Free AWS Certified AI Business Strategist (AIB-C01) Sample Questions

Correct answer: A — Tool use, because the assistant invokes external APIs to complete parts of the task

• A. Calling a calendar API and a payment API to take real actions in external systems is the definition of tool use, one of the core capabilities that separates an agent from a plain text-generating model.
• B. Fine-tuning refers to adapting a model's internal parameters with additional training data, not to calling external APIs during a live conversation.
• C. Prompt engineering is about how an instruction is worded to guide a model's response, not about invoking external systems to take action.
• D. Retrieval augmented generation refers to pulling supporting text from a knowledge base to ground a response, which is different from calling transactional APIs like calendars or payments.

Correct answer: A — Agent-to-agent communication, because two autonomous agents coordinate with each other to complete a shared goal

• A. Two separate autonomous agents exchanging dates and budget information to jointly complete a shared goal is exactly what agent-to-agent communication describes.
• B. This scenario involves two distinct agents exchanging information with each other, not one agent following a single, self-contained sequence of steps, so it is not simply an orchestration strategy.
• C. Calling a booking API is a separate capability from exchanging coordination information between two agents, and the scenario emphasizes the exchange between agents rather than an API call.
• D. Autonomy describes an agent's independent decision-making, but the scenario specifically highlights communication between two agents rather than one agent's internal ordering choice.

Correct answer: A — An AI solution is more appropriate, because the task involves interpreting varied, unstructured language rather than matching fixed patterns

• A. The keyword list's failure demonstrates that fixed patterns cannot capture the full variety of natural language, which is precisely the kind of unstructured-language task an AI solution is built to handle.
• B. Keyword lists cannot realistically be expanded to cover every possible phrasing of sentiment, which is exactly why the rule-based approach failed in the first place.
• C. Grouping and scoring customer reviews does not involve any external transactional system like a payment API, so an agent with tool use is not what this task calls for.
• D. Sentiment interpretation quality does depend heavily on the data a model has learned from, so this claim about training data being irrelevant is incorrect.

Correct answer: A — Use a predictive machine learning model for the default forecast and a generative AI model for drafting the letters

• A. This is correct because a predictive model fits a numeric risk forecast, while a generative model fits drafting new, personalized explanatory text.
• B. This is incorrect because it reverses the fit: forecasting a default score is a predictive task, and drafting new letters is a content-generation task.
• C. This is incorrect because relying on one generative model for a precise numeric risk score forgoes the reliability a purpose-built predictive model offers.
• D. This is incorrect because a predictive model is not designed to draft new explanatory prose, which is the strength of a generative approach.

Correct answer: A — Data such as emails, PDFs, images, or call recordings that does not fit a predefined schema

• A. This is correct because unstructured data covers formats like emails, PDFs, images, and call recordings that lack a predefined, tabular schema.
• B. This is incorrect because storage in relational tables with clearly defined columns and data types is the hallmark of structured, not unstructured, data.
• C. This is incorrect because being cleaned, labeled, and validated describes a data-quality state, not whether the data follows a fixed schema or not.
• D. This is incorrect because containing only numeric values with no free-form content describes structured data, not unstructured data.

Correct answers: A, B, C — Recorded customer support phone calls; Free-text customer product reviews; Scanned handwritten invoice documents

• A. This is correct because audio call recordings have no predefined tabular schema, making them a classic example of unstructured data.
• B. This is correct because free-text reviews are open-ended written content without fixed fields, which is a defining trait of unstructured data.
• C. This is correct because scanned PDF invoices are image-based documents without a consistent, machine-readable field layout, making them unstructured.
• D. This is incorrect because a sales table with defined columns for date, product, and revenue follows a predictable tabular format, making it structured data.
• E. This is incorrect because a database of customer IDs and balances is organized into clearly defined fields, which is the definition of structured data.
• F. This is incorrect because a spreadsheet of inventory counts follows a fixed row-and-column layout, which qualifies it as structured data.

Correct answer: A — Break the contract into smaller sections and process each section within the model's context window limit

• A. Chunking a long document into sections that each fit the model's token limit is the standard way to work within a context window constraint while still covering the full contract.
• B. Submitting content beyond the context window typically results in truncation or an error, so this does not reliably produce a usable review of the whole contract.
• C. This does not address the model's limitation at all and is not a workable business process for reviewing a legal document.
• D. The context window limit is measured in tokens derived from the text content, so changing the visual font size has no effect on how much text the model can process.

