the binding constraints are usually data quality and literacy; 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 pilot's programme stalls with no path to enterprise deployment. What is the MOST likely root cause?
Show answer
Answer: B.
Pilots that succeed but cannot scale usually lack the organisational foundations, ownership, data readiness, governance and operating model, that enterprise deployment needs. B is correct. A contradicts the stated success. C and D address budget and publicity, not the readiness gap that blocks scale.
A CEO wants to announce headcount reductions and then deploy an AI assistant to the affected team. What is the BEST leadership approach?
Show answer
Answer: B.
Pairing cuts with a rollout without honest communication poisons adoption; leaders should address fears and frame the role shift toward oversight and 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.
According to the AWS Cloud Adoption Framework, what is the correct order of the four transformation phases?
Show answer
Answer: B.
CAF's four phases run 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 planning time 30% in one depot; leadership wants it in all 40 depots next quarter, but the other depots keep data in incompatible spreadsheets. What is the BEST decision?
Show answer
Answer: B.
The pilot proved feasibility in one place; incompatible data across depots must be fixed before a phased, gated rollout. B is correct. A scales onto inconsistent data. C over-reacts to a solvable problem. D relies on unrepeatable manual heroics across 40 sites.
What is the primary purpose of an AI center of excellence (COE) when scaling AI?
Show answer
Answer: B.
A COE pools expertise, standards and reusable assets, lowering the cost and time of each new use case as the organisation scales. B is correct. A misframes a COE as a control that stifles teams. C confuses it with governance. D reduces it to a purchasing decision.
An organisation is transitioning contact-centre agents as AI takes routine queries. Which framing of the role change is BEST?
Show answer
Answer: B.
The intended transition shifts humans toward oversight and higher-value work that leverages judgement and empathy while AI handles routine volume. B is correct. A is both harmful and not the design intent. C denies the real change. D pits people against AI on the dimension AI is best at.
A cross-functional AI team is created, but 'everyone owns the outcome'. What is the problem and the fix?
Show answer
Answer: B.
When everyone owns an outcome, no one is accountable; the fix is clear owners and decision rights inside 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 enterprise. 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 waves with feedback. B is correct. A risks an enterprise-wide failure with no learning. C never starts. D forfeits shared standards and reuse.
An organisation's readiness scores are leadership 9, infrastructure 8, culture 7, governance 6 and data quality 3. What does this profile imply for its next action?
Show answer
Answer: B.
Readiness behaves like a weakest-link function; a data-quality score of 3 is the binding constraint and must be raised before scaling. B is correct. A hides the weak link behind an average. C invests where the organisation is already strong. D discards the assessment entirely.
What is the main risk of deploying a pilot build straight to enterprise production with no change?
Show answer
Answer: B.
A pilot proves feasibility but rarely has production-grade governance, monitoring, security and operations, so lifting it unchanged into enterprise use is risky. B is correct. A ignores the experimental-to-production gap. C and D describe non-issues rather than the real readiness risk.
Which AWS framework do leaders use to identify capability gaps across people, process, technology and governance when planning AI transformation?
Show answer
Answer: A.
AWS CAF and its perspectives are the tool for finding capability gaps across the organisation when planning transformation. A is correct. B is a cost-estimation tool. C is model documentation. D is a contractual availability commitment; none of these frames organisational capability gaps.
Which TWO are appropriate leadership interventions for cultural barriers such as risk aversion and fear of failure?
Show answer
Answer: A and B.
Risk aversion and fear of failure ease when leaders create safe experimentation (A) and model the behaviour while celebrating wins (B). C deepens resistance and produces gaming. D lets fear grow. E ignores the cultural barrier that is the actual constraint.
Which TWO mechanisms directly build AI literacy across a workforce?
Show answer
Answer: A and B.
Literacy grows 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.
Which TWO conditions should hold BEFORE a customer-facing AI pilot is allowed to scale to production?
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.
Which TWO statements about business continuity during an AI scale-out are correct?
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 assumes pilot results transfer unchanged. E stops the monitoring that continuity depends on.
Adoption of a well-built assistant has stalled at 11% of licensed users. Which TWO interventions are MOST likely to move it, given the technology is not the constraint?
Show answer
Answer: A and B.
When technology is not the constraint, manager modelling (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