AI Associate AI Fundamentals Practice Test 12 — 15 Questions

Question 1

A company uses an NLP model to detect intent in customer messages. The model works well for English but fails for Spanish messages. What is the most likely cause?

Accepted Answer

The training data is predominantly English. The most likely cause is that the training data is predominantly English, so the model never learned Spanish vocabulary, syntax, or intent patterns — this is a data coverage problem, not a model architecture or threshold problem. NLP intent classifiers are only as multilingual as their training corpus; without Spanish examples, the model cannot generalize to Spanish inputs regardless of algorithm choice.

Answer

The model is overfitted to English

Answer

The inference threshold is too high

Answer

The model uses the wrong algorithm

Question 2

What type of AI is designed to perform a specific task, such as playing chess or recommending products?

Accepted Answer

Narrow AI. Narrow AI (also called Weak AI) is designed to perform a specific, limited task such as playing chess, recommending products, or recognizing faces. It operates within a predefined domain and cannot transfer its capabilities to unrelated tasks. This matches the scenario described in the question.

Answer

Super AI

Answer

Reactive AI

Answer

General AI

Question 3

A company wants to use AI to reduce customer churn. Which TWO approaches are most appropriate? (Select 2)

Accepted Answer

Use predictive AI to score churn risk based on historical data. Option A is correct because predictive AI models (e.g., classification algorithms like logistic regression or gradient boosting) trained on historical customer data such as usage patterns, tenure, and billing history can output a churn probability score, directly enabling proactive retention targeting. Option E is correct because sentiment analysis (a natural language processing technique) applied to support tickets, chat logs, or call transcripts can surface dissatisfaction signals that often precede churn, giving an additional leading indicator beyond structured data. Option B is not the best fit because generative AI creating retention offers addresses the response after churn risk is identified, not the core churn-reduction analytics the scenario asks for, and it depends on first knowing who is at risk. Option C is irrelevant since computer vision analyzing customer photos has no meaningful connection to churn prediction. Option D is also off-target because reinforcement learning to train a chatbot focuses on conversational performance, not on identifying or reducing customer churn.

Answer

Use generative AI to create personalized retention offers

Answer

Use computer vision to analyze customer photos

Answer

Use reinforcement learning to train a chatbot

Question 4

Which THREE factors are most important for ensuring the accuracy of an AI model in a CRM context? (Select 3)

Accepted Answer

Relevant features selected for the prediction task. Option A is correct because feature relevance directly determines whether the model learns signals that actually predict the CRM outcome (e.g., churn, lead scoring); irrelevant or noisy features degrade accuracy regardless of algorithm choice. Option C is correct because representative training data ensures the model's learned distribution matches the real customer population it will score, preventing bias and poor generalization on live CRM records. Option E is correct because clean and complete data removes missing values, duplicates, and inconsistent entries that would otherwise corrupt training and skew predictions. Option B is not correct because a large number of parameters increases capacity and risks overfitting; it does not by itself improve accuracy. Option D is not correct because inference speed is a performance/latency concern, not an accuracy factor.

Answer

Large number of model parameters

Answer

Higher inference speed

Question 5

A company is deploying an AI chatbot for customer service. Which THREE ethical considerations should be addressed? (Select 3)

Accepted Answer

Protecting customer data privacy during model training and inference. Option A is correct because deploying an AI chatbot that processes customer conversations involves collecting and using personal data, so privacy must be protected during both model training and inference, typically through data minimization, anonymization, and compliance with regulations such as GDPR or CCPA. Option C is correct because biased training data can cause the chatbot to produce unfair, discriminatory, or offensive responses, so bias detection and mitigation are essential ethical safeguards. Option D is correct because transparency requires disclosing to customers when they are interacting with an AI rather than a human agent, which supports informed consent and trust. Option B is not correct because using the most complex deep learning architecture is an engineering choice, not an ethical consideration, and complexity does not inherently improve fairness or privacy. Option E is not correct because maximizing training epochs is a hyperparameter tuning decision that can cause overfitting and has no direct ethical justification.

Answer

Using the most complex deep learning architecture available

Answer

Maximizing the number of training epochs

Question 6

A sales operations manager wants to use AI to predict which leads are most likely to convert. The CRM has historical data on past leads, including whether they were won or lost, along with demographic and behavioral attributes. Which machine learning type should be used?

