AI Associate Practice Exam 54 — 20 Questions

Question 1

A Salesforce administrator is deploying an AI model that predicts customer churn. The model's decisions are not easily interpretable, and customers have asked how their data is used. Which ethical principle should the administrator prioritize to address customer concerns?

Choose an answer to begin — your selection is scored in the full session.

Accepted Answer

Transparency. Customers want to know how their data is used in the AI model, which is a matter of transparency. Transparency involves clear communication about data usage, model logic, and decision-making processes. While privacy, accountability, and inclusiveness are important ethical principles, they do not directly answer the customers' request for understanding. Therefore, prioritizing transparency is the correct action.

Answer

Accountability

Answer

Inclusiveness

Answer

Privacy

Question 2

A Salesforce admin is configuring Einstein Prediction Builder to predict which customers are likely to default on payments. The model will use historical data including payment history, credit score, and demographic information. Which TWO actions should the admin take to ensure the model aligns with ethical AI principles? (Choose two.)

Accepted Answer

Regularly audit the model's predictions for disparate impact across demographic groups.. Regular audits for disparate impact and thorough documentation are essential ethical AI practices. Audits ensure fairness by detecting bias, while documentation provides transparency and accountability. Together, they align with Salesforce's responsible AI principles. The other options either introduce bias, skip necessary checks, or remove human oversight, all of which are unethical. Thus, auditing and documenting are the correct actions.

Answer

Use the model to make final decisions without human review to increase efficiency.

Answer

Deploy the model immediately to test its real-world performance before conducting any bias checks.

Answer

Include all available demographic data to maximize predictive accuracy.

Question 3

A Salesforce developer is implementing an AI-powered service agent that uses the Einstein Trust Layer. The company wants to ensure ethical AI practices are followed. Which two actions should the developer take to uphold the principle of accountability? (Choose two.)

Accepted Answer

Maintain an audit trail of all AI interactions, including prompts and responses, for review.. Accountability in AI means having clear responsibility and oversight for the system's actions. Maintaining an audit trail ensures that AI interactions can be reviewed and traced, supporting transparency and responsibility. Designating a responsible AI officer or team establishes human ownership and decision-making authority. Together, these practices ensure that the AI system is monitored and that someone is answerable for its outcomes. Other options address privacy, safety, and user feedback, which are important but do not directly fulfill accountability.

Answer

Implement a feedback mechanism for users to report problematic AI responses.

Answer

Enable toxicity detection to filter out harmful language in AI responses.

Answer

Use data masking to redact personally identifiable information before sending prompts to the LLM.

Question 4

A Salesforce admin at a retail company is configuring Einstein Case Classification to automatically route customer support cases. The model will analyze case subject and description text. Which action best upholds the principle of transparency when deploying this AI feature?

Accepted Answer

Publish an internal wiki page explaining that Einstein Case Classification uses historical case data to suggest routing, and note that predictions may be overridden by support agents.. Transparency in AI means stakeholders understand how the system works, what data it uses, and its limitations. Informing support agents that Einstein Case Classification relies on historical case data and that predictions can be overridden gives them the context needed to use the tool responsibly. The other approaches either hide the AI's role or provide insufficient explanation, which undermines trust and ethical deployment.

Answer

Deploy the model without notifying support agents, because the predictions are only suggestions and agents retain final decision-making authority.

Answer

Enable Einstein Case Classification but restrict access to the model's prediction fields so only senior managers can view them.

Answer

Configure the model to output only a confidence score without any explanation, and share the score with the support team.

Question 5

A Salesforce admin is preparing a dataset for an Einstein Discovery model that predicts customer churn. They notice that the 'Customer_Email__c' field contains multiple blank values across thousands of records. Which data preparation step should they take FIRST to address this issue?

Accepted Answer

Analyze the missingness pattern to determine if it is random or systematic.. Before deciding how to handle missing values, it is essential to understand the nature of the missingness. Analyzing whether the missing email addresses are random or follow a pattern helps determine the best course of action, such as imputation, removal, or fixing data collection processes. This diagnostic step ensures that subsequent data preparation choices are informed and appropriate.

Answer

Create a new field that flags whether the email is missing.

Answer

Impute missing values using the most frequent email address.

Answer

Remove records with missing email addresses from the dataset.

