— Cisco AI Business Practitioner (AIBIZ)

Full question text, answer options, and explanations for this practice set — a spoiler-free alternative is the interactive quiz above for scored, shuffled practice.

1. . A company deploys a spam filter that learns from labeled email examples (spam vs. not spam) to classify new messages. Which category of machine learning does this represent?

   • A. Unsupervised learning
   • B. Reinforcement learning
   • C. Supervised learning(correct)
   • D. Transfer learning

   Explanation: Supervised learning trains a model on labeled examples — input data paired with correct output labels. A spam filter trained on emails already marked spam or not spam is a classic supervised classification task. Unsupervised learning finds patterns in unlabeled data, while reinforcement learning uses rewards and penalties to guide an agent.

2. . Which statement best distinguishes Artificial General Intelligence (AGI) from today's Narrow AI systems?

   • A. Narrow AI can only run on cloud infrastructure, whereas AGI runs on local hardware
   • B. AGI refers to AI that can perform any intellectual task a human can do, whereas Narrow AI is designed for a specific task or domain(correct)
   • C. AGI is currently deployed in most enterprise software products
   • D. Narrow AI requires more training data than AGI to achieve comparable accuracy

   Explanation: Narrow AI (also called Weak AI) is purpose-built for a single task — such as image recognition or language translation — and cannot generalize beyond it. AGI represents a hypothetical system with human-level flexibility to reason and learn across any domain. No AGI system exists today; all commercially deployed AI is Narrow AI.

3. . A retail company uses an AI system that groups customers into segments based on purchase behavior without any predefined labels. Which machine learning approach is being used?

   • A. Supervised learning — classification
   • B. Reinforcement learning
   • C. Unsupervised learning — clustering(correct)
   • D. Semi-supervised learning

   Explanation: Unsupervised learning discovers hidden patterns or groupings in data without labeled outcomes. Customer segmentation by purchase behavior is a clustering use case — the algorithm identifies natural groupings (e.g., high-value shoppers, deal seekers) without being told what the groups should be. This differs from supervised classification, which requires pre-labeled training examples.

4. . A manufacturing plant installs a computer vision system that inspects product images on an assembly line and flags defects with 98% accuracy. What is the primary business value of this application?

   • A. It replaces the need for all human workers in the plant
   • B. It enables automated quality inspection at scale, reducing defect escape rates and rework costs(correct)
   • C. It generates new product designs using generative AI
   • D. It translates operator manuals into multiple languages

   Explanation: Computer vision automates visual inspection tasks that were previously performed manually, enabling consistent quality checks at machine speed. The primary business value is catching defects before products reach customers, which reduces warranty costs, rework, and scrap. Quality inspection is one of the most widely cited industrial computer vision use cases.

5. . A business leader notices that the company's AI chatbot occasionally provides confident but factually incorrect answers. What AI limitation does this behavior illustrate?

   • A. Overfitting to training data
   • B. Hallucination — the tendency of language models to generate plausible-sounding but incorrect information(correct)
   • C. Underfitting due to insufficient model parameters
   • D. Reinforcement learning reward misalignment

   Explanation: Hallucination is a well-documented limitation of large language models (LLMs), where the model generates text that is fluent and confident-sounding but factually wrong or fabricated. This occurs because LLMs predict the next most likely token rather than retrieving verified facts. Business leaders should implement human-review workflows and source verification to mitigate this risk.

6. . A business analyst wants to understand the core difference between traditional rule-based automation and machine learning. Which statement most accurately describes this distinction?

   • A. Traditional automation uses statistical models; ML uses handwritten rules
   • B. ML systems learn patterns from data without being explicitly programmed with rules, whereas traditional automation executes predefined if-then logic(correct)
   • C. Traditional automation requires large labeled datasets; ML does not require any data
   • D. ML and traditional automation are functionally identical — the terms are interchangeable

   Explanation: Traditional rule-based automation relies on explicit, human-written logic (e.g., 'IF invoice amount > $10,000 THEN route for approval'). Machine learning instead infers rules from patterns in historical data, enabling it to handle ambiguous or variable inputs that would require thousands of manual rules to enumerate. This distinction is central to understanding when AI adds value beyond simple automation.

7. . A company is evaluating whether to build an AI model in-house, purchase an off-the-shelf AI solution, or partner with an AI vendor. Which factor most strongly favors the 'build' option?

   • A. The use case involves generic text classification with no proprietary data
   • B. The company has a small AI team and limited ML expertise
   • C. The AI use case requires training on highly proprietary, competitive data that cannot be shared with vendors(correct)
   • D. The company needs to deploy a solution within two weeks

   Explanation: Building an AI model in-house is justified when the competitive advantage comes from proprietary data that cannot be safely shared with third-party vendors, or when the use case is so unique that no suitable off-the-shelf solution exists. For generic use cases, buying or partnering is typically faster and less expensive. The build option requires significant ML expertise and long timelines.

