I85PREP
⌘ K
AMPI PIC study materials
AMPI PIC study resources

AMPI PIC practice questions

One question at a time. Build your AMPI PIC understanding.

First 50 of 200 questions. Each one lists the choices, the correct answer, and an explanation. The checker at the top of this page saves progress in this browser when it is available.

Continue in the practice tool above to load more questions, 50 at a time.

Question 1Digital strategy and continuous updating (AI, data, cloud)

In the context of digital strategy, what does "cloud computing" primarily change about how an organization obtains IT resources?

  1. It converts capital expenditure on infrastructure into variable operational expenditure consumed on demand Correct answer.
  2. It guarantees that legacy applications run faster without modification
  3. It eliminates the need for any information security controls
  4. It removes the need for data governance policies

Explanation

Cloud computing shifts infrastructure from owned capital assets to metered, on-demand services, converting capex into opex and enabling elastic scaling. It does not remove security, governance, or integration responsibilities, which remain with the organization under shared-responsibility models.

Question 2Digital strategy and continuous updating (AI, data, cloud)

A product team wants to add a recommendation feature using a pretrained model accessed via API rather than training its own. Which approach does this describe?

  1. Running a fully offline rule engine
  2. Using a pre-trained model through an API Correct answer.
  3. Conducting a manual A/B test
  4. Building a foundation model from scratch

Explanation

Consuming a pretrained model through a hosted API lets teams add machine-learning capability without owning training infrastructure or datasets. This is a common low-barrier entry point for embedding AI into products.

Question 3Digital strategy and continuous updating (AI, data, cloud)

Which statement best describes the purpose of a data strategy in a digital transformation program?

  1. It defines how data is governed, accessed, and turned into business value Correct answer.
  2. It specifies the exact vendor contracts the organization must sign
  3. It documents the organization's organizational chart
  4. It replaces the need for a technology roadmap

Explanation

A data strategy sets direction for data quality, governance, access, and value creation, aligning data assets with business goals. It complements rather than replaces technology and business planning.

Question 4Digital strategy and continuous updating (AI, data, cloud)

What is the main benefit of continuous delivery in a digital product organization?

  1. It eliminates the need for product owners
  2. It allows small, frequent releases that reduce deployment risk Correct answer.
  3. It removes the need for automated testing
  4. It guarantees zero production incidents

Explanation

Continuous delivery keeps software in a releasable state and ships changes in small increments, which makes failures easier to isolate and roll back. It depends on, rather than replaces, automated testing and product direction.

Question 5Digital strategy and continuous updating (AI, data, cloud)

Which practice best supports keeping an AI system's behavior trustworthy after deployment?

  1. Removing all human review to increase speed
  2. Relying solely on the vendor's marketing claims
  3. Monitoring inputs, outputs, and drift with periodic review Correct answer.
  4. Freezing the model permanently at launch

Explanation

Models degrade as real-world data drifts, so continuous monitoring and periodic review are needed to detect performance loss and harmful outputs. Governance should persist across the operational life of the system.

Question 6Digital strategy and continuous updating (AI, data, cloud)

What does the term "multi-cloud" describe in an enterprise architecture?

  1. Outsourcing the entire IT department
  2. Using services from more than one cloud provider Correct answer.
  3. Replacing networks with satellite links
  4. Running all workloads in a single private data center

Explanation

Multi-cloud means deliberately using two or more public cloud providers, often to avoid lock-in or to match workloads to provider strengths. It increases architectural complexity and requires consistent governance across environments.

Question 7Digital strategy and continuous updating (AI, data, cloud)

Which metric most directly indicates whether a newly launched digital feature is delivering value to users?

  1. Count of meetings held by the delivery team
  2. Number of lines of code written
  3. Total size of the source repository
  4. Adoption and engagement rate among target users Correct answer.

Explanation

Value delivery is measured by whether target users adopt and engage with the feature, ideally linked to a business outcome. Output metrics such as code volume do not demonstrate user or business value.

Question 8Digital strategy and continuous updating (AI, data, cloud)

Why is data quality considered foundational to successful AI initiatives?

  1. Data quality only matters for reporting dashboards
  2. Models trained on poor data tend to produce unreliable or biased outputs Correct answer.
  3. Data quality is irrelevant once a model is in production
  4. High-quality data removes the need for model evaluation

Explanation

Model behavior reflects the data it learns from, so errors, gaps, and bias in training data propagate into predictions. Data quality work is therefore a precondition for trustworthy AI, not an optional reporting concern.