Question 6

A Salesforce developer is building an Einstein Bot that handles customer service inquiries for a utility company. The bot must decide whether to escalate a customer's complaint to a human agent. The developer wants to ensure the bot's decision-making aligns with ethical AI principles. Which practice should be implemented?

Accepted Answer

Provide clear disclosure that the customer is interacting with an AI bot and offer an easy path to reach a human agent.. The ethical practice is to disclose the AI interaction and provide an easy path to a human agent. This respects customer autonomy, ensures transparency, and maintains human oversight for consequential decisions. Salesforce's responsible AI principles emphasize informing users when they are interacting with AI and preserving human control. Keyword escalation and unfiltered training do not systematically address fairness or accountability, and letting the bot make final decisions removes necessary human judgment.

Answer

Configure the bot to always escalate complaints containing keywords like 'lawsuit' or 'discrimination' to a human agent.

Answer

Allow the bot to make final decisions on all escalations to reduce call center costs and improve efficiency.

Answer

Train the bot on historical chat logs without filtering to maximize its ability to handle all complaint types.

Question 7

A data scientist is building an Einstein Discovery model to predict which customers are likely to default on a loan. The dataset has a field 'Income' with many missing values. The scientist decides to impute missing values using the median income. After training, the model performs well on the training data but poorly on new data. What is the most likely cause of this issue?

Accepted Answer

The median imputation introduced bias because missing values were not missing at random.. When missing values are not missing at random, median imputation can introduce bias because the imputed values do not reflect the true underlying distribution for those records. This can cause the model to learn spurious patterns that do not generalize to new data. The poor performance on new data despite good training performance suggests that the imputation method created an unrealistic representation of the data.

Answer

The median imputation reduced the variance of the 'Income' field, causing underfitting.

Answer

The model is overfitting because the dataset is too small.

Answer

The 'Income' field should have been removed entirely instead of imputed.

Question 8

A Salesforce administrator is preparing data for an AI model that predicts customer lifetime value. The dataset contains a 'Country' field with values like 'USA', 'Canada', and 'Mexico'. The administrator wants to use this field in the model but knows that machine learning algorithms require numeric input. What should the administrator do to make the 'Country' field suitable for the model?

Accepted Answer

Apply one-hot encoding to create binary columns for each country.. One-hot encoding is the correct method for nominal categorical variables like Country. It creates binary features that do not impose ordinality, allowing the model to learn separate effects for each country. This is a best practice in data preparation for AI models, ensuring that the algorithm interprets the data correctly.

Answer

Replace country names with arbitrary numeric codes (e.g., USA=1, Canada=2, Mexico=3).

Answer

Leave the field as text; Einstein Discovery automatically handles categorical variables.

Answer

Use a formula field to extract the country's ISO code and treat it as a number.

Question 9

A Salesforce admin is preparing a dataset for an Einstein Discovery model that predicts customer churn. The dataset contains a field named 'CustomerSince' with values like '2021-03-15'. The model treats this field as a categorical variable, which lowers accuracy. What should the admin do to improve the model?

Accepted Answer

Convert the 'CustomerSince' field to a numeric value representing the number of days since that date.. Dates stored as strings or treated as categories can mislead a model because the algorithm cannot interpret chronological order or duration. Transforming the date into a numeric value such as days since a reference date converts it into a continuous feature that captures tenure, which is directly relevant to churn. This is a standard feature engineering step in data preparation for AI.

Answer

Replace the date with the customer's age to provide a more stable numeric feature.

Answer

Change the field type to 'Text' in the dataset so the model treats it as a category.

Answer

Remove the 'CustomerSince' field entirely because dates are not useful for churn prediction.

Question 10

A Salesforce administrator is setting up an Einstein AI feature that generates email responses to customers. The administrator wants to ensure that customers are aware they are interacting with an AI system. Which action should the administrator take?

Accepted Answer

Add a disclaimer in the email footer stating that the response was generated by AI.. The administrator should add a disclaimer in the email footer to inform customers that the response was generated by AI. This is a straightforward way to achieve transparency, ensuring customers are aware of the AI's involvement. The other options either mislead customers, focus on privacy, or gather feedback without disclosure. Transparency about AI usage is a key ethical requirement in customer-facing AI systems.

Answer

Enable data masking for all customer personal information in the email.

Answer

Configure the AI to use a human-like name and persona in the email signature.

