AWS AI Business Strategist (AIB-C01): 5 Practice Questions on AI Fundamentals and Literacy
AI Fundamentals and Literacy is a recurring theme in the AWS AI Business Strategist (AIB-C01) exam. It carries about 24% of the exam score. The 5 questions below come with a full explanation of why each option is right or wrong.
AWS AI Business Strategist (AIB-C01) · AI Fundamentals and Literacy · 5 questions with answers and explanations
🧠Before you start
Expect business situations rather than technical ones: a forecast that arrives too late, a team with little time, a calculation that must be exact. Each one asks which kind of AI, if any, suits the need. The right option matches the approach to the problem and the data available, while distractors offer a more ambitious tool than the situation calls for.
Keep in mind
• Ask first whether the rule can be written down. If the result must be exact and repeatable, ordinary software can be a better choice than a learned model.
• When a model does badly on something new, look at the examples it learned from before blaming the model itself; missing history explains a lot.
• If a prediction comes too late to act on, the fix lies in what the model is trained to predict and how far ahead, not in faster delivery of the alert.
📊 This set at a glance
Exam: AIB-C01 · 85 questions · 130 minutes · passing score 700
This set: 3 applied, 2 advanced · 3 scenario, 1 comparison, 1 trade-off
Topics: Amazon Personalize, Amazon SageMaker AI
Question 1 of 5
A company's model predicts equipment failures but only after the failure has begun. Operations want earlier warnings. What change to the model's design is needed?
Show answer
✅ Answer: D
Difficulty: applied
Detecting versus predicting
A model that flags a failure once it has started is a detection model: it learned what "failing" looks like. To warn earlier, the problem must be framed as prediction over a time horizon: "will this machine fail within the next 48 hours?"
• Labels: for each past failure, mark the readings from a window before it began (for example, 48 to 6 hours prior) as "failure coming".
• Features: use sensor trends from that earlier window, such as rising vibration or temperature drift, that precede the failure.
• Lead time: chosen so operations have enough time to act, such as scheduling maintenance.
Example
If bearing temperature starts creeping up two days before a breakdown, a model trained on those earlier readings can raise an alert while there is still time to replace the bearing.
Why each option works or fails
• ✗ A. Use less sensor data: faster inference does not change what the model learned to recognize; it still detects failures in progress and may lose useful early signals.
• ✗ B. Change alert colors: better visibility helps operators notice alerts sooner, but the alert itself still arrives after the failure has begun.
• ✗ C. More people reading late alerts: more readers do not move the alert earlier; the model still only fires once the failure has already started.
• ✓ D. Predict a set time ahead using pre-failure data (correct): training on signals from before past failures, labeled with a future horizon, teaches the model the early warning patterns operations need.
Question 2 of 5
Which statement best compares a chatbot, an assistant grounded in company data, and an agent?
Show answer
✅ Answer: A
Difficulty: advanced
Three levels of AI assistants
• Chatbot: holds a conversation using the model's general knowledge or scripted flows. Example: answers "What are your opening hours?" from a fixed script.
• Grounded assistant: retrieves company sources (policies, manuals, product data) and answers from them, often with citations. Example: "What is our refund policy for damaged items?" answered from the current policy document.
• Agent: can also plan multiple steps and take actions through tools or APIs. Example: checks an order, confirms eligibility and starts the refund.
Why the difference matters
Each level adds capability and risk. Agents can change records, so they need permissions, limits and human approval for sensitive actions. On AWS, Amazon Bedrock Knowledge Bases supports grounded answers, and Amazon Bedrock AgentCore supports building new tool-using agents (Amazon Bedrock Agents remains available to existing customers).
Note: Amazon Bedrock Agents was renamed Bedrock Agents Classic and closed to new customers on 2026-07-30; existing customers can keep using it, and AWS points new agent builds to Amazon Bedrock AgentCore. Exam questions may still name Bedrock Agents.
Why each option works or fails
• ✓ A. (correct): Chatbot converses, grounded assistant uses company sources, agent also acts: this matches the increasing capability from conversation, to answers grounded in company data, to planning and executing actions.
• ✗ B. Identical except for price: the three differ in what they can do (use company data, call tools, take actions), which changes their value, risk and required controls.
• ✗ C. Chatbot acts, agent only answers: this reverses the roles. Taking actions through tools is what defines an agent, while a basic chatbot only converses.
• ✗ D. Grounded assistant cannot use company data: being grounded means answering from company sources. A plain chatbot is the one that usually relies on general knowledge.
Question 3 of 5
A mid-sized retailer wants product recommendations on its website within one quarter. It has two data scientists and clean purchase history. Which approach best balances speed and fit for this business?
Show answer
✅ Answer: C
Difficulty: advanced · Topic: Amazon Personalize
Build, wait or use a managed service
A small team with a firm deadline should spend its time on the business problem, not on reinventing algorithms.
• Amazon Personalize is a managed service that trains recommendation models on a company's own interaction data (who bought or viewed what), item data and user data. It offers ready-made recipes such as "recommended for you" and "similar items" and serves results through an API.
• The team's work becomes preparing clean data, choosing a use case, integrating the API into the website and measuring results.