Correct answer: A — The report exceeded the context window, so the earliest content was truncated before use

• A. When document length exceeds the context window, content is truncated and the model can only work with what remains, which can drop earlier sections such as an opening caveat.
• B. Models do not intentionally suppress unfavorable information; this describes a motive rather than a token or context window limitation.
• C. Account permissions govern access to a system, not which parts of a supplied document a model can process within a single request.
• D. Summaries are generated from the supplied source text, so this contradicts how the task was performed in the first place.

Correct answers: A, B — Teaching the model to follow a specialized internal writing style across future outputs; Teaching the model to produce output in a company-specific format used in every document

• A. A specialized writing style that should apply consistently across all future outputs is a behavioral pattern well suited to fine-tuning, since it is not tied to a single changing fact.
• B. A consistent structured output format is a stable behavioral pattern that fine-tuning can bake into the model's default responses.
• C. Pricing that updates as soon as it is published is time-sensitive information best served by retrieval from a live source rather than a technique that adapts general behavior.
• D. A contract signed an hour ago is new, specific information that a retrieval-based approach can surface immediately, rather than a general behavior pattern to train into the model.
• E. A live count of open tickets changes continuously and needs to be pulled from a current source, which fits retrieval rather than a fixed behavioral pattern learned through fine-tuning.

Correct answer: A — Deploy a generative AI assistant that resolves routine password and billing tickets automatically, reducing average ticket handling time and freeing agents for complex cases

• A. This directly matches the high-volume, predictable ticket pattern to a measurable customer operations metric: average handling time.
• B. Documentation automation is useful but does not address the described high-volume repetitive ticket problem, so it is not the highest-impact match for this scenario.
• C. Marketing email generation belongs to a different business function and does not address the customer operations volume described in the scenario.
• D. Competitive research automation is a different function; it does not map to the ticket-handling problem stated in the scenario.

Correct answer: A — Buy a managed OCR and data-extraction service from a vendor, since the capability is standardized and the timeline does not allow for building it in-house

• A. Buying a managed OCR service fits a standardized capability and a short deadline, since a mature commercial product avoids the time cost of building the same capability internally.
• B. Building in-house does not guarantee the fastest delivery; without an internal ML team, in-house development would likely take longer than the three-month deadline allows.
• C. A fully custom co-developed model is a heavier undertaking than the timeline supports; partnering for a custom build is not the fastest path to a mature, standardized capability.
• D. Delaying the initiative to hire an ML team is unnecessary when a mature commercial product already meets the need within the required timeline.

Correct answers: A, B, C — The budget available for upfront development and ongoing operating costs; The timeline by which the capability needs to be in production; Whether vendor proposals meet the regulatory compliance requirements that apply to the retailer's data

• A. Available budget for upfront and ongoing costs is a standard factor in any build-buy-partner evaluation.
• B. The required time to production is one of the standard factors that determines whether building, buying, or partnering is realistic.
• C. Confirming vendor proposals meet applicable regulatory compliance requirements is a standard, necessary part of the evaluation for regulated data.
• D. The relative size of the vendor's office has no bearing on whether building, buying, or partnering is the right choice.
• E. Whether a product's marketing name includes the word 'AI' says nothing about its fit for the retailer's actual requirements.
• F. Whether a vendor's logo matches the retailer's brand colors is irrelevant to a build-buy-partner decision.

Correct answer: A — Comparing an organization's AI capabilities and adoption maturity against industry peers and competitors

• A. Comparing AI capabilities and adoption maturity against peers is correct because a competitive landscape assessment is fundamentally an external, relative comparison of where the organization stands versus its industry.
• B. Reviewing internal governance policies is incorrect because it looks inward at controls and compliance rather than outward at how competitors are adopting AI.
• C. Auditing infrastructure architecture is incorrect because it is a technical readiness exercise, not a comparison against the competitive market.
• D. Calculating ROI on one pilot is incorrect because it measures a single initiative's value rather than the organization's standing relative to competitors.

Correct answer: A — Neither retailer gains a durable advantage, because the identical, easily replicated tool is available to both

• A. This is correct because an identical, off-the-shelf tool available to any competitor is easily replicated and does not, by itself, create a defensible or lasting advantage.
• B. This is incorrect because a few days of earlier adoption of an identical, uncustomized tool is not enough to secure a permanent lead.
• C. This is incorrect because using the same shared, third-party tool does not generate proprietary data that is unique to either retailer.
• D. This is incorrect because a shared chatbot with no customization does not inherently produce a network effect tied to store count.

