AI Associate Practice Test 23 — 15 Questions

Question 1

A company wants to use AI to analyze customer feedback from surveys and social media. Which TWO capabilities are most relevant?

Accepted Answer

Sentiment analysis. Sentiment analysis (B) is correct because it is the natural language processing capability specifically designed to determine the emotional tone (positive, negative, neutral) of customer feedback from surveys and social media, which is exactly what the company wants to analyze. Text classification (D) is also correct because it categorizes unstructured text into predefined topics or labels (e.g., complaints, praise, product issues), enabling systematic analysis of large volumes of survey and social media text. Speech recognition (A) is not relevant here because it converts spoken audio to text and the scenario involves written survey responses and social media posts, not voice data. Image recognition (C) is not relevant because it analyzes visual content such as photos or videos, which is outside the scope of text-based customer feedback. Predictive lead scoring (E) is not relevant because it ranks sales prospects by likelihood to convert, which is a sales/marketing task rather than analyzing customer feedback sentiment or topics.

Answer

Speech recognition

Answer

Image recognition

Answer

Predictive lead scoring

Question 2

A financial services company uses Einstein Bots to answer customer inquiries. A customer asks the bot to explain why their loan application was rejected. The bot provides a response based on AI predictions. Which Salesforce Trusted AI Principle is MOST directly addressed by the bot's ability to explain the decision?

Accepted Answer

Transparency. The Salesforce Trusted AI Principle of Transparency requires that AI systems be able to explain their decisions and actions in a way that is understandable to users. In this scenario, the Einstein Bot is providing an explanation for why the loan application was rejected, which directly addresses the need for transparency in AI-driven decisions. This principle ensures that customers and stakeholders can understand the reasoning behind AI outcomes, building trust and accountability.

Answer

Honesty

Answer

Accuracy

Answer

Safety

Question 3

A financial services firm uses a deep learning model to approve loans. The model is highly accurate but cannot explain its decisions. Regulators now require the firm to provide reasons for loan denials. What is the best approach to address this ethical concern?

Accepted Answer

Replace the deep learning model with a simpler, interpretable model like logistic regression.. The correct option is C: replacing the deep learning model with a simpler, interpretable model like logistic regression. In this regulated financial scenario, the firm must provide actual reasons for loan denials, and logistic regression offers inherently interpretable coefficients that directly show how each feature contributes to the decision, satisfying the regulatory requirement for explainability. Options A and B do not address explainability: removing influential features can degrade accuracy without making the model transparent, and retraining with more data only improves performance, not interpretability. Option D, post-hoc tools like LIME, only approximate the model's behavior locally and may not provide the faithful, legally defensible reasons regulators require.

Answer

Remove the most influential features from the model.

Answer

Retrain the model with more data to improve accuracy further.

Answer

Use a post-hoc explanation tool like LIME to approximate decisions.

Question 4

Refer to the exhibit. Which ethical principle is most at risk with this AI model configuration?

Accepted Answer

Transparency. The configuration has explainability disabled, which directly violates transparency. Therefore, Option D (Transparency) is the correct answer. Option A (Accountability) is at risk because human review is not required, but the most direct ethical risk is transparency. Option B (Fairness) is partially addressed by the fairness check, though not fully. Option C (Privacy) is not directly affected by this configuration.

Answer

Accountability

Answer

Fairness

Answer

Privacy

Question 5

Refer to the exhibit. What does the "Status: FAIL" indicate?

Accepted Answer

The model shows a significant disparity in true positive rates across age groups.. The correct answer is D: the FAIL status indicates the model shows a significant disparity in true positive rates across age groups. In AI/ML governance and responsible-AI tooling, a fairness check evaluates metrics such as true positive rate (equal opportunity) across protected or sensitive groups like age_group, and a FAIL flag is raised when the disparity exceeds the configured threshold. This is a fairness/bias violation rather than a general performance or data-quality problem. Option A is incorrect because overall accuracy is a separate performance metric, not what a fairness FAIL reports. Option B is incorrect because missing age_group values would be a data-completeness issue, not a true-positive-rate disparity. Option C is incorrect because a FAIL status flags a detected fairness violation; it does not by itself state that deployment is prohibited.

Answer

The model's overall accuracy is low.

Answer

The data is missing age_group information.

Answer

The model is not allowed for deployment.

Question 6

A Salesforce admin is configuring Einstein Next Best Action. Which TWO actions demonstrate ethical AI practices? (Choose two.)