Why it fits this retailer
Clean purchase history is the key input and already exists. Two data scientists can take a managed service live within a quarter, while building from research papers would consume that time on algorithm and infrastructure work. Because the model trains on the retailer's own data, recommendations reflect its customers and catalog.
Why each option works or fails
• ✗ A. Custom algorithm from research papers: turning papers into a production system is slow and risky for two people and would likely miss the one-quarter goal with no guarantee of better results.
• ✗ B. Wait for five years of data: the retailer already has clean purchase history that is enough to start, and waiting years gives up value with no clear benefit.
• ✓ C. Managed recommendation service on its own data (correct): proven algorithms, managed infrastructure and the retailer's own purchase history together give relevant recommendations within a quarter for a two-person team.
• ✗ D. General-purpose chatbot: a chatbot does not know each visitor's purchase behavior or the live catalog and is not built to rank products for recommendations, so its suggestions would be generic.
Question 4 of 5
A finance team needs to calculate exact, deterministic monthly interest accrual for millions of loan accounts using a fixed, well-defined formula, with results that must be 100% precisely reproducible for audit purposes. Which approach is most appropriate?
Show answer
✅ Answer: A
Difficulty: applied
When not to use machine learning
Machine learning is useful when the rule is unknown or too complex to write down, so it is learned from examples. Its predictions are statistical estimates and can be slightly wrong.
Interest accrual is the opposite case. The rule is known exactly, for example interest = balance * annual_rate / 12, and auditors need the same input to give the same result every time, to the cent.
Deterministic software fits
• Ordinary code (a SQL query, a batch job, a spreadsheet formula) computes the exact answer.
• It is reproducible: rerunning it yields identical results.
• It is explainable: the formula itself is the audit trail.
An ML model trained to imitate the formula could at best match it approximately while adding cost, error and harder audits. A good rule of thumb: if you can write the formula, write the code.
Why each option works or fails
• ✓ A. Standard deterministic software (correct): a fixed, known formula is computed exactly and reproducibly by ordinary code, while any ML approach would only approximate it and add error.
• ✗ B. Deep learning regression: a neural network would learn an approximation of a formula that is already known, introducing small errors that fail the 100% reproducibility requirement.
• ✗ C. Foundation model generating figures: foundation models produce probabilistic text and can state numbers that look plausible but are wrong, and output may vary between runs, which is unacceptable for audited figures.
• ✗ D. Clustering and per-cluster estimates: grouping accounts and estimating one figure per group replaces each account's exact interest with an approximation, which defeats precise per-account accrual.
Question 5 of 5
A retailer's leadership asks why its new demand-forecasting model, trained on three years of sales, badly misjudged demand for a product line launched last month. The data team confirms the model worked as designed. What is the most likely explanation to give leadership?
Show answer
✅ Answer: A
Difficulty: applied · Topic: Amazon SageMaker AI
Models learn from the examples they have seen
A machine learning model finds patterns in its training data and applies them to new inputs. A demand-forecasting model learns things like seasonality, promotions and trends for each product from past sales.
• When a new input looks like the training data, predictions are usually reasonable.
• When it does not, the model is extrapolating: it has to guess with patterns learned from other things, and accuracy drops sharply.
The "cold start" problem
A product line launched last month has only a few weeks of sales and none of the seasonal history the model relies on. This is often called a cold-start problem, and it appears in forecasting and in recommendations alike.
Common remedies include using data from similar products as a proxy, adding product attributes as features, blending the forecast with expert judgment for new items, and retraining as history builds up.
Why each option works or fails
• ✓ A. No history for the new line (correct): the model worked as designed, but it had no examples of this product's behavior, so its forecast for it was an extrapolation rather than a learned pattern.
• ✗ B. Only multi-region products: forecasting does not require sales in more than one region; single-store or single-region forecasts are common, and region count has nothing to do with the new line's missing history.
• ✗ C. Larger GPU instance: compute size changes how fast a model trains or runs, not what it knows; more hardware cannot create sales history that does not exist.
• ✗ D. Train on fewer years: cutting history would remove seasonal patterns for existing products and still leave zero examples of the new line, so it would not fix this miss.
🎯 Key ideas from this set
• Earlier warnings require reframing the model to forecast a future window from pre-failure signals; faster processing or better alert display does not change timing.
• Chatbot, grounded assistant and agent add capabilities in order: conversation, answers from company data, then planning and actions.
• With good data, a small team and a short deadline, use a managed service trained on your own data rather than building algorithms from scratch.
• When the rule is known and exact, reproducible results are required, use deterministic code; ML is for patterns you cannot write down.
• A model is only as reliable as the examples behind it; for a brand-new product with no history, weak predictions are expected, not a malfunction.
🔗 Keep practicing on this blog
• AI Strategy and Business Value: 5 AIB-C01 Questions with Full Explanations
• 5 Questions to Test Your AI Governance and Responsible AI Knowledge (AIB-C01)
• All AIB-C01 practice sets
• Every exam on this blog
📚 Keep studying AIB-C01
More AIB-C01 questions: https://hyunhp.github.io/cert-bank/#exam=AIB-C01
Study guide: https://hyunhp.github.io/cert-bank/study.html#exam=AIB-C01
These are original practice questions, not real exam questions, and cert-bank is not affiliated with any certification vendor.
Original practice questions written for this blog, not exam dumps.
Comments
Post a Comment