Correct answer: A — A measured, pilot-stage investment focused on learning and validating high-potential use cases

• A. This is correct because in a low-AI-maturity industry, a measured pilot-stage investment lets the manufacturer learn and validate use cases without overcommitting capital before value is proven.
• B. This is incorrect because an immediate enterprise-wide rollout skips the learning and validation that a low-maturity market calls for, raising unnecessary risk.
• C. This is incorrect because low competitor activity reflects industry timing, not evidence that AI would provide no value to this manufacturer.
• D. This is incorrect because matching an unrelated industry's budget ignores this manufacturer's own competitive dynamics and maturity level.

Correct answer: D — The reduction in average customer resolution time compared with the pre-AI baseline.

• A. A count of experimental models reflects activity level, not a business outcome tied to cost, revenue, or customer experience.
• B. Dataset size is a technical input measure and has no direct connection to a business outcome leadership should evaluate.
• C. Counting mentions of a word in press releases reflects messaging volume, not any real business result, making it a classic vanity metric.
• D. A measurable drop in resolution time compared with a recorded baseline ties directly to customer experience and operational cost, making it a genuine business KPI.

Correct answer: C — Developer output must have been measured before rollout as a baseline comparison.

• A. The pricing tier chosen for the assistant affects cost, not whether a productivity claim can be validated against prior performance.
• B. Productivity is itself a legitimate metric to report; avoiding it entirely does not make the claim more credible.
• C. Recording developer output before the rollout provides the baseline needed to credibly attribute any measured change to the AI assistant.
• D. Team size has no bearing on whether a productivity improvement claim is supported by measured evidence.

Correct answer: A — Maintain documented records of model decisions, data sources, and approvals across the lifecycle.

• A. Keeping documented records of decisions, data sources, and approvals gives auditors a traceable history to review, which is exactly what supports demonstrating compliance later.
• B. Restricting processing hours changes operations but produces no documentation an auditor could use to verify compliance.
• C. Broad source code access increases exposure risk and does not itself create the documentation an audit requires.
• D. Reducing reviewer staffing changes workload but does not create the records needed to demonstrate compliance during an audit.

Correct answer: A — Apply stricter oversight and review to the supplier certification system because its decisions carry higher consequences.

• A. The supplier certification system directly determines a consequential approve-or-deny outcome, so a risk classification framework correctly assigns it stricter oversight based on the impact of its decisions.
• B. Using the same team to build both systems says nothing about the impact of the decisions each system makes, which is the basis a risk classification framework should use.
• C. How often a system runs does not reflect the consequence of its decisions, so frequency of use is not the right basis for prioritizing oversight.
• D. Treating both as low-priority ignores that one system makes a consequential approval decision, which a risk classification framework would flag for more oversight.

Correct answer: A — Incorporate the legal team's concern into the project's risk review before the system launches.

• A. Incorporating the legal team's concern into the pre-launch risk review lets the governance structure evaluate and address the issue before it becomes a live regulatory problem, which is the purpose of cross-functional review.
• B. Waiting for regulators to raise the issue is reactive and abandons the proactive role governance review is meant to play.
• C. Removing legal from the review process eliminates exactly the cross-functional input that surfaced the concern in the first place.
• D. Deferring the concern to a future budget cycle delays addressing a live regulatory risk for reasons unrelated to the risk itself.

Correct answers: A, B, D — The decision is high-stakes and difficult to reverse, such as a loan denial or termination.; The model was only recently deployed and has not yet been validated against real-world outcomes.; A regulator or internal policy requires explicit human sign-off for this category of decision.

• A. High-stakes, hard-to-reverse decisions such as loan denials or terminations carry serious consequences if the AI system errs, which is exactly the situation where a human should confirm the outcome before it takes effect.
• B. A model that has not yet been validated against real-world outcomes is more likely to behave unpredictably, so human review provides a safety net until its real-world performance is established.
• C. Choosing a font size or color for an internal newsletter carries essentially no risk of harm, so mandatory human review would add cost without a corresponding benefit.
• D. When a regulator or internal policy explicitly requires human sign-off for a category of decision, that requirement itself establishes the need for human review regardless of the AI system's apparent accuracy.
• E. Reformatting a spreadsheet into a different file type is a low-risk, easily reversible task, so it does not present the kind of stakes that warrant mandatory human review.

Correct answer: A — To disclose a service's intended use cases, known limitations, and responsible AI design considerations.