Answer

Set up a feedback survey after each AI-generated email to gauge customer satisfaction.

Question 11

A Salesforce developer is integrating an AI model that generates personalized marketing emails. To adhere to ethical AI principles, which two actions should the developer take? (Choose two.)

Accepted Answer

Ensure the model does not use sensitive personal data without explicit consent.. Ethical AI integration requires protecting sensitive data and respecting user choice. Ensuring sensitive data is not used without consent upholds privacy, while providing an opt-out respects autonomy. These two actions align with principles of privacy, transparency, and user control. The other options involve unrestricted data access, deception, or lack of oversight, which are unethical and should be avoided.

Answer

Hide the fact that emails are AI-generated to maintain a human touch.

Answer

Allow the model to access all customer data to maximize personalization.

Answer

Use the model to automatically send emails without any human review.

Question 12

A Salesforce administrator is using Einstein Discovery to analyze customer feedback and identify factors leading to churn. The model reveals that customers with lower income levels are more likely to churn. The administrator wants to use this insight to improve retention but is concerned about ethical implications. Which action best aligns with ethical AI principles?

Accepted Answer

Investigate whether the model is biased against low-income customers and address any bias before acting.. The administrator should first investigate whether the model is biased against low-income customers. Acting on a potentially biased insight without verification could lead to discrimination. By addressing bias before taking action, the administrator upholds fairness and accountability. The other options either exploit the insight, ignore the issue, or conceal it, all of which fail to meet ethical AI standards.

Answer

Use the insight to target low-income customers with aggressive retention offers.

Answer

Share the insight with the marketing team without mentioning the income correlation.

Answer

Ignore the income factor and focus on other variables to avoid ethical issues.

Question 13

A data scientist is preparing data for an Einstein Discovery model to predict customer lifetime value. They have a dataset with 50,000 records and 200 features. They notice that one feature, 'Total_Purchases__c', has a highly skewed distribution with a few extreme outliers. Which data preparation technique should they consider to reduce the impact of outliers?

Accepted Answer

Apply a logarithmic transformation to 'Total_Purchases__c'.. For highly skewed numerical features with outliers, a logarithmic transformation is often effective. It reduces the leverage of extreme values, making the distribution more symmetric and improving the performance of models that assume normality or are sensitive to scale. Other techniques like standardization or removal do not address the skewness directly and may introduce other issues.

Answer

Standardize 'Total_Purchases__c' by subtracting the mean and dividing by the standard deviation.

Answer

Replace outliers with the median value of 'Total_Purchases__c'.

Answer

Remove all records where 'Total_Purchases__c' is above the 99th percentile.

Question 14

A Salesforce admin is deploying an Einstein Prediction Builder model to identify customers likely to churn. The model will be used to offer retention incentives. Which TWO actions align with ethical AI principles for this deployment? (Choose two.)

Accepted Answer

Document the model's intended use, training data sources, and known limitations in a model card.. Documenting the model with a model card and monitoring for disparate impact with retraining are both ethical practices. The model card supports transparency and accountability, while bias monitoring ensures fairness across customer segments. Automatically denying service, using sensitive attributes, and failing to disclose data use all violate core ethical AI principles. These two actions help ensure the churn model is deployed responsibly and aligns with Salesforce's guidance.

Answer

Use the model to automatically deny service to predicted churners without human review.

Answer

Include sensitive attributes like race and religion as features to improve prediction accuracy.

Answer

Deploy the model without informing customers that their data is being used to predict churn.

Question 15

A Salesforce administrator is setting up a Data Cloud data stream to ingest customer order data from an external system. The external system sends records with a field 'OrderDate' in the format 'MM/DD/YYYY'. The administrator needs to ensure the data is correctly mapped and usable for AI models. What should the administrator do first?

Accepted Answer

Create a data stream and map the 'OrderDate' field to a Date data type in Data Cloud.. When ingesting data into Data Cloud, the first step is to map fields to appropriate data types. Mapping 'OrderDate' to a Date type ensures that Data Cloud can correctly parse and use the field for temporal calculations and AI features. This foundational step prevents downstream issues such as incorrect sorting or failed date functions.

Answer

Use Data Prep to split 'OrderDate' into separate month, day, and year fields.

Answer

Transform the 'OrderDate' field to a text string to avoid parsing errors.