Question 9Digital strategy and continuous updating (AI, data, cloud)

An organization wants to experiment quickly with several AI use cases while limiting sunk cost. Which approach best fits this goal?

  1. Purchasing hardware for every possible future model
  2. Waiting until the technology is fully mature before starting
  3. Running time-boxed pilots on managed services with clear success criteria Correct answer.
  4. Commissioning a five-year custom platform before any pilot

Explanation

Time-boxed pilots on managed services let teams test value hypotheses with limited investment and exit quickly if results disappoint. This staged approach reduces risk while building organizational learning.

Question 10Digital strategy and continuous updating (AI, data, cloud)

A team is designing a cloud migration and wants to minimize rework for workloads that will eventually be refactored. What sequencing principle should guide them?

  1. Prioritize workloads by business value and technical feasibility, choosing a migration pattern per workload Correct answer.
  2. Defer all decisions until the data center lease expires
  3. Move the most complex workloads first regardless of value
  4. Migrate every workload identically with a single pattern

Explanation

Effective migration portfolios triage applications by value and feasibility, applying rehost, replatform, or refactor patterns as appropriate to each. A single blanket pattern ignores differences in business criticality and technical debt.

Question 11Digital strategy and continuous updating (AI, data, cloud)

When introducing generative AI into a customer-facing workflow, which control most directly addresses the risk of fabricated responses?

  1. Grounding responses in a curated knowledge source with citation and review Correct answer.
  2. Removing logging to protect privacy
  3. Shortening the user interface text
  4. Increasing the model's temperature setting

Explanation

Retrieval grounding constrains generation to vetted content and enables citation, reducing hallucination risk. Logging and review provide the audit trail needed to detect and correct failures.

Question 12Digital strategy and continuous updating (AI, data, cloud)

A leadership team asks how to govern AI use across business units without blocking innovation. Which model best balances these goals?

  1. A rule that only the IT department may use AI
  2. Unrestricted use with no documentation requirements
  3. A central policy framework with risk-tiered review and delegated execution Correct answer.
  4. A total ban on AI tools until further notice

Explanation

Risk-tiered governance sets guardrails and escalation criteria centrally while letting teams move quickly on low-risk use cases. This balances consistency with the speed innovation requires.

Question 13Digital strategy and continuous updating (AI, data, cloud)

Which indicator best signals that a cloud cost problem requires architectural attention rather than simple tagging fixes?

  1. Costs grow faster than usage and revenue over several periods Correct answer.
  2. The monthly invoice arrived a day late
  3. A single team forgot to tag one resource
  4. One developer attended a training course

Explanation

Sustained cost growth outpacing usage and revenue points to structural inefficiency such as over-provisioning or poor workload design. Tagging corrections address visibility, not underlying architecture.

Question 14Digital strategy and continuous updating (AI, data, cloud)

A company wants its data platform to support both regulatory reporting and machine-learning experimentation. Which design choice best serves both needs?

  1. Storing all data only in spreadsheets owned by analysts
  2. Prohibiting experimentation to protect reporting data
  3. A single ungoverned data lake with no access controls
  4. A governed data platform with curated, access-controlled datasets and separate exploratory sandboxes Correct answer.

Explanation

A governed platform with curated datasets supports reliable reporting, while isolated sandboxes allow experimentation without contaminating production data. Access controls satisfy regulatory obligations in both contexts.

Question 15Digital strategy and continuous updating (AI, data, cloud)

What is the primary reason to adopt infrastructure as code in a cloud operating model?

  1. It eliminates the need for cloud engineers
  2. It makes infrastructure changes reproducible, reviewable, and reversible Correct answer.
  3. It guarantees that no configuration errors will occur
  4. It removes the need for change management entirely

Explanation

Treating infrastructure as versioned code enables peer review, automated testing, and reliable rollback of environment changes. It strengthens change management rather than abolishing it.

Question 16Digital strategy and continuous updating (AI, data, cloud)

A team reports that an ML model performs well offline but poorly in production. Which cause is most plausible?