Accepted Answer

Regularly review recommendation logs for patterns of bias or unfair treatment. Option D is correct because regularly reviewing recommendation logs for patterns of bias or unfair treatment is a core ethical AI practice, enabling the admin to detect and correct discriminatory outcomes in Einstein Next Best Action recommendations. Option E is correct because setting up human review for recommendations involving sensitive customer decisions preserves human oversight and accountability, which is essential for ethical AI in high-impact scenarios. Options A, B, and C do not belong: A removes human oversight entirely, which is unethical for consequential actions; B applies a one-size-fits-all recommendation that ignores individual context and can itself be unfair; and C hides the AI's reasoning, undermining transparency and the ability of business users to validate or challenge recommendations.

Answer

Allow the AI to automatically take actions without human approval for all recommendations

Answer

Use the same recommendation for all customers to ensure fairness

Answer

Keep the AI's reasoning hidden from business users to avoid confusion

Question 7

A company wants to build a chatbot that can understand natural language queries and escalate to a human agent when needed. Which tool should they use?

Accepted Answer

Einstein Bots. Einstein Bots is the correct tool because it is specifically designed to handle natural language queries in Service Cloud and can seamlessly escalate to a human agent when the bot cannot resolve the issue. It uses intent recognition and dialog flows to understand user input, and it supports handoff to live agents via Omni-Channel routing. This makes it the ideal choice for building a chatbot that requires escalation capabilities.

Answer

Einstein GPT for Service

Answer

Einstein Next Best Action

Answer

Einstein Copilot

Question 8

A company is developing an AI model to screen job applications. The training data is heavily skewed toward candidates from a specific demographic. What is the most important step the team should take to address potential ethical concerns?

Accepted Answer

Ensure the training data is diverse and representative of all demographics.. Using diverse and representative training data helps mitigate bias by ensuring the model learns from a balanced distribution of candidates. Option A is wrong because simply removing sensitive attributes does not guarantee fairness; the model can still learn proxies for those attributes, perpetuating bias. Option C is wrong because deploying the model quickly and monitoring complaints is reactive and may cause harm before issues are detected. Option D is wrong because increasing dataset size without addressing class imbalance can amplify existing biases rather than reduce them.

Answer

Remove all sensitive attributes from the data before training.

Answer

Deploy the model quickly and monitor for complaints.

Answer

Increase the overall size of the training dataset without regard to distribution.

Question 9

Which TWO actions promote transparency in AI decision-making?

Accepted Answer

Document the model's limitations and potential biases.. Option B is correct because documenting a model's limitations and potential biases is a core transparency practice: it discloses where the system may fail or produce skewed outcomes, enabling stakeholders to interpret results responsibly and satisfy governance and audit requirements. Option E is correct because providing clear explanations for significant AI decisions makes the reasoning behind outcomes understandable to users, auditors, and regulators, which is the essence of transparent decision-making (e.g., via explainability techniques such as feature attribution or model cards). The unmarked options do not belong: A (black-box models for accuracy) reduces interpretability and hides how decisions are reached; C (keeping algorithms proprietary) restricts disclosure and thus transparency; and D (sharing only final results) withholds the rationale and process, preventing meaningful scrutiny.

Answer

Use black-box models that maximize accuracy.

Answer

Keep the AI algorithms proprietary to protect intellectual property.

Answer

Only share the final results of the AI system.

Question 10

A data scientist is evaluating the performance of an Einstein Discovery model. They observe that the model has high accuracy but low precision for a specific prediction class. What does this indicate?

Accepted Answer

The model correctly predicts most instances but has many false positives for that class.. High accuracy with low precision for a specific class indicates that while the model correctly classifies the majority of instances overall, it produces a high number of false positives for that class. Precision measures the proportion of positive identifications that were actually correct, so low precision means many of the predicted positive cases are false alarms. In Einstein Discovery, this trade-off is critical when optimizing for business outcomes where false positives are costly.

Answer

The model is overfitted to the training data.

Answer

The model correctly predicts most instances but has many false negatives for that class.

Answer

The model rarely predicts that class, leading to high accuracy.

Question 11

A developer notices that an AI model performs differently for different age groups. What should be done?

