AI Associate Practice Exam 61 — 20 Questions

Question 1

A marketing operations team is preparing to use a Salesforce predictive scoring feature on leads. The team wants to understand what the underlying model actually requires to produce meaningful scores. Which two statements correctly describe requirements or characteristics of supervised predictive scoring? (Choose two.)

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

Accepted Answer

Field values used as predictors should be populated consistently enough to be informative.. Supervised predictive scoring depends on historical records with known outcomes to learn from and on reasonably complete predictor fields to carry signal. It does not require live external data-lake connections, manual notebook retraining, or produce guaranteed accuracy, since scores are probabilistic estimates shaped by data quality and relevance.

Answer

The model requires a live internet connection to an external data lake at scoring time.

Answer

The model must be retrained manually by a data scientist using Python notebooks.

Answer

Scores are guaranteed to be perfectly accurate as long as the feature is enabled.

Question 2

A company is implementing an AI-powered chatbot using Salesforce Einstein Bots to handle customer inquiries. To ensure ethical AI practices, which two actions should the company take? (Choose two.)

Accepted Answer

Clearly disclose to users that they are interacting with an AI and not a human.. Ethical AI chatbots must prioritize transparency and user control. Disclosing AI interaction ensures users are informed, and providing an escalation path to humans respects their needs for complex issues. These actions build trust and align with principles of honesty, accountability, and user-centric design. The other options involve deception, overcollection of data, or aggressive tactics that undermine ethical standards.

Answer

Collect as much personal data as possible to improve the chatbot's responses.

Answer

Provide a seamless experience by mimicking human responses without indicating AI involvement.

Answer

Use the chatbot to upsell products aggressively to maximize revenue.

Question 3

A data engineer is preparing a dataset for an Einstein Discovery model to predict customer lifetime value. The dataset includes a field 'TotalSpend' that is highly skewed with a few extreme outliers. Which transformation should the engineer apply to improve model performance?

Accepted Answer

Apply a logarithmic transformation to the 'TotalSpend' field.. A logarithmic transformation is effective for highly skewed data with extreme outliers. It reduces skewness, making the distribution more symmetric and improving model performance. Standardization, binning, or removal do not address the underlying skewness and can lead to information loss or bias.

Answer

Bin the 'TotalSpend' values into equal-width bins.

Answer

Standardize the 'TotalSpend' field by subtracting the mean and dividing by the standard deviation.

Answer

Remove all records with 'TotalSpend' values beyond three standard deviations from the mean.

Question 4

A Salesforce administrator is configuring Einstein Trust Layer for a new AI-powered case summarization feature in Service Cloud. The legal team requires that no customer data be retained by the external large language model (LLM) provider after the response is generated. Which setting should the administrator enable to meet this requirement?

Accepted Answer

Zero Data Retention. Zero Data Retention is the Einstein Trust Layer setting that contractually and technically ensures the external LLM provider does not store any prompt or response data. Data Masking, Toxicity Detection, and Prompt Defense address different aspects of AI security and privacy but do not control post-processing data retention by the provider. Therefore, Zero Data Retention is the correct choice to meet the legal requirement.

Answer

Data Masking

Answer

Prompt Defense

Answer

Toxicity Detection

Question 5

A Salesforce admin is using Einstein Discovery to build a model predicting opportunity win probability. The dataset includes a 'CloseDate' field. The admin notices that the model's performance is poor and suspects data leakage. Which action should be taken to prevent leakage from the 'CloseDate' field?

Accepted Answer

Exclude 'CloseDate' from the model because it is not known at prediction time.. Data leakage occurs when a feature used in training is not available at prediction time. CloseDate is typically set when an opportunity is closed, so it cannot be used to predict win probability for open opportunities. Excluding it prevents the model from learning spurious correlations that won't generalize.

Answer

Convert 'CloseDate' to a numeric value (e.g., days since epoch) and include it as a feature.

Answer

Create a formula field that calculates the number of days until CloseDate and use that instead.

Answer

Leave 'CloseDate' in the model but monitor for overfitting using cross-validation.