  1. The offline evaluation used too many metrics
  2. The model was trained on a GPU
  3. Training and production data distributions differ, or feature pipelines behave differently in production Correct answer.
  4. The team documented the model too thoroughly

Explanation

Training-serving skew, caused by distribution shift or inconsistent feature computation, is a classic reason offline metrics fail to hold in production. Consistent feature pipelines and drift monitoring are the standard mitigations.

Question 17Digital strategy and continuous updating (AI, data, cloud)

Which combination best describes a responsible approach to automating a decision that affects customers?

  1. Automation only when no data is available
  2. Full automation with no appeal mechanism
  3. Random sampling of decisions only after complaints
  4. Human oversight for high-impact cases, explainability, and an appeal path Correct answer.

Explanation

High-impact automated decisions require meaningful human review, explanations that affected parties can understand, and a route to contest outcomes. These controls address fairness, accountability, and regulatory expectations.

Question 18Digital strategy and continuous updating (AI, data, cloud)

An organization measures digital transformation progress only by the number of tools deployed. Why is this an inadequate indicator?

  1. Tool counts measure activity, not the capability or outcome change the transformation intends Correct answer.
  2. Tools cannot be counted reliably across departments
  3. Transformation never involves new tools
  4. Tool counts are always inaccurate

Explanation

Counting deployments captures activity rather than whether processes, skills, and customer outcomes have changed. Mature programs pair output metrics with outcome and adoption measures.

Question 19Digital strategy and continuous updating (AI, data, cloud)

A regulated firm wants to use a public cloud for sensitive workloads. Which arrangement best addresses data residency and control concerns?

  1. Avoiding written contracts to speed procurement
  2. Assuming all cloud regions offer identical legal protections
  3. Choosing a provider region and service tiers that meet residency requirements, with encryption and customer-managed keys Correct answer.
  4. Storing encryption keys only with the provider

Explanation

Residency obligations are met by selecting appropriate regions and service configurations, supported by encryption and key control that limit provider access. Contractual and technical controls must be aligned to the regulatory requirement.

Question 20Digital strategy and continuous updating (AI, data, cloud)

Which practice most improves the reliability of an AI feature whose outputs feed directly into a downstream automated process?

  1. Increasing output volume to overwhelm reviewers
  2. Relying on a single unchanging prompt forever
  3. Adding confidence thresholds and a fallback path for low-confidence outputs Correct answer.
  4. Removing validation to reduce latency

Explanation

Confidence thresholds let low-certainty outputs route to human review or a safe fallback instead of propagating errors downstream. This contains the blast radius of model mistakes in automated chains.

Question 21Digital strategy and continuous updating (AI, data, cloud)

An enterprise is evaluating whether to build or buy an analytics capability. Which factor most strongly favors buying a managed service?

  1. The organization has a large team of specialists eager to build
  2. The capability is a commodity need with mature market offerings Correct answer.
  3. Regulatory constraints forbid any third-party processing
  4. The capability is a source of unique competitive differentiation

Explanation

Commodity capabilities with mature vendors are usually cheaper and faster to buy, freeing internal effort for differentiating work. Build decisions are justified when the capability is strategic or no adequate vendor exists.

Question 22Digital strategy and continuous updating (AI, data, cloud)

A model's accuracy is stable, but subgroup error rates diverge significantly across customer segments. What does this most likely indicate?

  1. The issue can only be solved by removing the affected segments
  2. The model should be deployed immediately because overall accuracy is unchanged
  3. The divergence is irrelevant if the model is retrained monthly
  4. The aggregate metric masks performance disparities that require segment-level evaluation and mitigation Correct answer.

Explanation

Aggregate accuracy can conceal materially different error rates across groups, which raises fairness and safety concerns. Segment-level evaluation and targeted mitigation are needed before scaling deployment.

Question 23Digital strategy and continuous updating (AI, data, cloud)

A leadership team wants a single metric to track cloud modernization benefit. Which approach is most defensible?

  1. Report only the number of applications migrated
  2. Report a composite that combines cost, delivery speed, and reliability, with the underlying components visible Correct answer.
  3. Avoid measurement until the program is complete
  4. Report only the total cloud spend because it is easy to obtain

Explanation

Modernization benefits span cost, speed, and reliability, so a composite with transparent components prevents gaming and preserves diagnostic value. Single-dimension metrics invite optimization that harms other dimensions.