8. . An organization wants to prioritize which AI initiatives to pursue first. It plots potential projects on a 2x2 matrix with business impact on one axis and implementation feasibility on the other. Which quadrant should receive the highest priority?

   • A. Low impact, high feasibility — easy wins regardless of value
   • B. High impact, low feasibility — biggest transformational potential
   • C. High impact, high feasibility — maximum value with achievable execution(correct)
   • D. Low impact, low feasibility — minimal investment required

   Explanation: The impact vs. feasibility matrix is a standard AI project prioritization tool. High-impact, high-feasibility projects deliver the most business value with the greatest chance of successful execution, making them ideal starting points. High-impact, low-feasibility projects may be aspirational long-term goals, while low-impact, high-feasibility projects may not justify the investment.

9. . An organization assesses its AI maturity and finds that AI projects exist in isolated pockets with no shared standards, governance, or infrastructure. According to AI maturity models, which stage does this represent?

   • A. Managed — AI is measured and controlled across the organization
   • B. Ad-hoc — AI initiatives are uncoordinated and lack formal processes(correct)
   • C. Defined — standard AI processes are documented and followed
   • D. Optimizing — continuous improvement is embedded in AI operations

   Explanation: In AI maturity models, the Ad-hoc (or Initial) stage is characterized by isolated, uncoordinated AI experiments with no governance, shared infrastructure, or repeatable processes. Organizations at this stage often have a few enthusiastic individuals driving AI projects without executive sponsorship or cross-functional alignment. Progressing to higher maturity stages requires standardizing processes and establishing centralized governance.

10. . Which type of business process is generally the BEST candidate for an AI automation use case?

    • A. A strategic planning meeting that requires executive judgment and qualitative debate
    • B. A high-volume, repetitive task with clear patterns in historical data (e.g., invoice processing)(correct)
    • C. A one-time creative brainstorming session for a new product concept
    • D. A legal contract negotiation requiring ethical judgment and relationship management

    Explanation: AI delivers the highest ROI on high-volume, repetitive tasks that have sufficient historical data to train models. Invoice processing is a canonical example: the volume is large, the inputs are structured, and the desired outputs are well-defined. Tasks requiring human judgment, empathy, or creativity are poor AI candidates in isolation, though AI may augment humans in those roles.

11. . A retailer implements an AI-driven personalization engine that increases average order value by 12% and reduces customer acquisition cost by 8%. How should the business case for continued AI investment be framed to leadership?

    • A. Frame it purely as a technology modernization initiative requiring capital expenditure
    • B. Quantify the revenue uplift and cost savings as ROI metrics, then project the value of expanding personalization to additional customer segments(correct)
    • C. Focus exclusively on the data science team's model accuracy scores as the primary KPI
    • D. Propose replacing the entire marketing team with AI to maximize cost reduction

    Explanation: Communicating AI ROI to leadership requires translating technical outcomes into business financial metrics — revenue uplift, cost reduction, and payback period. Framing the 12% AOV increase and 8% CAC reduction as dollar figures, then projecting the expansion opportunity, demonstrates clear business value and builds the case for continued investment. Model accuracy alone does not resonate with executive stakeholders.

12. . An AI hiring tool trained on historical hiring data consistently recommends fewer female candidates for engineering roles. What is the most likely root cause of this bias?

    • A. The model uses too many parameters and has overfit to noise
    • B. Historical bias in the training data — the model learned patterns from past decisions that reflected gender imbalance in hiring(correct)
    • C. The model was trained using reinforcement learning instead of supervised learning
    • D. The hiring tool lacks sufficient compute resources to process all candidates equally

    Explanation: Historical bias occurs when training data reflects past human decisions that were themselves biased. If historical hiring decisions underrepresented women in engineering, an AI trained on that data will perpetuate and potentially amplify the same bias. This is one of the most important AI ethics concerns — organizations must audit training data and model outputs for demographic disparities before deployment.

13. . A data scientist must choose between a highly accurate deep learning model (99% accuracy) and a logistic regression model (93% accuracy) for a loan approval system. A key stakeholder requires that every decision can be explained to a customer. Which model should be preferred, and why?

    • A. The deep learning model — higher accuracy always takes precedence in financial decisions
    • B. The logistic regression model — it is interpretable, allowing specific features (e.g., income, credit score) to be cited as decision factors(correct)
    • C. The deep learning model — regulators do not require explainability for automated lending decisions
    • D. Neither model — AI should not be used for loan approvals under any circumstances

    Explanation: The explainability vs. accuracy trade-off is a core Responsible AI consideration. Logistic regression produces interpretable coefficients that directly explain why a decision was made — critical for regulatory compliance (e.g., Fair Credit Reporting Act requirements) and customer communication. Black-box deep learning models may be more accurate but cannot easily articulate a specific reason for a denial, creating legal and ethical risks.