Accepted Answer

Investigate the cause and evaluate fairness metrics.. The correct option is C: Investigate the cause and evaluate fairness metrics. When a model performs differently across age groups, this indicates potential bias or disparate impact, so the proper first step is to diagnose the root cause (e.g., data imbalance, proxy features, or label noise) and quantify the disparity using fairness metrics such as demographic parity, equalized odds, or disparate impact ratio. Only after this analysis can an informed mitigation be chosen. Option A is premature because simply adding data may not address the underlying cause and could even worsen bias. Option B is not automatically valid because removing age can still leave correlated proxy features and may reduce model utility. Option D is incorrect because high overall accuracy can mask subgroup performance gaps and does not satisfy fairness or compliance requirements.

Answer

Retrain with more data from all ages.

Answer

Remove age as a feature.

Answer

Ignore it if overall accuracy is high.

Question 12

A financial services firm wants to use AI to detect fraudulent transactions. They have a dataset with 1% fraudulent and 99% legitimate transactions. Which THREE actions should they take to address class imbalance? (Select THREE.)

Accepted Answer

Use F1 score as the evaluation metric. Option C is correct because F1 score balances precision and recall, making it far more informative than accuracy when the positive (fraud) class is only 1% of the data; a model predicting all legitimate transactions would score 99% accuracy but near-zero F1. Option D is correct because assigning higher class weights to the minority class during training penalizes misclassification of fraudulent transactions more heavily, forcing the model to learn the minority pattern without altering the dataset. Option E is correct because oversampling the minority class (e.g., SMOTE or random duplication) or undersampling the majority class rebalances the class distribution so the learning algorithm sees a more representative mix of fraud and legitimate cases. Option A is not appropriate because a larger but even more imbalanced dataset typically worsens the imbalance problem and does not by itself correct skewed class priors. Option B is not appropriate because accuracy is misleading on imbalanced data, as a trivial majority-class classifier achieves 99% accuracy while detecting no fraud.

Answer

Use a larger dataset even if it is more imbalanced

Answer

Use accuracy as the primary evaluation metric

Question 13

A service agent needs quick access to relevant knowledge articles while handling a case. Which Einstein feature provides article recommendations automatically?

Accepted Answer

Einstein Article Recommendations. Einstein Article Recommendations is the correct feature because it uses AI to automatically suggest relevant knowledge articles to service agents based on the context of the case, such as subject, description, and product. This reduces search time and improves case resolution efficiency by surfacing the most pertinent articles without manual querying.

Answer

Einstein Bots

Answer

Einstein Discovery

Answer

Einstein Case Classification

Question 14

A financial services company uses Salesforce Service Cloud with Einstein Bots to handle account balance inquiries. The bot currently uses a standard intent 'CheckBalance' which recognizes phrases like 'What is my balance?' and 'Show my account balance.' The company wants to expand the bot to also answer questions about recent transactions, such as 'What were my last five deposits?' and 'Show my recent withdrawals.' The system administrator has added a new intent called 'RecentTransactions' and mapped it to a new flow. However, during testing, the bot often misclassifies 'CheckBalance' requests as 'RecentTransactions' when the user mentions a specific amount or date. Which action should the administrator take to resolve this misclassification?

Accepted Answer

Add sample utterances containing amounts and date ranges to the 'CheckBalance' intent to differentiate it.. Adding sample utterances that include amounts and date ranges to the 'CheckBalance' intent provides the Einstein Bot's natural language processing (NLP) model with more training data to distinguish between balance inquiries and transaction requests. This improves intent classification accuracy by reducing overlap in the phrases the bot recognizes, directly addressing the misclassification issue.

Answer

Reduce the confidence threshold for both intents to allow more matches.

Answer

Disable the 'RecentTransactions' intent and handle transaction requests using a flow without intents.

Answer

Create a new Einstein Bot specifically for transaction inquiries and route users there.

Question 15

Which three practices help maintain data quality for AI models in Salesforce? (Choose three.)

Accepted Answer

Monitor data freshness with Data Check. Option A is correct because Data Check in Salesforce Data Cloud lets you monitor data freshness and other quality metrics so AI models are trained on current, reliable data. Option D is correct because scheduling regular data audits proactively identifies stale, incomplete, or inconsistent records before they degrade model accuracy. Option E is correct because validation rules on critical fields enforce data integrity at the point of entry, preventing invalid or malformed values from reaching AI models. Option B does not belong because disabling duplicate matching rules increases duplicate records and harms data quality rather than improving it. Option C does not belong because manual Excel updates are error-prone, unscalable, and bypass Salesforce validation and governance controls.

Answer

Disable duplicate matching rules for faster load

Answer

Use Excel for manual data updates