Question 24Digital strategy and continuous updating (AI, data, cloud)

An organization adopts retrieval-augmented generation for an internal knowledge assistant. Which failure mode is most specific to this architecture and needs explicit testing?

  1. The application uses HTTPS for transport
  2. The retriever returns outdated or unauthorized documents, causing the model to answer confidently from the wrong source Correct answer.
  3. The model's tokenizer splits words into subword units
  4. The assistant runs on a managed cloud service

Explanation

In retrieval-augmented systems, answer quality is bounded by retrieval quality and access control, so stale or unauthorized documents can produce confident but wrong or improper answers. Testing must cover retrieval relevance and permission filtering, not just generation.

Question 25Digital strategy and continuous updating (AI, data, cloud)

A firm finds that its AI governance board reviews every use case, creating a multi-month backlog. Which redesign best preserves oversight while restoring delivery speed?

  1. Eliminating the board and letting teams self-certify all cases
  2. Adding more board members without changing the process
  3. Requiring board review only after a production incident
  4. Tiered review where low-risk cases follow a documented self-assessment and only high-risk cases reach the board Correct answer.

Explanation

Risk-tiered review concentrates scarce expert attention on high-impact cases while standardizing low-risk approvals through self-assessment. This maintains accountability without making central review a bottleneck for all work.

Question 26Digital strategy and continuous updating (AI, data, cloud)

In a digital strategy context, what does "data gravity" primarily explain about cloud adoption?

  1. Large datasets attract the applications and services that need them, making migration progressively harder Correct answer.
  2. Data automatically replicates to every cloud region once it exceeds a threshold
  3. Storage costs always fall as dataset size grows
  4. Regulators require data to remain in its original format

Explanation

Data gravity describes how the cost and complexity of moving large datasets cause dependent applications and analytics to accumulate around wherever the data already resides. This creates lock-in effects that shape sequencing decisions in a cloud strategy. It is not an automatic replication or pricing rule.

Question 27Digital strategy and continuous updating (AI, data, cloud)

Which statement best captures the difference between a data lake and a data warehouse?

  1. A lake is always on-premises, while a warehouse is always cloud-hosted
  2. A lake enforces strict schema at ingestion, while a warehouse is schema-on-read
  3. A lake only supports unstructured data, while a warehouse only supports streaming data
  4. A lake stores raw data in native formats, while a warehouse stores modeled, schema-on-write data for structured query Correct answer.

Explanation

A data lake holds raw or lightly processed data in its native format, deferring structure until read time. A data warehouse stores curated, modeled data with schema applied on write, optimized for structured analytical queries. The two differ in modeling discipline, not in deployment location.

Question 28Digital strategy and continuous updating (AI, data, cloud)

What is the main purpose of a cloud landing zone in an enterprise digital strategy?

  1. To host the organization's public marketing website with maximum caching
  2. To provide a pre-configured, governed baseline of accounts, networking, identity, and controls for workloads Correct answer.
  3. To replace the need for any security operations team
  4. To benchmark cloud provider pricing across vendors

Explanation

A landing zone is a governed foundation that standardizes identity, networking, logging, and guardrails so teams can deploy workloads safely and repeatedly. It accelerates adoption by removing repeated setup work while embedding compliance controls. It does not eliminate security operations or serve as a price benchmark.

Question 29Digital strategy and continuous updating (AI, data, cloud)

Which description best matches supervised machine learning?

  1. A model discovers clusters in unlabeled data without any target variable
  2. A system follows hand-written business rules to produce outputs
  3. A model learns from labeled examples to predict an outcome for new inputs Correct answer.
  4. An agent learns by receiving rewards for actions in an environment

Explanation

Supervised learning trains on input-output pairs where the target label is known, so the model can generalize to unseen inputs. Clustering and reinforcement learning are distinct paradigms, and rule-based systems are not learning at all. The presence of labeled targets is the defining feature.

Question 30Digital strategy and continuous updating (AI, data, cloud)

What does "elasticity" mean when describing cloud infrastructure?

  1. Capacity can scale up or down automatically in response to demand Correct answer.
  2. Data is permanently replicated across three regions by default
  3. All workloads are migrated to serverless functions
  4. Compute pricing is fixed for the duration of a contract

Explanation

Elasticity is the ability to provision and release resources automatically as demand changes, matching cost to actual usage. It is a dynamic scaling property rather than a fixed pricing or replication guarantee. Serverless is one implementation option, not a definition.

