AI Associate Practice Exam 52 — 20 Questions

Question 1

A marketing team at a retail company configures Einstein Content Selection to personalize product recommendations. During testing, they notice that the model disproportionately recommends luxury items to customers from affluent ZIP codes and budget items to those from lower-income ZIP codes, even though the training data did not include income directly. Which ethical principle is most directly violated by this outcome?

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

Accepted Answer

Fairness. The core ethical violation is fairness, because the model produces disparate outcomes for customers grouped by a socioeconomic proxy. Even if income was not a direct feature, ZIP code can encode socioeconomic status and lead to discriminatory personalization. Salesforce's responsible AI principles emphasize avoiding bias and ensuring equitable treatment, so the team must address the proxy variable and retrain or adjust the model to mitigate the disparate impact.

Answer

Reliability and safety

Answer

Accountability

Answer

Transparency

Question 2

A Salesforce admin is preparing to implement Einstein Prediction Builder to predict which customers are likely to churn. Which two data quality considerations are most critical for the success of this predictive model? (Choose two.)

Accepted Answer

The historical data must be representative of the current customer base to avoid bias.. The two critical data quality considerations for Einstein Prediction Builder are representativeness of historical data and a clearly defined, consistently labeled target variable. Representative data ensures the model generalizes to the current customer base, while consistent labeling of churn allows the model to learn accurate patterns. Other factors like dataset size, field types, or storage location are not as critical for model success.

Answer

The data must be stored in an external data warehouse to ensure processing power.

Answer

The dataset must include at least 10,000 records to train a reliable model.

Answer

All fields used as predictors must be numeric; text fields should be excluded.

Question 3

A Salesforce administrator is preparing to enable Einstein Activity Capture for a sales team that wants emails and events logged automatically. The team also wants to understand how the feature interacts with Salesforce data storage and sharing. (Choose two.)

Accepted Answer

Activity sharing settings determine which users can see captured emails and events, so admins must configure sharing before enabling capture.. Einstein Activity Capture stores captured emails and events in a separate activity data store and governs visibility through activity sharing settings. Admins planning a rollout must account for both the non-standard storage model and the sharing configuration. Historical migration, automatic record creation, and mandatory desktop plugins are not behaviors of this feature, so they should not factor into the implementation plan.

Answer

Users must install a desktop plugin on every machine before any email or event can be captured by Einstein Activity Capture.

Answer

Einstein Activity Capture automatically creates new lead and contact records whenever it detects an unknown email address in captured messages.

Answer

Enabling Einstein Activity Capture immediately converts all historical emails in connected mailboxes into standard Salesforce EmailMessage records.

Question 4

A service manager is evaluating Einstein Case Classification to automatically route cases. Which two statements accurately describe how Einstein Case Classification works in Salesforce? (Choose two.)

Accepted Answer

It can suggest values for a picklist field on the case record, and the suggestion can be accepted or overridden by the service agent.. Einstein Case Classification uses NLP to predict picklist values on cases based on subject and description. It is configured declaratively and provides suggestions that agents can accept or override. It does not require custom code and does not send customer emails. The model is trained on your data, making it adaptable to your processes.

Answer

It automatically sends an email to the customer with the predicted case category for confirmation.

Answer

It requires the admin to manually write Apex triggers that call the Einstein Language API for each new case.

Answer

It uses a prebuilt model that cannot be customized or retrained with your own data.

Question 5

A marketing team wants to use Salesforce's AI to analyze customer data and predict which customers are likely to churn. They want to understand the factors that influence the prediction so they can take proactive retention actions. Which Einstein feature should they use?

Accepted Answer

Einstein Discovery. Einstein Discovery is a predictive analytics and AI platform within Salesforce that can analyze data to predict outcomes like churn and explain the contributing factors. It provides insights that help teams understand why a prediction is made. The other options are specialized features for activity logging, lead conversion scoring, and case classification, none of which address churn prediction with explainability.