Question 6

A financial institution uses Salesforce Einstein to generate personalized investment advice. The AI model suggests higher-risk investments to younger clients and lower-risk investments to older clients, based on age. A compliance officer raises concerns about ethical and regulatory implications. Which action should the company take to align with Salesforce's responsible AI principles?

Accepted Answer

Conduct a fairness assessment, document the use of age, and implement safeguards such as human review for high-risk recommendations.. The correct action is to conduct a fairness assessment, document the use of age, and implement safeguards like human review. This ensures the model is used ethically and transparently while still leveraging its benefits. It aligns with Salesforce's principles of fairness, accountability, and transparency. Simply removing age or disabling the model does not adequately address the ethical and regulatory concerns.

Answer

Disable the AI model and revert to manual investment advice to avoid any ethical risk.

Answer

Continue using age as a factor because it is a legitimate business criterion for investment risk.

Answer

Remove age from the model and retrain it using only financial goals and risk tolerance assessments.

Question 7

A data engineer is integrating external data into Data Cloud for an AI model. The external data contains a 'Gender' field with values 'M', 'F', 'Male', 'Female', and 'Unknown'. Which Data Cloud feature should be used to standardize these values before mapping to the Data Model Object?

Accepted Answer

Data Cloud Data Transform. Data Cloud Data Transform is the correct feature for standardizing values during ingestion. It enables mapping of inconsistent categorical values to a unified set, ensuring data quality before the data is used for AI models. Other tools operate after ingestion or in different contexts.

Answer

Salesforce Flow

Answer

Einstein Discovery

Answer

CRM Analytics Data Prep Recipe

Question 8

A data engineer is using Data Cloud to prepare a training dataset for an Einstein Discovery model that predicts customer lifetime value (CLV). The dataset contains transactional records at the line-item level, with multiple rows per customer. The model requires one row per customer. Which Data Cloud feature should the engineer use to aggregate the line items into a single customer-level record?

Accepted Answer

Data Cloud calculated insight that aggregates transaction metrics by customer.. Calculated insights in Data Cloud are purpose-built for aggregating data across multiple rows and grouping by a dimension such as customer ID. They produce one row per customer with metrics like total spend, which is exactly what the CLV model needs. Streams, row-level formulas, and identity resolution do not perform the required aggregation.

Answer

Data Cloud data stream with a transformation that groups by customer ID.

Answer

Data Cloud data model object with a formula field that sums transaction amounts.

Answer

Data Cloud identity resolution ruleset that merges transaction records by customer.

Question 9

A financial services company uses Einstein Prediction Builder to forecast loan default risk. The model includes zip code as a feature. An audit reveals that zip code strongly correlates with race due to historical housing segregation. Which ethical principle is most directly violated if the model is deployed as is?

Accepted Answer

Fairness. Zip code can act as a proxy for race, and using it in a loan default model can produce disparate impact against racial minorities. This violates the ethical principle of fairness, which demands that AI systems avoid unjust discrimination. Even if race is not explicitly used, the correlation makes the model unfairly biased, so it should not be deployed without mitigation.

Answer

Privacy

Answer

Accountability

Answer

Transparency

Question 10

A Salesforce developer is integrating an external large language model with Service Cloud to summarize customer cases. The legal team is concerned about data residency and wants to ensure customer data is not used to train the external model. Which Einstein Trust Layer capability directly addresses this concern?

Accepted Answer

Zero data retention, which ensures prompts and responses are not stored or used for model training by the external LLM provider.. Zero data retention is the Einstein Trust Layer commitment that directly addresses concerns about external LLM providers storing or training on customer data. It ensures prompts and responses are not retained beyond the immediate transaction. Dynamic grounding, toxicity detection, and prompt defense serve different purposes, such as accuracy, safety, and security, but none of them guarantee that customer data will not be used for model training.

Answer

Dynamic grounding, which retrieves relevant Salesforce records to enrich the prompt with accurate context.

Answer

Toxicity detection, which scans AI responses for harmful or offensive language before displaying them to agents.

Answer

Prompt defense, which protects against prompt injection attacks that could manipulate the model's behavior.