Question 31Digital strategy and continuous updating (AI, data, cloud)

Which practice best supports continuous updating of a deployed machine-learning model as real-world behavior shifts?

  1. Freezing the model permanently after the first successful launch
  2. Retraining only when the annual budget cycle permits
  3. Removing all logging to reduce storage cost
  4. Monitoring input and output distributions and retraining on a defined trigger Correct answer.

Explanation

Model behavior degrades when live data drifts from training data, so production monitoring of distributions and performance, tied to retraining triggers, keeps the model current. Permanent freezing and budget-bound retraining ignore drift. Removing logging eliminates the evidence needed to detect it.

Question 32Digital strategy and continuous updating (AI, data, cloud)

What distinguishes generative AI from classical predictive AI?

  1. Generative models run only on-premises, while predictive models run only in the cloud
  2. Generative models never require training data, while predictive models always do
  3. Generative models produce new content such as text or images, while predictive models estimate a label or value Correct answer.
  4. Generative models are always more accurate than predictive models

Explanation

Generative AI learns patterns that let it synthesize new content, whereas classical predictive AI outputs a classification, score, or forecast. Both rely on training data, and deployment location or accuracy is not the dividing line. The nature of the output is the key distinction.

Question 33Digital strategy and continuous updating (AI, data, cloud)

Why does a cloud-first strategy typically favor managed platform services over self-managed infrastructure for commodity capabilities?

  1. It shifts undifferentiated operational work to the provider so teams can focus on differentiating capabilities Correct answer.
  2. It makes vendor lock-in impossible
  3. It removes the need for any architecture or security decisions
  4. It guarantees lower total cost for every workload regardless of scale

Explanation

Managed services handle patching, scaling, and availability for commodity functions, letting teams invest engineering effort where they create unique value. Cost outcomes vary by workload, and customers still own architecture and security choices. Managed services often increase, not remove, provider dependency.

Question 34Digital strategy and continuous updating (AI, data, cloud)

An insurer wants to add AI-assisted claims triage but has no labeled historical triage decisions. Which approach is the most practical first step?

  1. Start with a rules-based or retrieval-assisted assistant that surfaces evidence to human adjusters while labeled data is collected Correct answer.
  2. Abandon AI and rely solely on manual triage indefinitely
  3. Deploy a fully autonomous decision engine and audit it after launch
  4. Train a supervised classifier immediately on unlabeled claim documents

Explanation

Without labeled outcomes, supervised training has no reliable target, so an assistive approach that supports human judgment delivers value while generating the labels needed later. Autonomy without validation data is unsafe. Waiting indefinitely forfeits incremental benefit.

Question 35Digital strategy and continuous updating (AI, data, cloud)

A retail analytics team needs sub-second dashboard queries over curated sales aggregates while data scientists explore raw event logs. Which architecture fits best?

  1. A transactional database used simultaneously for checkout and analytics
  2. A single raw data lake queried directly by all business dashboards
  3. A warehouse or lakehouse serving modeled aggregates alongside a raw zone for exploratory access Correct answer.
  4. One spreadsheet export refreshed manually each night

Explanation

Separating curated, modeled aggregates for fast BI from raw storage for exploration serves both access patterns without contention. Querying raw data directly for dashboards is slow and error-prone, and mixing transactional and analytical load degrades both. Manual exports cannot meet sub-second needs.

Question 36Digital strategy and continuous updating (AI, data, cloud)

A team wants to reduce the time from code commit to production for a cloud-native application. Which combination most directly shortens that cycle?

  1. Manual release checklists executed by a single release manager
  2. Automated CI/CD pipelines with infrastructure as code and automated testing Correct answer.
  3. Quarterly change-approval boards with manual environment builds
  4. Separate hand-built environments per developer with no version control

Explanation

Automated pipelines, codified infrastructure, and automated tests remove manual handoffs and make deployments repeatable and fast. Manual approval boards, hand-built environments, and single-person release gates add delay and inconsistency. Speed comes from automation with embedded quality checks.

Question 37Digital strategy and continuous updating (AI, data, cloud)

What is the primary role of a feature store in an ML platform?