Accepted Answer

Supervised learning. Supervised learning is correct because the CRM contains labeled historical data — each past lead is tagged as 'won' or 'lost' — and the goal is to predict that same label for new leads. Classification models (a supervised technique) learn the mapping from demographic and behavioral features to the conversion outcome. This is the textbook definition of a supervised classification problem.

Answer

Generative AI

Answer

Unsupervised learning

Answer

Reinforcement learning

Question 7

A data scientist evaluates a churn prediction model. On the test set, the model achieves 99% accuracy, but the business reports that the model rarely flags actual churners. Which metric should the data scientist focus on to improve the model?

Accepted Answer

Recall. Recall measures the proportion of actual churners that the model correctly identifies, which is exactly what the business complaint describes — the model misses real churners (false negatives). Since the dataset is likely imbalanced (few churners), 99% accuracy can be achieved by predicting 'no churn' for everyone, making recall the metric that exposes this failure. Optimizing recall directly addresses the goal of flagging more true churners.

Answer

F1 score

Answer

Accuracy

Answer

Precision

Question 8

A customer support team wants to automatically categorize incoming cases into predefined categories such as Billing, Technical, or Account. Which NLP task is most appropriate?

Accepted Answer

Text classification. Text classification assigns predefined labels to text, making it the right choice for categorizing support cases.

Answer

Intent detection

Answer

Entity extraction

Answer

Sentiment analysis

Question 9

A CRM administrator is planning to implement predictive AI for lead scoring. Which TWO actions should be taken to ensure data quality?

Accepted Answer

Ensure lead source values are standardized (e.g., 'Web' not 'website' or 'web form'). Option A is correct because predictive lead scoring models depend on consistent categorical inputs; standardizing lead source values such as 'Web' instead of 'website' or 'web form' prevents the model from treating the same source as multiple distinct categories, which would fragment training data and degrade scoring accuracy. Option B is correct because duplicate leads inflate certain patterns and skew the model's learned relationships, while stale leads older than 5 years with no activity add noise and outdated behavioral signals that no longer reflect current buyer behavior, so removing both improves data quality. Option C is incorrect because adding as many fields as possible increases model complexity and risks overfitting rather than improving data quality. Option D is incorrect because manually relabeling all historical leads is impractical, subjective, and not a standard data-quality step for lead scoring. Option E is incorrect because using a smaller dataset reduces training signal and can worsen model performance rather than ensure data quality.

Answer

Use as many fields as possible to increase model complexity

Answer

Manually label all historical leads again to remove bias

Answer

Use a smaller dataset to train faster

Question 10

A company uses AI to generate personalized email content for marketing campaigns. They notice the AI occasionally produces factually incorrect statements. Which THREE actions should they take to mitigate this?

Accepted Answer

Implement retrieval-augmented generation (RAG) to ground outputs in verified data. Option A is correct because retrieval-augmented generation (RAG) retrieves relevant, verified documents at inference time and injects them into the prompt, grounding the model's output in authoritative source data and directly reducing fabricated statements (hallucinations). Option B is correct because lowering the temperature parameter reduces sampling randomness, making the model favor higher-probability tokens and produce more deterministic, conservative outputs that are less likely to drift into invented facts. Option C is correct because a human-in-the-loop review step catches residual factual errors before emails reach customers, providing a final verification layer that neither RAG nor temperature tuning can fully guarantee. Option D is not appropriate because training on more unlabeled internet emails adds noisy, unverified data that can increase hallucination and bias rather than improve factual accuracy. Option E is not appropriate because removing safety guardrails increases the risk of harmful or inappropriate content and does nothing to improve factual correctness.

Answer

Train the model on more unlabeled emails from the internet

Answer

Remove all safety guardrails to allow more creative content

Question 11

A sales team wants to use AI to get product recommendations for customers. Which TWO types of machine learning could be used?