Question 11

A Salesforce developer is integrating an external large language model (LLM) with Salesforce to generate personalized marketing emails. The developer wants to ensure that the LLM does not generate harmful, biased, or inappropriate content. Which Einstein Trust Layer feature should be configured to filter out such content?

Accepted Answer

Toxicity detection. Toxicity detection is the Einstein Trust Layer feature that actively scans and filters AI-generated content for harmful, biased, or inappropriate language. By configuring it, the developer ensures that marketing emails generated by the LLM adhere to ethical standards and do not damage the brand or offend customers.

Answer

Data masking

Answer

Zero data retention

Answer

Audit trail

Question 12

A company uses Salesforce Einstein to analyze customer feedback and identify sentiment. They want to ensure the AI does not inadvertently discriminate against certain demographic groups. Which practice should they implement?

Accepted Answer

Regularly test the model's outputs for bias across different demographic groups.. To prevent discrimination, the company must proactively test for bias across demographic groups. This allows them to identify and address any disparities in the model's outputs. The other options either ignore the need for testing, use data improperly, or filter data in a way that hides issues.

Answer

Deploy the model without testing, as Einstein AI is pre-trained to be unbiased.

Answer

Limit the AI's analysis to only positive feedback to avoid negative sentiment.

Answer

Use a single, large dataset that includes all demographic groups without any preprocessing.

Question 13

A Salesforce admin is preparing a dataset for an Einstein Prediction Builder model that predicts whether a lead will convert. The dataset contains several fields, including 'Industry', 'AnnualRevenue', 'NumberOfEmployees', and 'LeadSource'. The admin notices that 'Industry' has many missing values, 'AnnualRevenue' has some outliers, and 'LeadSource' has inconsistent entries (e.g., 'Web', 'Website', 'web'). Which TWO data preparation steps should the admin take to improve data quality for the model? (Choose two.)

Accepted Answer

Standardize 'LeadSource' entries by mapping 'Web', 'Website', and 'web' to a single value.. Imputing missing categorical values with the mode and standardizing inconsistent categorical entries are both essential data cleaning steps. They address the specific problems of missing 'Industry' values and inconsistent 'LeadSource' entries, improving data completeness and consistency. These steps directly enhance the dataset's quality for training a reliable model.

Answer

Apply a logarithmic transformation to 'AnnualRevenue' to handle outliers.

Answer

Remove records with missing 'Industry' values.

Answer

Bin 'AnnualRevenue' into categories such as low, medium, high.

Question 14

A sales operations lead at Get Cloudy Consulting is reviewing an Einstein Lead Scoring model. The lead conversion rate is very low, and the model consistently assigns high scores to almost every lead. Which action best addresses the root cause?

Accepted Answer

Adjust the scoring model by ensuring there is a sufficient number of converted leads in the historical data used for training.. Einstein Lead Scoring depends on historical leads that converted and those that did not. When conversions are very rare, the model has little positive signal and tends to produce scores that do not separate leads well. Ensuring a sufficient number of converted leads in the training data restores the contrast the model needs. Adding fields, disabling scoring, or moving a threshold does not correct that fundamental data imbalance.

Answer

Lower the lead score threshold so that fewer leads are classified as high scoring.

Answer

Increase the number of fields Einstein Lead Scoring evaluates by adding custom fields to the lead object.

Answer

Disable Einstein Lead Scoring and rely on manual lead prioritization until conversion rates improve.

Question 15

A Salesforce developer is building an AI-powered service that recommends product bundles to customers. The model uses historical purchase data and demographic attributes such as age and gender. During testing, the developer notices that the model recommends premium bundles almost exclusively to older male customers. The developer wants to address this ethical concern. Which action should the developer take first?

Accepted Answer

Remove age and gender from the model's training data and retrain the model.. The model is exhibiting bias by recommending premium bundles based on age and gender. The first ethical action is to remove those sensitive attributes from the training data and retrain the model. This directly tackles the source of discrimination. Other options either ignore the bias, worsen it, or apply superficial fixes that do not address the root cause, so removing sensitive attributes is the correct initial step.

Answer

Adjust the recommendation threshold to favor premium bundles for all customers.

Answer

Deploy the model as-is and monitor for customer complaints.