  1. To store the final trained model artifacts for archival
  2. To compute and serve consistent feature values for both training and online inference Correct answer.
  3. To provide a dashboard for executives to view model accuracy
  4. To replace the need for a data warehouse entirely

Explanation

A feature store centralizes feature definitions and computes them consistently for offline training and online serving, reducing training-serving skew. Model artifact storage and executive dashboards are separate concerns. It complements rather than replaces the warehouse.

Question 38Digital strategy and continuous updating (AI, data, cloud)

An organization wants to cut cloud spend without harming reliability. Which action addresses the largest avoidable cost first?

  1. Adding more monitoring dashboards without changing resource allocation
  2. Purchasing long-term commitments for every instance regardless of utilization
  3. Right-sizing over-provisioned instances and shutting down idle non-production resources Correct answer.
  4. Migrating all workloads to the most expensive region for latency

Explanation

Over-provisioning and idle non-production capacity are common sources of avoidable spend, and correcting them reduces cost without touching production reliability. Commitments purchased blindly can lock in waste. Dashboards alone do not change consumption.

Question 39Digital strategy and continuous updating (AI, data, cloud)

Which control most directly limits the exposure of sensitive data when employees use a third-party generative AI assistant?

  1. Enabling the assistant for all staff without logging
  2. Increasing the token limit of the assistant
  3. A policy document asking employees to be careful
  4. A data loss prevention layer that detects and blocks sensitive content in prompts Correct answer.

Explanation

Technical enforcement through DLP inspection of prompts prevents sensitive data from leaving the boundary, whereas policy alone relies on compliance. Raising limits or enabling broad access without logging increases exposure. Effective controls are enforced, not merely documented.

Question 40Digital strategy and continuous updating (AI, data, cloud)

A firm wants to compare two cloud providers for a latency-sensitive trading workload. Which evaluation approach is most defensible?

  1. Compare published marketing latency figures from each provider
  2. Choose the provider with the largest global region count
  3. Select the provider offering the deepest discount
  4. Run a representative benchmark from the actual user locations and measure tail latency Correct answer.

Explanation

Latency depends on network path, workload profile, and user location, so a representative benchmark measuring tail latency reflects real experience. Marketing figures, region counts, and discounts do not predict latency for this workload. Empirical measurement is the defensible basis.

Question 41Digital strategy and continuous updating (AI, data, cloud)

What does "training-serving skew" refer to in a production ML system?

  1. A difference between the cloud region used for training and the one used for serving
  2. A discrepancy between the model's accuracy and its business value
  3. A mismatch between how features are computed during training and during live inference Correct answer.
  4. The lag between model training completion and deployment approval

Explanation

Training-serving skew occurs when feature transformations differ between the offline training pipeline and the online serving path, degrading live predictions. Region placement and deployment lag are operational factors but not this concept. Business value mismatch is a separate concern.

Question 42Digital strategy and continuous updating (AI, data, cloud)

An organization wants to modernize a legacy application for the cloud while reducing risk. Which sequencing is generally most prudent?

  1. Refactor the entire application in one release before any migration
  2. Lift the database only and leave the application on aging hardware
  3. Rehost first to capture quick migration benefits, then refactor high-value components incrementally Correct answer.
  4. Rewrite from scratch without preserving existing business rules

Explanation

Rehosting establishes a cloud foothold and immediate operational benefits, after which high-value components can be refactored with reduced risk. A single large refactor or full rewrite concentrates risk and delays value. Migrating only the database leaves the application constrained.

Question 43Digital strategy and continuous updating (AI, data, cloud)

Which data governance capability most directly enables trustworthy AI across business units?

  1. A policy that forbids any data sharing between units
  2. A catalog with lineage, ownership, and quality metadata for each dataset Correct answer.
  3. A shared drive where teams store copies of datasets
  4. Annual manual audits of a sample of spreadsheets

Explanation

Cataloging with lineage, ownership, and quality metadata lets teams find, assess, and trust data before using it in models. Shared drives without metadata create duplication and ambiguity, blanket prohibitions block value, and sparse manual audits do not scale. Governance depends on visibility into data.

Question 44Digital strategy and continuous updating (AI, data, cloud)

A team is choosing between serverless functions and container orchestration for an event-driven workload with highly variable traffic. Which consideration most favors serverless?