Answer

Einstein Lead Scoring

Answer

Einstein Case Classification

Answer

Einstein Activity Capture

Question 6

A Salesforce administrator is preparing to use Einstein Prediction Builder to predict which customers are likely to renew their service contracts. The admin must ensure the model has suitable data and a valid target. Which two conditions are required for this predictive model to be built successfully? (Choose two.)

Accepted Answer

A defined outcome field that records whether each historical customer renewed. To build a supervised prediction, Einstein Prediction Builder needs a clearly defined outcome field that captures the historical result, such as renewal or non-renewal, and enough historical records containing both outcomes to learn meaningful patterns. These two conditions supply the labels and the contrast required for training. Rules engines, streaming feeds, and dashboards are not prerequisites for model creation.

Answer

A manually written rules engine that maps each customer to a renewal score

Answer

A real-time external data stream that updates customer records every second

Answer

A pre-built dashboard that visualizes renewal rates by region

Question 7

A Salesforce admin is preparing to enable Einstein Activity Capture for a sales team so that emails and events sync automatically to Salesforce. The admin must evaluate which considerations affect data visibility and governance before rollout. (Choose two.)

Accepted Answer

Admins can configure sharing settings that determine whether users see only their own captured activities or activities shared across the team.. Einstein Activity Capture stores synced activities in its own data store rather than as standard EmailMessage or Event records, and it provides sharing controls that govern whether users see only their own activities or team-wide activities. Both considerations directly affect visibility, reporting, and governance, so the admin must evaluate them before rolling out the feature to the sales team.

Answer

Einstein Activity Capture requires every user to install a desktop plugin before any email or event can sync.

Answer

Einstein Activity Capture permanently prevents admins from disabling it once enabled, so the decision is irreversible.

Answer

Einstein Activity Capture automatically deletes all existing manually logged activities to avoid duplication with synced items.

Question 8

A Salesforce admin is setting up Einstein Lead Scoring. The org has 50,000 leads, but only 2,000 have been converted, and of those, many have missing values in key fields like Industry and Lead Source. The admin wants to improve the model's predictive accuracy. What should the admin do first?

Accepted Answer

Ensure that the lead fields used for scoring are populated consistently and that there are enough converted leads to train the model.. The accuracy of Einstein Lead Scoring depends on the quality and quantity of training data. Missing values in key fields and a low number of converted leads hinder the model's ability to learn patterns. The admin should first ensure data completeness and sufficient conversion volume. Other actions like adding fields or adjusting thresholds do not fix the underlying data issues.

Answer

Enable Einstein Activity Capture to automatically log more activities, which will be used as training data.

Answer

Manually adjust the score thresholds to make the model more selective.

Answer

Increase the number of lead fields included in the scoring model to cover more attributes, even if they are sparsely populated.

Question 9

A retail company uses Salesforce Einstein to generate personalized product recommendations. The data science team discovers that the model recommends high-interest credit products primarily to customers from lower-income zip codes, even when those customers have similar credit profiles to others. The company wants to address this ethical concern. Which action best aligns with Salesforce's responsible AI principles?

Accepted Answer

Conduct a bias audit, document the findings, and apply mitigation techniques such as reweighting or adversarial debiasing before redeploying.. The correct action is to conduct a bias audit, document findings, and apply mitigation techniques. This aligns with Salesforce's responsible AI principles of fairness, transparency, and accountability. By systematically identifying and addressing bias, the company can reduce disparate impact and ensure ethical deployment. Simply removing a feature or adding a disclaimer does not adequately resolve the underlying issue.

Answer

Deploy the model as-is but add a disclaimer that recommendations may be biased.

Answer

Remove the zip code field from the training dataset and retrain the model.

Answer

Increase the model's accuracy by adding more training data from the same zip codes.

Question 10

A Salesforce administrator is configuring Einstein Generative AI for a customer service org. To adhere to ethical AI principles, which two actions should they take? (Choose two.)

Accepted Answer

Implement data masking for sensitive fields before sending data to the LLM.. Ethical AI requires accountability and privacy protection. Audit logging ensures transparency and traceability, while data masking prevents sensitive data exposure. The other options involve unrestricted data access, disabling human oversight, and using biased data, all of which violate ethical principles and increase risk.