Answer

Increase the model's complexity by adding more layers to improve accuracy.

Question 16

A Salesforce data architect is preparing a Data Cloud dataset to train an Einstein Studio model for predicting customer lifetime value. The dataset includes fields such as 'CustomerID', 'LastPurchaseDate', 'TotalSpend', and 'PreferredChannel'. The architect must ensure the data is ready for AI. Which two actions should the architect take to address potential data quality issues that could degrade model performance? (Choose two.)

Accepted Answer

Ensure 'LastPurchaseDate' is stored in a consistent date format and convert it to a recency feature.. Ensuring uniqueness of CustomerID prevents data leakage and incorrect aggregations, which are critical for reliable model training. Converting LastPurchaseDate to a consistent format and deriving a recency feature addresses both consistency and feature engineering, making the data more usable for the model. These two actions directly improve data quality and readiness for Einstein Studio, whereas imputation, arbitrary removal, and binning are either risky or premature.

Answer

Remove all records where 'TotalSpend' is zero to reduce noise.

Answer

Replace missing 'PreferredChannel' values with the most frequent channel.

Answer

Convert 'TotalSpend' to a categorical field with high, medium, and low bins.

Question 17

A service manager wants to automatically route incoming cases to the correct queue based on the case description, using AI to classify the case. Which Salesforce feature should they configure?

Accepted Answer

Einstein Case Classification. Einstein Case Classification analyzes case text to predict a category, such as Type or Reason, which can then trigger assignment rules to route cases. It is the only option that directly addresses automatic classification for routing. The other features serve different purposes: scoring predicts escalation risk, activity capture logs emails, and article recommendations suggest knowledge.

Answer

Einstein Case Scoring

Answer

Einstein Article Recommendations

Answer

Einstein Activity Capture

Question 18

A sales manager wants to use AI to improve the productivity of the sales team. Which two Salesforce Einstein capabilities can help sales reps prioritize their work and engage customers more effectively? (Choose two.)

Accepted Answer

Einstein Opportunity Scoring. Einstein Opportunity Scoring helps sales reps prioritize deals by predicting close likelihood, while Einstein Activity Capture reduces manual data entry by syncing emails and events. Together, these capabilities enable reps to focus on the right opportunities and spend more time selling. Both are part of Salesforce Sales Cloud and directly contribute to sales productivity.

Answer

Einstein Case Classification

Answer

Einstein Bots

Answer

Einstein Article Recommendations

Question 19

A data scientist is building a propensity-to-buy model in Salesforce using Einstein Discovery. The dataset includes a field 'ProductCategory' with 50 distinct values. After training, the model shows high accuracy but performs poorly on new data. What is the most likely cause and the best action to take?

Accepted Answer

The model is overfitting due to high cardinality; reduce the number of categories by grouping rare ones into an 'Other' category.. High cardinality in a categorical feature like 'ProductCategory' with 50 values can lead to overfitting, as the model may memorize rare categories. Grouping infrequent categories into an 'Other' bucket reduces the number of levels, improving generalization. This action directly addresses the discrepancy between training and new data performance without discarding the feature's predictive potential.

Answer

The model is overfitting due to too many features; apply feature selection to remove other variables.

Answer

The model is suffering from data leakage; ensure that 'ProductCategory' is not derived from the target variable.

Answer

The model is underfitting because the feature is not informative; remove 'ProductCategory' entirely.

Question 20

A Salesforce administrator at a retail company notices that an Einstein Discovery model predicting customer churn is performing poorly for customers over 65. An analysis shows the training dataset contains only 2% of records from that age group. Which action best addresses this ethical concern?

Accepted Answer

Collect additional representative data for the underrepresented age group and retrain the model.. The model's poor performance for older customers is a classic symptom of underrepresentation in training data. The ethical and effective fix is to improve data diversity by collecting more examples from that group and retraining. This reduces bias at the source and supports fairness, aligning with responsible AI principles of inclusive and representative data.

Answer

Remove the age field from the model's training dataset.

Answer

Increase the model's complexity by adding more features to improve overall accuracy.

Answer

Deploy the model only for customers under 65 and use a separate manual process for older customers.