Answer

Create a calculated insight to convert 'OrderDate' to a Unix timestamp before ingestion.

Question 16

A Salesforce developer is building an AI model to predict customer lifetime value. The training data includes historical purchase records that reflect past discriminatory pricing practices against certain demographic groups. The developer wants to mitigate the risk of perpetuating this bias. Which action should the developer take?

Accepted Answer

Audit the training data for disparate impact and apply bias mitigation techniques.. The developer should audit the training data for disparate impact and apply bias mitigation techniques. This approach directly addresses the historical discriminatory pricing practices embedded in the data, reducing the risk that the model perpetuates bias. Simply removing demographics or adding more data does not solve the problem and may even exacerbate it. Proactive bias mitigation before deployment is essential for ethical AI.

Answer

Remove all demographic attributes from the training data.

Answer

Increase the size of the training dataset by adding more historical records.

Answer

Deploy the model and monitor for bias in production, then retrain if issues arise.

Question 17

A Salesforce partner is building an AI-powered customer service chatbot using Einstein Bots and the Einstein Trust Layer. The client wants to ensure the deployment aligns with ethical AI practices. Which two actions should the partner take? (Choose two.)

Accepted Answer

Provide customers with a clear notice that they are interacting with an AI chatbot and offer an option to reach a human agent.. Ethical AI deployment with Einstein Bots and the Einstein Trust Layer requires protecting customer data and being transparent about AI use. Masking PII through the Trust Layer prevents sensitive information from reaching the LLM, while disclosing AI interaction and offering a human escalation path respects user autonomy. The other options either weaken accountability, introduce unreliable training data, or expose sensitive financial information.

Answer

Disable all audit logging to reduce storage costs, since chatbot conversations are ephemeral.

Answer

Allow the chatbot to collect and store credit card numbers in plain text to speed up payment processing.

Answer

Train the chatbot exclusively on unverified public internet data to maximize the breadth of its knowledge.

Question 18

A data engineer is preparing a dataset for an Einstein Discovery model to predict customer churn. The dataset has a field 'TotalSpend' that is highly skewed, with a few customers having extremely high values. The engineer wants to reduce the impact of these outliers on the model. Which data transformation should the engineer apply?

Accepted Answer

Log transformation. Log transformation is effective for highly skewed data because it compresses the range of extreme values, making the distribution more symmetric. This reduces the leverage of outliers, allowing the model to learn from the majority of the data without being dominated by a few extreme points. It is a standard preprocessing step for skewed numerical predictors.

Answer

Binning

Answer

Min-max normalization

Answer

Z-score standardization

Question 19

A Salesforce user is configuring an Einstein AI feature that generates email responses to customers. The user wants to ensure the AI does not generate harmful or biased content. Which Salesforce feature should be enabled to help filter inappropriate outputs?

Accepted Answer

Einstein Trust Layer. The Einstein Trust Layer is designed to help ensure responsible AI use by providing toxicity detection, data masking, and audit trails. It filters harmful or biased outputs from generative AI features, making it the correct choice for safeguarding customer email responses. The other options serve different purposes: Prediction Builder for predictions, Einstein Bot for chatbots, and Shield for security and compliance, none of which directly filter AI-generated content.

Answer

Einstein Prediction Builder

Answer

Salesforce Shield

Answer

Einstein Bot

Question 20

A marketing team at a Salesforce customer wants to use Einstein Generative AI to draft personalized email content for a campaign. Before enabling the feature, the compliance officer asks what step ensures the AI-generated content is appropriate for the brand. Which action should the team take?

Accepted Answer

Configure a human review step so that a marketing manager approves AI-generated email drafts before they are sent to customers.. Human review of AI-generated customer-facing content is essential for ethical deployment. It ensures brand alignment, catches errors or inappropriate language, and maintains accountability. The other options either bypass oversight, reduce traceability, or expose sensitive data unnecessarily. Salesforce's responsible AI principles consistently emphasize human oversight for generative AI, particularly when output reaches customers directly.

Answer

Allow the AI to access all customer records, including sensitive financial data, to make emails more personalized.

Answer

Disable all logging of AI-generated content to protect the creative process from scrutiny.

Answer

Enable the feature and send emails immediately to maximize campaign velocity, trusting the AI to produce on-brand content.