14. . Under GDPR, which principle is most relevant when an organization wants to use customer data collected for order fulfillment to train an AI marketing recommendation model?

    • A. Data minimization — collect as much data as possible to improve model accuracy
    • B. Purpose limitation — personal data collected for one purpose cannot be used for a materially different purpose without additional consent(correct)
    • C. Accuracy — the model must achieve 95% precision before data can be reused
    • D. Storage limitation — data can be used for any AI purpose within the original retention period

    Explanation: GDPR's purpose limitation principle requires that personal data be collected for specified, explicit, and legitimate purposes and not further processed in a manner incompatible with those purposes. Using order-fulfillment data to train a marketing AI model is a different purpose, requiring the organization to obtain explicit consent or establish another valid legal basis. This is a common compliance pitfall when building AI systems.

15. . Which combination of components is essential for a complete AI governance framework within an enterprise?

    • A. A GPU cluster, a model accuracy dashboard, and a data lake
    • B. Policy (defining acceptable AI use), oversight mechanisms (human review of high-risk decisions), and accountability structures (clear ownership of AI outcomes)(correct)
    • C. An open-source LLM, a prompt library, and an API gateway
    • D. Vendor contracts, SLA agreements, and disaster recovery plans

    Explanation: An AI governance framework must address three pillars: Policy (what is and is not permitted in AI development and deployment), Oversight (human-in-the-loop controls and auditing mechanisms for high-stakes decisions), and Accountability (named owners responsible for AI system outcomes, including bias monitoring and incident response). Technical infrastructure alone does not constitute governance.

16. . A company is deploying an AI system that makes autonomous decisions in a safety-critical environment. Which Responsible AI safety consideration should take highest priority?

    • A. Maximizing model inference speed to reduce latency
    • B. Ensuring containment mechanisms and human oversight are in place so that the AI cannot take irreversible harmful actions without human approval(correct)
    • C. Using the largest available model to minimize error rates
    • D. Encrypting all model weights to protect intellectual property

    Explanation: In safety-critical AI deployments (medical devices, autonomous systems, critical infrastructure), the highest priority is containment — designing the system so it cannot take catastrophic or irreversible actions autonomously. This includes human-in-the-loop oversight for high-consequence decisions, fail-safe defaults, and the ability to override or shut down the system. These principles align with Cisco's Responsible AI emphasis on safety and human oversight.

17. . An organization is launching an AI initiative and surveys employees about their concerns. The most common responses are 'I'm worried AI will replace my job' and 'I don't know how to use these tools.' Which AI adoption barriers do these responses reflect?

    • A. Data readiness and infrastructure limitations
    • B. Fear of job displacement and skill gaps(correct)
    • C. Budget constraints and vendor lock-in
    • D. Model accuracy and latency issues

    Explanation: Fear of job displacement (psychological resistance) and skill gaps (lack of competency to use AI tools) are the two most frequently cited human-side barriers to AI adoption in the AIBIZ curriculum. Addressing them requires transparent communication about AI's role as an augmentation tool, combined with structured upskilling programs to build employee confidence and capability.

18. . A Chief Digital Officer wants to structure an AI Center of Excellence (CoE). Which description best captures the primary purpose of an AI CoE?

    • A. A dedicated team that builds all AI models in-house and prohibits business units from using external AI tools
    • B. A centralized function that sets AI standards, shares best practices, provides governance, and accelerates AI capability building across business units(correct)
    • C. An executive committee that approves all technology spending above $1M
    • D. A help desk for employees encountering errors in AI-powered software

    Explanation: An AI Center of Excellence is a cross-functional team that establishes AI standards, governance policies, reusable frameworks, and shared infrastructure. Its goal is to prevent siloed AI efforts, ensure ethical and consistent practices, and accelerate adoption by providing templates, training, and guidance to business units. The CoE enables scale without requiring each team to reinvent foundational AI capabilities.

19. . When designing an AI pilot (proof of concept), which approach best sets up the initiative for a credible go/no-go decision?

    • A. Define success criteria and measurable KPIs before the pilot begins, and scope the pilot to a representative but limited dataset and use case(correct)
    • B. Deploy the AI system organization-wide immediately to gather the maximum amount of feedback
    • C. Allow the AI team to decide success criteria after reviewing pilot results to avoid setting unrealistic targets
    • D. Run the pilot indefinitely until the model achieves 100% accuracy

    Explanation: A well-designed AI pilot pre-defines success criteria (e.g., 'reduce processing time by 30% on a sample of 500 invoices within 8 weeks') before execution. Pre-defined KPIs prevent post-hoc rationalization of results and enable a genuine go/no-go decision. Limiting scope to a representative subset reduces risk while producing generalizable insights. This structured approach is a key change management best practice.