Answer

Train the AI on all available data, including outdated and biased datasets, to maximize knowledge.

Answer

Allow the AI to access all customer data without restrictions to improve personalization.

Answer

Disable all human oversight to ensure faster AI responses.

Question 11

A Salesforce developer is implementing an AI-powered chatbot for customer service that uses Einstein Bots and an LLM integration. The company wants to ensure the solution adheres to ethical AI principles, particularly around transparency and user consent. Which two actions should the developer take? (Choose two.)

Accepted Answer

Allow users to opt out of AI interactions and connect with a human agent.. Transparency and user consent are fundamental ethical principles for AI chatbots. Informing users they are interacting with AI and allowing them to opt out to a human agent respects their autonomy and ensures informed interaction. Other actions like indefinite data storage or lack of content moderation violate ethical guidelines.

Answer

Collect user feedback only after the conversation ends and only if the user initiates it.

Answer

Store all conversation logs indefinitely for future model training.

Answer

Use the LLM to generate responses without any content moderation.

Question 12

A company uses Salesforce Einstein to analyze customer emails and extract sentiment. The AI model was trained on data that included personal health information without consent. Which ethical principle is most directly violated?

Accepted Answer

Privacy. The scenario describes training an AI model on personal health information without consent, which is a clear violation of privacy. Privacy is the ethical principle that safeguards individuals' control over their personal data, especially sensitive information like health records. Without consent, the company infringes on individuals' rights and may violate regulations like HIPAA or GDPR. Thus, privacy is the most directly violated principle.

Answer

Accountability

Answer

Transparency

Answer

Fairness

Question 13

A service organization wants its AI to detect whether incoming customer emails are frustrated or satisfied so supervisors can prioritize negative interactions. The team must choose a Salesforce capability that analyzes free-text messages and returns a sentiment value. Which capability should they use?

Accepted Answer

Einstein Sentiment Analysis. Analyzing free-text customer messages to determine emotional tone is the specific purpose of Einstein Sentiment Analysis, which returns positive, negative, or neutral values that supervisors can use to prioritize interactions. Case Classification predicts field values, Prediction Builder produces custom outcome scores, and Activity Capture only logs communications, so none of them provide sentiment.

Answer

Einstein Prediction Builder

Answer

Einstein Activity Capture

Answer

Einstein Case Classification

Question 14

A Salesforce admin is evaluating Einstein Prediction Builder for a use case that predicts whether a custom object record will exceed a service-level threshold. The admin wants to understand what type of problem the tool solves and how it produces its output. Which statement is accurate?

Accepted Answer

Einstein Prediction Builder creates a custom predictive model on a selected object and field, then writes a predicted score or outcome back to a field on each record.. Einstein Prediction Builder is a declarative tool that trains a custom model on a chosen object and target field, then writes predictions back to records. It supports custom objects and fields, making it suitable for the described service-level prediction. It does not produce narrative summaries, require external model uploads, or limit predictions to standard objects, so those alternatives mischaracterize the capability.

Answer

Einstein Prediction Builder only works on standard objects and cannot be used with custom objects or custom fields.

Answer

Einstein Prediction Builder generates a natural language summary of each record and posts it as a Chatter feed item.

Answer

Einstein Prediction Builder requires an external data science team to train models in a separate environment and upload them as packages.

Question 15

A Salesforce admin is implementing Einstein Prediction Builder for a retail company's churn prediction. The admin selects 'Customer Gender' and 'Ethnicity' as predictor fields to improve model accuracy. Which action should the admin take before deploying the model?

Accepted Answer

Remove the sensitive fields from the model and document the exclusion in the model's metadata.. Sensitive attributes like gender and ethnicity should not be used as predictors in AI models because they can lead to biased and discriminatory decisions. Removing them ensures fairness and compliance with ethical AI principles. This aligns with Salesforce's guidance to avoid protected characteristics in model training, thereby promoting equitable outcomes and maintaining trust.

Answer

Keep the fields but apply a post-processing adjustment to balance predictions across groups.

Answer

Replace the sensitive fields with anonymized versions that retain the same predictive power.