Accepted Answer

Unsupervised learning. Supervised learning (E) is correct because product recommendation can be framed as a classification or regression task trained on labeled historical data, such as past customer purchases and ratings, to predict which products a customer is likely to buy. Unsupervised learning (D) is also correct because techniques like clustering and association rules can group customers with similar behavior or find products frequently bought together, enabling recommendations without explicit labels. Generative AI (A) is not the intended answer here since it creates new content rather than being the core ML paradigm for recommendation, and reinforcement learning (B) is typically used for sequential decision-making and reward optimization, not standard product recommendation. Computer vision (C) is unrelated because it processes images and video, not the tabular customer-product interaction data used for recommendations.

Answer

Generative AI

Answer

Reinforcement learning

Answer

Computer vision

Question 12

A company wants to use AI to automatically extract key information (e.g., invoice number, date, total amount) from scanned invoices. Which THREE technologies should be combined?

Accepted Answer

Optical Character Recognition (OCR). The scenario requires converting scanned invoice images into usable data, so Optical Character Recognition (OCR) (B) is essential because it converts the pixel-based text in scanned documents into machine-readable characters. Computer vision (E) is also required because it handles image preprocessing and layout analysis, such as detecting regions, correcting skew, and isolating text blocks before OCR runs. Natural Language Processing (NLP) with entity extraction (D) is the third necessary technology because it identifies and classifies the specific fields requested, such as invoice number, date, and total amount, from the OCR output. Generative AI for text generation (A) is not appropriate here because the goal is extraction, not creating new text, and sentiment analysis (C) is irrelevant since invoices contain no opinion or emotional content to classify.

Answer

Generative AI for text generation

Answer

Sentiment analysis

Question 13

A bank is implementing an AI system to approve small business loans. Which TWO ethical considerations should be addressed?

Accepted Answer

Ensure training data does not contain historical biases against certain demographics. Option A is correct because biased historical lending data can cause the AI to systematically discriminate against protected demographic groups, so the bank must audit and remediate training data to ensure fair and equitable loan decisions. Option D is correct because loan denials are adverse actions, and customers need transparency and explanations (e.g., reason codes) to understand and potentially contest the AI's decision, which also supports regulatory compliance such as ECOA/Regulation B adverse action notice requirements. Option B is not necessarily an ethical consideration; more data can improve predictions but does not by itself address fairness, privacy, or accountability. Option C is not ethically required and could remove human oversight and accountability, which is generally undesirable in high-stakes lending decisions. Option E is not ethically sufficient because maximizing accuracy alone can worsen bias and ignores fairness, explainability, and customer rights.

Answer

Use as many data points as possible to improve predictions

Answer

Replace all human loan officers with AI

Answer

Maximize model accuracy at all costs

Question 14

A Salesforce administrator at a retail company is asked to explain to the sales team how Einstein Lead Scoring determines which leads receive a high score. The team wants to know what the model actually evaluates to produce its predictions. Which statement best describes the basis of Einstein Lead Scoring?

Accepted Answer

It compares each lead's field values against patterns learned from the company's own historically converted and unconverted leads.. Einstein Lead Scoring is a supervised learning feature that examines the org's own historical leads, learns which attributes and behaviors correlate with conversion, and scores new leads by similarity to those past patterns. The score is therefore data-driven and org-specific, not a manually weighted formula, not a shared cross-customer model, and not a simple view-count ranking.

Answer

It scores leads using a pre-trained model that Salesforce trained on aggregated, anonymized data from all customer orgs.

Answer

It applies a static scoring formula that the administrator defines by assigning point values to selected lead fields.

Answer

It ranks leads purely by the number of times each lead has been viewed or opened by sales reps in the previous quarter.

Question 15

A sales manager at a manufacturing company wants to forecast next quarter's revenue more accurately. They have two years of historical opportunity data in Salesforce, including amounts, close dates, and stages. Which AI capability in Salesforce should they use to generate a predictive forecast?

Accepted Answer

Einstein Forecasting. Einstein Forecasting is the only Salesforce AI feature designed to predict future revenue based on historical opportunity data. It uses machine learning to analyze patterns and produce a forecast, which is exactly what the sales manager needs. The other options serve different purposes, such as scoring leads or opportunities individually, or capturing activities, and do not provide aggregate revenue predictions.

Answer

Einstein Activity Capture

Answer

Einstein Lead Scoring

Answer

Einstein Opportunity Scoring