20. . A business leader needs to communicate the value of an AI customer service initiative to a non-technical executive team. Which communication approach is most effective?

    • A. Present the model's F1 score, precision, and recall curves in detail
    • B. Show the neural network architecture diagram and explain transformer attention mechanisms
    • C. Translate AI outcomes into business terms: reduced average handle time by 40%, deflected 25% of calls, projected annual savings of $2.4M(correct)
    • D. Share the Python training code to demonstrate technical rigor

    Explanation: Non-technical executives respond to business impact metrics — cost savings, time reductions, and revenue implications — not technical model performance metrics. Translating AI outcomes into financial and operational terms (call deflection, handle time, dollar savings) makes the value tangible and directly relevant to leadership priorities. This skill is a core competency in the AIBIZ change management curriculum.

21. . When applying Kotter's 8-step change model to an AI transformation, which step is most commonly skipped — and most likely to cause AI initiative failure?

    • A. Step 8: Anchoring changes in culture — organizations do not embed AI practices into performance management and onboarding(correct)
    • B. Step 1: Creating urgency — executives always feel urgency about technology trends
    • C. Step 3: Forming a strategic vision — AI projects always have a clear vision from day one
    • D. Step 6: Generating short-term wins — teams naturally celebrate milestones without structure

    Explanation: Kotter's final step — anchoring change in organizational culture — is the most frequently skipped in AI transformations. Organizations often celebrate the pilot launch but fail to embed AI usage into job descriptions, performance reviews, onboarding processes, and workflows. Without cultural institutionalization, AI adoption regresses when project champions leave or initial enthusiasm fades. Sustained transformation requires structural reinforcement.

22. . A user provides an AI assistant with a single instruction and no examples: 'Summarize this quarterly earnings report in 3 bullet points.' What prompting technique is this?

    • A. Few-shot prompting
    • B. Chain-of-thought prompting
    • C. Zero-shot prompting(correct)
    • D. Fine-tuning

    Explanation: Zero-shot prompting provides the AI with only an instruction and no examples of the desired output format. The model relies entirely on its pre-trained knowledge to complete the task. Few-shot prompting provides 2–5 examples within the prompt, while chain-of-thought prompting instructs the model to reason step-by-step. Zero-shot is the simplest and most common approach for straightforward business tasks.

23. . Cisco Webex AI features include automatic meeting summaries, live transcription, and action item extraction. What category of AI capability underlies these features?

    • A. Computer vision — analyzing video frames to detect speaker attention
    • B. Natural Language Processing (NLP) and speech recognition — converting audio to text and extracting semantic meaning(correct)
    • C. Reinforcement learning — optimizing meeting schedules based on participant feedback
    • D. Generative adversarial networks — synthesizing realistic meeting transcripts

    Explanation: Webex AI productivity features are built on NLP and automatic speech recognition (ASR). Speech recognition converts spoken audio to text (transcription), NLP extracts named entities and semantic content (action items, decisions), and summarization models condense the transcript into a meeting summary. These are core NLP use cases that the AIBIZ curriculum highlights as examples of AI-powered collaboration tools.

24. . A finance team is using an AI tool to automate transaction review and flag potentially fraudulent charges. Which AI application domain does this represent?

    • A. AI in HR — candidate screening and resume analysis
    • B. AI in marketing — customer personalization and recommendation engines
    • C. AI in finance — fraud detection using anomaly detection or classification models(correct)
    • D. AI in supply chain — demand forecasting and inventory optimization

    Explanation: Fraud detection is one of the most mature and widely deployed AI applications in finance. ML models (anomaly detection for unsupervised scenarios, or supervised classifiers trained on labeled fraud cases) analyze transaction patterns in real time to identify suspicious activity. This is a canonical example used in the AIBIZ curriculum to illustrate AI's value in financial operations.

25. . A business team receives an AI-generated market research report. Before acting on it, which quality evaluation steps are most critical?

    • A. Check the model's parameter count and training dataset size — larger models produce more reliable outputs
    • B. Verify key factual claims against authoritative sources, check for signs of hallucination, and assess whether the content reflects any demographic or selection bias(correct)
    • C. Confirm the report was generated with a temperature setting below 0.5 for determinism
    • D. Accept the output without modification since enterprise AI tools are always accurate

    Explanation: Evaluating AI tool output quality requires: (1) hallucination checking — verifying that specific facts, statistics, and citations exist in authoritative sources; (2) source verification — confirming the AI did not fabricate references; and (3) bias detection — identifying whether the analysis disproportionately favors certain perspectives due to training data patterns. These human-verification steps are essential before AI outputs inform business decisions.