Answer

Proceed with deployment since Einstein Prediction Builder automatically handles bias mitigation.

Question 16

A developer at a financial services company is using Einstein Prediction Builder to create a model that predicts whether a customer will default on a loan. The training dataset includes a field 'ZipCode' that strongly correlates with race due to historical redlining. The developer wants to ensure the model is ethically sound. What is the most appropriate action?

Accepted Answer

Remove the 'ZipCode' field from the training dataset to avoid proxy discrimination.. Using ZipCode in a loan default model can introduce proxy discrimination because ZipCode often correlates with race due to historical housing policies. The most ethical action is to remove such proxy variables from the training data to prevent the model from learning and perpetuating bias. Post-hoc mitigation or using less granular data does not fully address the root cause.

Answer

Include 'Race' as a feature and then apply fairness constraints to balance outcomes.

Answer

Replace 'ZipCode' with a less granular geographic field like 'State' to reduce bias.

Answer

Keep 'ZipCode' but add a bias mitigation algorithm after model training.

Question 17

A company wants to use AI to automatically suggest the next best action for sales reps during an opportunity, based on the opportunity's stage, amount, and recent activity. They want the recommendations to appear directly on the opportunity record. Which Salesforce feature should they implement?

Accepted Answer

Einstein Next Best Action. Einstein Next Best Action is designed to provide context-aware recommendations to users. It uses a recommendation strategy that can incorporate business rules and predictive models to suggest actions. These recommendations can be embedded in the opportunity record, giving reps immediate guidance. The other features are for scoring, activity logging, or case classification, and do not provide next action suggestions.

Answer

Einstein Opportunity Scoring

Answer

Einstein Case Classification

Answer

Einstein Activity Capture

Question 18

A Salesforce administrator is configuring Einstein Trust Layer for a customer service chatbot. The company operates in a region with strict data protection laws. The administrator wants to ensure that no personally identifiable information (PII) from chat transcripts is stored or used to train the underlying large language model. Which configuration should the administrator verify?

Accepted Answer

Confirm that the Einstein Trust Layer zero-retention setting is enabled so prompts and responses are not stored by the LLM provider.. The Einstein Trust Layer's zero-retention setting ensures that prompts and responses are not stored or used for training by the LLM provider. This directly satisfies the data protection requirement while allowing the chatbot to operate. Encryption at rest, DLP blocking, or private VPC deployment do not address the specific concern of external retention or training use.

Answer

Set up a Data Loss Prevention (DLP) policy to block PII from being included in prompts sent to the LLM.

Answer

Enable field-level encryption on all chat transcript objects to protect PII at rest within Salesforce.

Answer

Configure the LLM to run in a private instance within the customer's own virtual private cloud (VPC).

Question 19

A retail company uses Salesforce Einstein AI to generate product recommendations. The AI model was trained on purchase data that underrepresents rural customers, leading to poor recommendations for that demographic. Which ethical principle is most directly violated?

Accepted Answer

Fairness. The scenario describes an AI system that delivers inferior recommendations to rural customers because the training data underrepresents them. This is a classic example of algorithmic bias, which violates the ethical principle of fairness. Fairness demands that AI systems treat all demographic groups equitably, and when training data is not representative, the model can systematically disadvantage certain groups. Thus, fairness is the most directly violated principle.

Answer

Privacy

Answer

Accountability

Answer

Transparency

Question 20

A Salesforce administrator is configuring Einstein Trust Layer for a new AI-powered customer service bot. The bot will summarize customer conversations and suggest responses. The administrator wants to ensure that personally identifiable information (PII) such as credit card numbers and social security numbers is not sent to the large language model (LLM). Which feature of Einstein Trust Layer should be enabled to achieve this?

Accepted Answer

Data masking. Data masking in Einstein Trust Layer identifies and redacts sensitive information such as credit card numbers and social security numbers before the data is sent to the LLM. This directly addresses the administrator's goal of preventing PII from reaching the external model, thereby upholding ethical AI practices around privacy and data minimization.

Answer

Zero data retention

Answer

Toxicity detection

Answer

Audit trail