  1. Guaranteed access to host-level kernel tuning
  2. Predictable per-hour pricing independent of request volume
  3. Simplified management of long-running stateful processes
  4. Scale-to-zero behavior that avoids paying for idle capacity during quiet periods Correct answer.

Explanation

Serverless functions scale to zero and bill per invocation, which suits spiky event-driven traffic where idle capacity would be wasted. Host-level tuning and long-running stateful processes are better served by container orchestration. Per-hour pricing independent of volume describes provisioned capacity, not serverless.

Question 45Digital strategy and continuous updating (AI, data, cloud)

What is the purpose of a model registry in an MLOps workflow?

  1. To version, store, and track the lineage and approval status of trained models Correct answer.
  2. To store raw training data in its original format
  3. To manage the cloud billing for GPU instances
  4. To execute the model's predictions for end users

Explanation

A model registry records model versions with metadata such as lineage, metrics, and approval state, enabling controlled promotion and rollback. Raw data storage, live inference, and cost management are handled by other components. The registry is the system of record for models.

Question 46Digital strategy and continuous updating (AI, data, cloud)

A leadership team wants AI initiatives to scale beyond isolated pilots. Which organizational enabler matters most?

  1. A shared platform and reusable components that let teams deploy models with common governance Correct answer.
  2. Requiring every team to build its own data pipeline and tooling
  3. Centralizing all AI work in a single team with no business-unit involvement
  4. Funding each pilot as a fully independent one-off build

Explanation

Reusable platforms with common governance let teams move from pilot to production faster and consistently. Independent one-off builds and duplicated pipelines multiply cost and risk, while full centralization without business involvement slows relevance. Shared enablement is the scaling lever.

Question 47Digital strategy and continuous updating (AI, data, cloud)

An enterprise wants to adopt a multi-cloud approach but is concerned about complexity. Which governance mechanism best contains that complexity?

  1. Using a different toolchain for every provider with no shared policy
  2. Allowing each business unit to define its own security and identity standards
  3. A common control plane with standardized identity, policy, and observability across providers Correct answer.
  4. Prohibiting any cross-provider networking

Explanation

Standardizing identity, policy, and observability across providers gives consistent control while permitting provider choice, which is how multi-cloud complexity is contained. Fragmented per-unit standards and disjoint toolchains multiply risk. Blanket networking bans do not address governance.

Question 48Digital strategy and continuous updating (AI, data, cloud)

A model's outputs feed an automated pricing system. Which practice most reduces the risk of harmful automated decisions when the model is uncertain?

  1. Increasing the prediction volume to average out errors
  2. Removing confidence scores and applying every prediction directly
  3. Retraining only after a customer complains
  4. Routing low-confidence predictions to human review with calibrated confidence thresholds Correct answer.

Explanation

Calibrated confidence thresholds let the system escalate uncertain cases to humans, containing harm where the model is least reliable. Discarding confidence and acting on all outputs removes the safeguard. Complaint-driven retraining is reactive, and higher volume does not eliminate systematic errors.

Question 49Digital strategy and continuous updating (AI, data, cloud)

An organization finds that its cloud bill grows faster than usage. Which analysis most directly distinguishes waste from legitimate growth?

  1. Compare the total bill month over month without further breakdown
  2. Reduce all services by a flat percentage across the board
  3. Assume growth is legitimate because usage also rose
  4. Decompose spend by service, tag, and unit economics such as cost per transaction Correct answer.

Explanation

Breaking spend into service, tag, and unit-cost components reveals whether cost growth tracks value delivery or reflects idle and over-provisioned resources. Comparing totals alone cannot separate the causes. Flat cuts risk harming productive workloads.

Question 50Digital strategy and continuous updating (AI, data, cloud)

A firm wants to use a large language model for regulated customer communications while limiting hallucination risk. Which design choice addresses the root cause most directly?

  1. Remove human review to speed up responses
  2. Ground responses in vetted internal sources with retrieval and require citations to approved content Correct answer.
  3. Raise the model's temperature to make answers more varied
  4. Rely on the model's training knowledge alone for regulatory facts

Explanation

Retrieval grounding ties responses to vetted, citable sources, reducing fabrication of regulatory facts. Higher temperature increases variability and risk, and relying on training knowledge alone invites outdated or invented content. Removing review eliminates a key safeguard.