50 IBM A1000-068 Practice Questions: Question Bank 2025

A product manager says, “We need AI that can summarize customer emails and draft responses.” Which AI capability best matches this requirement?

A team is building a model to predict whether a transaction is fraudulent (yes/no). Which type of machine learning problem is this?

A developer wants to quickly build an AI assistant that can ground responses in a company’s documents without training a new foundation model. Which approach is most appropriate?

A data science team has an imbalanced dataset: 1% fraud and 99% legitimate. Accuracy is 99% but fraud detection is poor. Which metric is most appropriate to evaluate this model?

A bank wants to deploy an AI model and must be able to explain why an application was approved or denied. Which best practice supports this requirement?

A retailer wants to reduce call center volume by automating answers to common questions (order status, returns) while handing off complex cases to human agents. Which solution pattern best fits?

An organization wants to operationalize AI and ensure models can be versioned, promoted from development to production, and monitored over time. Which practice most directly addresses this?

A team built a text-generation assistant for internal HR policies. In testing, it sometimes fabricates policy details not found in the official documents. What is the most effective mitigation?

A company wants to use customer chat transcripts to train a model. The transcripts contain names, phone numbers, and account IDs. What should be done FIRST to follow responsible AI practices?

A team deployed a loan-approval model. After several months, approval rates for one demographic group drop significantly even though overall accuracy looks stable. What is the best next step?

A project team is deciding whether to use supervised or unsupervised learning. They have historical data where each customer record is labeled as "churn" or "not churn." What approach is most appropriate?

A business stakeholder asks why an AI model that performs well in testing is failing after deployment, even though the code has not changed. Which concept best explains this issue?

An organization wants to enable employees to ask questions over internal policy documents and get grounded answers with citations. Which solution approach best fits this requirement?

A team using IBM watsonx wants to reduce hallucinations in a generative AI assistant for customer support. Which practice is most effective?

A retail company needs to build a model to predict demand and wants an end-to-end way to manage data preparation, model training, evaluation, and deployment within a governed platform. Which IBM capability best aligns?

A bank must comply with internal policies requiring that model decisions are explainable to auditors and that model risk controls are documented. Which combination best supports this requirement?

A data scientist evaluates a binary classifier and sees high accuracy, but the business reports many missed positive cases (false negatives). The positive class is rare. Which metric is most appropriate to focus on improving?

A team builds a text classification model and gets strong results in training but much weaker performance on new, unseen data. Which action is the best first step to address this?

A healthcare organization wants to use an LLM to summarize clinician notes. The notes may include protected health information (PHI). Which is the BEST governance-oriented approach to reduce risk while still enabling value?

A company is building an AI solution that combines: (1) a vector store for document retrieval, (2) prompt templates, (3) an LLM for generation, and (4) monitoring for quality and safety. Which architecture pattern does this describe?

A product team is building an internal chatbot to answer employee questions about HR policies. They want responses grounded in approved policy documents and want to reduce hallucinations. Which approach is most appropriate?

Which statement best describes the difference between supervised and unsupervised learning?

A business analyst wants to quickly explore a dataset and build a basic model without writing code, then share results with stakeholders. Which IBM capability best aligns with this need?

A data science team reports that a classifier performs very well in training but poorly on new, unseen data. Which issue is most likely?

A company wants to deploy an AI model and must be able to explain key factors influencing predictions to meet internal governance requirements. Which IBM capability best addresses this need?

A customer support solution must categorize incoming emails into topics and route them to the correct queue. The organization has thousands of historical emails already labeled by topic. Which approach is most appropriate?

A team wants to reuse a curated set of enterprise documents for multiple AI projects (search, Q&A, summarization). They need consistent indexing and governed access controls. Which design is best?

A bank deploys a loan approval model. After launch, approval rates for one protected group decrease significantly compared to historical patterns. What is the BEST immediate action aligned with responsible AI practices?

A team fine-tunes a generative model with internal chat transcripts. Later, the model occasionally outputs sensitive customer details. Which control would have MOST directly reduced this risk during solution design?

An organization wants an end-to-end approach to manage foundation model lifecycle: prompt assets, model evaluation, risk controls, and deployment oversight across teams. Which combination best fits this objective?

A product team wants a chatbot for internal HR questions. They have a small set of approved policy documents and want answers grounded only in those documents to reduce hallucinations. What is the recommended approach?

A team notices their classification model performs well in the lab but poorly in production because the live data has different feature distributions than the training data. Which issue is MOST likely occurring?

A retail bank wants to deploy an AI model but must ensure decisions can be explained to customers and auditors. Which model choice BEST supports this requirement at a foundational level?

A company is creating an AI assistant and wants to reduce the risk of exposing sensitive customer data in prompts and logs. Which practice is MOST appropriate?

A team is selecting an evaluation metric for a fraud detection model where fraud is rare. Which metric is generally MORE informative than accuracy for this scenario?

A solution architect wants to operationalize an AI model with repeatable deployments, model versioning, and automated testing. Which approach BEST aligns with industry best practices?

An insurance company wants to triage incoming claims emails by extracting key fields (policy number, date of loss, claimant name) and routing them to the right queue. Which AI capability is MOST directly applicable?

A team wants to use IBM watsonx for building a generative AI application and needs to manage prompts, evaluate outputs, and iterate quickly with governance controls. Which watsonx component is MOST aligned to this work?

A company is required to demonstrate that its AI system can be audited end-to-end (training data sources, feature engineering steps, model versions, approvals, and deployment history). Which set of artifacts is MOST critical to maintain?

A generative AI system is used to draft customer support responses. The business wants to reduce harmful or policy-violating outputs while keeping responses helpful. Which combined strategy is MOST effective?

A business stakeholder says, "We want an AI model that can explain every individual decision in plain language." Which model approach best aligns with this requirement at a foundational level?

A team wants to improve a customer support assistant so it gives answers grounded in the company’s policy documents, and reduces hallucinations. Which approach is most appropriate?

A project manager asks what “inference” means in an AI solution lifecycle. Which description is correct?

A data science team needs a governed environment to build models, track experiments, and manage deployments while integrating with other IBM data and AI capabilities. Which IBM platform best fits this need?

A retail company built a demand-forecasting model that performed well during training. After rollout, forecasting accuracy steadily degrades as customer behavior changes. What is the most likely issue and best next step?

A bank is evaluating an AI model for loan approvals. Which evaluation approach best supports both performance and fairness considerations?

A team is building an internal chatbot that can access HR policies and also take actions like creating IT tickets. What is the best practice to reduce risk while still enabling automation?

A product owner asks whether a proposed AI solution is supervised or unsupervised. The team has historical examples where each support ticket has a labeled category (e.g., billing, technical, account). What type of learning is this?

An organization wants to deploy an LLM-based assistant. Legal requires that no sensitive customer data is used to train the base model, and security requires that prompts and outputs are retained only as long as necessary for audit. Which governance controls best address these requirements?

A team implements RAG for a technical support assistant. Users report confident answers that cite irrelevant documents. The retrieval step returns many documents, but ranking seems poor. What is the most effective next improvement?