• A. AI Service Cards are AWS's documentation resource that discloses a service's intended use cases, known limitations, and responsible AI design considerations, giving customers transparency into how a given AI service was built and where it should and should not be used.
• B. AI Service Cards are informational documentation, not a contract, and they do not guarantee any specific accuracy level for a service.
• C. AI Service Cards focus on responsible AI transparency rather than Region-specific deployment configuration steps.
• D. Pricing comparisons are covered by AWS pricing pages, not AI Service Cards, which focus on responsible AI disclosures instead.

Correct answer: A — Integrating responsible AI practices, such as risk identification, from a project's earliest stages.

• A. Governance by design means embedding responsible AI practices, such as identifying risks and defining required controls, from the earliest planning stages of a project, so those considerations shape the system as it is built rather than being retrofitted afterward.
• B. Requiring a single executive sign-off describes one possible accountability mechanism, but it does not by itself describe integrating responsible AI practices throughout a project's lifecycle.
• C. Conducting a one-time review only once the system is fully built is the opposite of governance by design, which calls for integrating these practices from the start of the project.
• D. Publishing a policy publicly is a transparency practice, but it does not describe when or how responsible AI practices are integrated into a project's planning and development.

Correct answer: A — Bias drift, where a model's fairness characteristics change over time as the population and inputs it sees evolve.

• A. Fairness outcomes can change after launch as the applicant population and input distributions shift, and this gradual change in disparate outcomes over time is exactly what bias drift describes.
• B. A one-time coding defect would typically appear immediately at launch rather than emerge gradually years later.
• C. Hosting cost changes affect the cloud bill, not the statistical fairness of the model's approval decisions.
• D. This scenario shows the opposite: fairness characteristics changed after launch, so treating them as fixed and unchangeable is incorrect.

Correct answer: A — Intellectual property risk, from generated content that may infringe on rights held by original content creators.

• A. Output that closely mirrors copyrighted source material raises the risk of infringing on the rights of the original creators, which is precisely what intellectual property risk describes.
• B. Response generation time is a performance concern and has no connection to whether output content infringes on someone else's copyright.
• C. Exchange rate movements affect infrastructure spending, not whether generated text resembles protected source material.
• D. Insufficient compute capacity affects throughput, not the likelihood that a model reproduces copyrighted phrasing.

Correct answer: A — Add a grounding verification step that checks each summary claim against the source paper before it is shared.

• A. Verifying each claim in a summary against its source paper directly targets the described problem of summaries containing findings that are not actually present in the source material.
• B. Increasing throughput addresses volume, not whether the content of each summary is factually accurate to its source.
• C. A layout redesign changes presentation, not whether the substance of the summary reflects the actual paper.
• D. Restricting user access controls who can use the tool but does not address whether the tool's summaries are factually accurate.

Correct answer: C — Data quality

• A. The scenario describes inconsistent and incomplete records, not employee attitudes or resistance to change, so cultural preparedness is not the dimension exposed.
• B. Incompatible formats and missing fields are data problems, not descriptions of compute capacity, network capability, or platform limitations.
• C. Inconsistent formats, missing values, and no clear data ownership are textbook data quality problems, since accurate and complete data is a prerequisite for a reliable forecasting model.
• D. The scenario says nothing about whether executives support the initiative, so leadership alignment is not the gap being illustrated.

Correct answer: A — Technical infrastructure

• A. A nightly batch-processing mainframe with no API layer cannot support real-time predictions, which is a direct limitation of the company's technical infrastructure.
• B. The scenario describes a systems limitation, not employee attitudes toward adopting AI, so cultural preparedness is not the obstacle described.
• C. There is no mention of executive endorsement or the lack of it, so leadership alignment is not what is being tested here.
• D. The problem described is about system architecture and processing speed, not about the accuracy or completeness of the shipment data itself.

Correct answer: A — Technology

• A. Disconnected databases, no shared data pipeline, and no platform capable of training or hosting models at scale are direct examples of a technology capability gap.
• B. The scenario explicitly states a documented AI project workflow already exists, so process is not the gap being illustrated.
• C. The scenario explicitly states the analytics team is experienced, so people is not the capability gap described here.
• D. The scenario does not mention approval authority or risk policy, so governance is not the gap being tested.

