AI Associate Data for AI Practice Test 4 — 15 Questions

Question 1

Which THREE of the following are required when setting up a data stream from Salesforce to Data Cloud?

Accepted Answer

Data Stream object definition. Option A (Data Stream object definition) is correct because creating a data stream requires you to define the Data Stream object, which specifies the source object, its fields, and the ingestion settings that Data Cloud uses to pull the data. Option C (Data Source connection) is correct because a data stream must be tied to a configured Data Source (such as a Salesforce CRM connector), which establishes the connection and credentials needed to access the source data. Option D (Data Model mapping) is correct because the incoming fields from the data stream must be mapped to the Data Cloud Data Model (DMO) so the ingested data is stored in the correct model objects and fields. Option B (Data Transform) is not required, since transformations are optional and only used when you need to reshape or derive data before mapping. Option E (Data Action) is not required, as Data Actions are used for downstream activation or engagement use cases, not for the initial setup of a data stream.

Answer

Data Transform

Answer

Data Action

Question 2

A large retail company uses Data Cloud to consolidate customer data from e-commerce, POS, and loyalty programs. They plan to use Einstein Studio to build a churn prediction model. The data architect notices that the churn model's accuracy is below expectations. Upon investigation, they find that the customer entity in Data Cloud has multiple records for the same customer with slightly different spellings and addresses. The data comes from different streams. What should the data architect do to improve the model?

Accepted Answer

Create a Data Transform to merge duplicate records based on fuzzy matching on name and address fields. The churn model's accuracy is degraded because the customer entity contains duplicate records for the same real-world customer, which fragments behavioral signals and creates inconsistent training features. A Data Transform with fuzzy matching on name and address fields is the correct Data Cloud mechanism to identify and consolidate these near-duplicate records into a single unified profile before the model consumes the data. This directly addresses the root cause — data quality/identity resolution — rather than the symptoms.

Answer

Increase the data stream frequency to get more recent data

Answer

Change the primary key in the data model to use a different identifier

Answer

Use a Calculated Insight to aggregate customer behavior over time

Question 3

A healthcare provider implements Data Cloud to predict patient readmission rates. They have HIPAA compliance requirements. The data includes sensitive patient health information (PHI). The AI model must be trained without exposing PHI to unauthorized users. The data architect uses Data Cloud's data masking on PHI fields. However, model performance drops significantly after masking because the masked values lose predictive value. What additional step should the architect consider to maintain model performance while protecting PHI?

Accepted Answer

Use tokenization for highly predictive fields like diagnosis codes instead of masking. The correct option is A: use tokenization for highly predictive fields like diagnosis codes instead of masking. Masking typically replaces values with generic or null data, which destroys the statistical signal in fields such as diagnosis codes, whereas tokenization substitutes a reversible or irreversible surrogate that preserves the one-to-one mapping and thus the predictive relationships the model needs while still preventing unauthorized users from seeing the actual PHI. Differential privacy (B) adds noise to protect individuals but would also degrade model accuracy and is not the targeted fix for lost predictive value in specific fields. Removing masking and relying on permissions (C) violates the HIPAA-driven requirement to avoid exposing PHI to unauthorized users. Simply adding more training data (D) cannot restore information that masking has already destroyed.

Answer

Implement differential privacy within Einstein Studio

Answer

Remove masking and rely on user permissions to restrict access

Answer

Increase the volume of training data to compensate for masking

Question 4

Data quality is critical for AI model performance. Which three data quality dimensions should be monitored? (Choose three.)

Accepted Answer

Completeness. Completeness (A) is a core data quality dimension because it measures whether all required values and records are present, and missing fields directly degrade model training and inference accuracy. Consistency (B) is correct because it verifies that the same data is represented uniformly across systems, formats, and time, preventing contradictory signals that confuse model learning. Timeliness (D) is correct because data must be current and available when needed; stale features or labels cause models to make decisions on outdated patterns and reduce real-world performance. Uniqueness (C) and Volume (E) are not among the three dimensions marked correct here: uniqueness addresses duplicate records and volume addresses data quantity, both of which are important but are not the dimensions selected as the correct answers for this question.

Answer

Uniqueness

Answer

Volume

Question 5

A marketing team wants to use Einstein Recommendations to personalize product offers on their e-commerce site. They have a dataset of 50,000 customers with purchase history. However, 40% of customers have no purchase history (new registrations). The model performs well for returning customers but gives generic recommendations for new ones. The team wants to improve recommendations for new customers. What data preparation step should they take?

Accepted Answer

Impute missing purchase history with the average purchase frequency across all customers.. Imputing missing purchase history with the average purchase frequency across all customers is a standard data preparation technique for handling missing values. It allows the model to make reasonable inferences for new customers without discarding them, improving recommendations by providing a baseline behavior. This approach maintains dataset size and avoids bias from removal.

Answer

Remove all customers with missing purchase history from the training set.

Answer

Assign a random purchase frequency to each new customer to add variety.

Answer

Use only customers with complete purchase history to train a more accurate model.