Correct answer: A — A senior executive publicly sponsors the initiative and empowers AI champions in each business unit

• A. A visible executive sponsor signals organizational priority and unblocks resources, while AI champions embedded in each unit keep momentum alive day-to-day instead of fading after a single announcement. This combination is the standard way leadership sustains an enterprise AI effort past the kickoff phase.
• B. A one-time memo with no follow-up does not sustain momentum because employees have no ongoing point of contact or visible leadership involvement. Cost figures alone do not address the lack of active sponsorship that caused the stall.
• C. Excluding business units removes the cross-functional input needed to keep the initiative relevant to daily work and does nothing to restore visible senior support. It also risks building a solution business teams will not adopt.
• D. Delaying communication with other teams leaves the broader workforce without any sponsorship signal, which is the exact problem causing the stall. Momentum requires ongoing visible support, not a wait-and-see approach.

Correct answer: A — Recognizing bias, verifying outputs, and knowing when to escalate AI-assisted decisions

• A. Responsible AI training focuses on judgment skills like spotting potential bias, checking outputs before acting on them, and knowing when a decision needs human escalation. These are the practical, non-technical skills a broad workforce needs to use AI tools safely.
• B. Fine-tuning production code is a technical AI skill aimed at practitioners building models, not the responsible-use judgment training intended for the general workforce. This falls outside what the general-audience training is meant to cover.
• C. Configuring infrastructure permissions is an operational and security task for technical teams, not a responsible-use topic for general workforce training. It requires specialized access most employees would never need.
• D. Compute instance selection is a cost and infrastructure decision made by technical or platform teams, unrelated to the bias-awareness and escalation judgment that responsible AI training targets. It does not build general workforce readiness.

Correct answers: A, B, C — Deciding who is authorized to access the dataset and for what purposes; Being accountable for the ongoing accuracy and quality of the dataset; Approving changes to how the dataset is classified and governed

• A. Deciding who may access a dataset and for what purpose is a core data ownership responsibility. This access decision authority is central to keeping data use accountable.
• B. A data owner is accountable for the dataset's ongoing accuracy and quality, since they are the responsible party if the data degrades or becomes unreliable. This accountability is a defining feature of ownership.
• C. Approving how a dataset is classified and governed falls under the owner's accountability for how the data is treated and protected. This ensures classification decisions have a clear responsible party.
• D. Writing training code and tuning hyperparameters is a data science or machine learning engineering task, not a data ownership responsibility. Ownership concerns the data itself, not the model-building work.
• E. Provisioning cloud servers is an infrastructure or operations task performed by engineering teams, not something a data owner personally does. It falls outside the scope of data ownership.
• F. Designing a marketing campaign is a business and communications function unrelated to being accountable for a dataset's quality and access. It is not part of the data owner role.

Correct answers: A, B — Alignment between how data is managed and the organization's AI and business priorities; Defined accountability for data quality, ownership, and access across the organization

• A. A data strategy is meant to connect how data is collected, governed, and used to what the organization is actually trying to accomplish with AI and the business. This alignment is one of its central purposes.
• B. Defining accountability for data quality, ownership, and access is a core outcome of a data strategy, since it clarifies who is responsible for the data that AI initiatives depend on. Without this, data foundations remain unclear and inconsistent.
• C. A data strategy cannot guarantee project budgets, since cost outcomes depend on scope, staffing, and execution, not data management practices. Budget assurance is not something a data strategy is designed to deliver.
• D. A blanket ninety-day deletion rule is an arbitrary retention policy, not a general outcome of a data strategy, and it could conflict with legitimate business or regulatory needs. Retention rules should be tailored to actual requirements rather than fixed universally.
• E. Committing to purchase new laptops annually is a procurement decision unrelated to how an organization manages and governs its data. It has no place as an outcome of a data strategy.

Correct answer: A — Establish ongoing model monitoring, defined incident-response procedures, and clear operational ownership before go-live.

• A. Ongoing monitoring, defined incident response, and clear operational ownership are exactly the governance and operational requirements that mark a production-grade transition.
• B. Reassigning the team without setting up monitoring or ownership leaves the production system without the operational support it needs.
• C. Expanding without reviewing monitoring or support processes skips the operational readiness that a production-grade transition requires.
• D. Indefinitely postponing deployment to evaluate an unrelated vendor does not address the operational and governance gaps the model actually needs closed.

Correct answer: A — The pilot that delivered a clearly measurable cost reduction and received strong support from its business unit.

• A. A clearly measurable cost reduction combined with strong business-unit support is exactly the kind of demonstrated short-term win that justifies further investment and expansion.
• B. Internal excitement without clear results does not provide the evidence needed to justify scaling investment.
• C. Build speed alone says nothing about whether the pilot delivered value worth scaling.
• D. Using new technology without an assessed business impact does not demonstrate the proof point that scaling decisions should be based on.