Question 6

A sales operations team is training an AI model to forecast quarterly revenue. They have five years of historical data, which includes a strong seasonal pattern but also a significant outlier: during the pandemic year, revenue dropped by 70% from typical values. The model trains with high accuracy on historical data but fails to predict future quarters accurately, consistently overestimating revenue. What should the data scientist do to improve forecast accuracy?

Accepted Answer

Remove the data points corresponding to the pandemic year from the training set.. Removing the pandemic year data eliminates the extreme outlier that is causing the model to learn a distorted seasonal pattern. The 70% revenue drop is not representative of future quarters, so including it forces the model to overestimate revenue to compensate for the anomaly. By training only on typical data, the model can learn the true seasonal pattern and generalize better to future quarters.

Answer

Add a binary feature indicating whether each quarter was during the pandemic.

Answer

Normalize the entire dataset using Z-scores to reduce the impact of the outlier.

Answer

Include the outlier data and increase the model capacity to capture the anomaly.

Question 7

A financial services company uses Salesforce AI to detect fraudulent transactions. The dataset has 1 million legitimate transactions and only 1,000 fraudulent ones. The model trained with default parameters achieves 99.9% accuracy but identifies no fraud (precision and recall of 0). The data scientist wants to maximize fraud detection (recall) while minimizing false positives. Which approach is most effective?

Accepted Answer

Use SMOTE to generate synthetic fraud samples to balance the dataset.. The dataset is extremely imbalanced (1,000 fraud vs. 1,000,000 legitimate), causing the model to achieve high accuracy by simply predicting the majority class. SMOTE (Synthetic Minority Over-sampling Technique) generates synthetic examples of the minority class (fraud) to balance the class distribution, forcing the model to learn patterns that distinguish fraud. This directly improves recall for the minority class while still allowing the model to learn from the majority class, thus minimizing false positives. Other methods like class weighting can help, but SMOTE is specifically designed to address imbalance by creating new, plausible minority instances, making it highly effective for this scenario.

Answer

Increase the weight of the majority class in the loss function.

Answer

Train multiple models on different random subsets and average predictions.

Answer

Use a simpler model to avoid overfitting on the majority class.

Question 8

A company is preparing their Salesforce Data Cloud for Einstein AI predictions. They need to ensure data quality and governance. Which TWO actions should they take? (Choose two.)

Accepted Answer

Create profiling and auditing dashboards to monitor data health.. Option B is correct because Data Cloud provides Data Health and Data Profiling dashboards (under Data Cloud Setup / Data Governance) that surface metrics such as completeness, freshness, and duplicate rates, letting the team continuously audit data quality before feeding it into Einstein AI predictions. Option D is correct because the Data Cloud data model (DMO-based, e.g., Individual, Contact Point, Engagement) defines the relationships and identity resolution needed to unify profiles, which is the foundation for accurate, governed features used by Einstein predictions. Option A is not appropriate because calculated insights are derived metrics, not persisted objects that support uniqueness constraints. Option C, while good security practice, addresses access control rather than the data quality and governance focus of this scenario. Option E is not recommended because automatic field mapping can introduce incorrect mappings and degrade data quality without review.

Answer

Declare uniqueness rules on calculated insights.

Answer

Set role-based access controls on data model objects.

Answer

Enable automatic field mapping for all data sources.

Question 9

A retail company uses Salesforce Data Cloud to power Einstein AI for personalized product recommendations. They have integrated customer data from multiple sources: ERP (order history), marketing automation (email engagement), and web analytics (browsing behavior). The data model includes a unified Customer__dlm object with fields: Age__c, TotalSpend__c, LastPurchaseDate__c, EmailEngagementScore__c, and WebSessionCount__c. The AI model is configured to predict "LikelyToPurchaseNextWeek__c" (Boolean). The data team has noticed that the predictions are less accurate for new customers (those with less than 30 days of data). The model was trained on all customer data without any filtering. The team wants to improve model performance without increasing training frequency. What should they do?

Accepted Answer

Include a new feature such as 'DaysSinceFirstPurchase__c' to capture customer maturity.. Including a feature like 'DaysSinceFirstPurchase__c' captures customer maturity, allowing the model to differentiate patterns between new and established customers, which improves prediction accuracy for new customers. Option A is incorrect because increasing the training window to 180 days does not provide information about customer tenure; it only includes more historical data but still lacks a feature indicating how long a customer has been active. Option C is incorrect because changing from classification to regression would alter the prediction target but does not address the lack of tenure information. Option D is incorrect because excluding new customers from the training set would bias the model and ignore a valuable segment, reducing the model's applicability.

Answer

Increase the training window from Last_90_Days to Last_180_Days to include more data.

Answer

Change the model from classification to regression to predict a probability instead of binary.

Answer

Exclude new customers from the training set to focus on established customers.

Question 10

A Salesforce admin is preparing a dataset to train an Einstein Discovery model that predicts whether an opportunity will close. The dataset contains a field 'CloseDate' that is currently a text field with values like '2024-03-15'. The admin wants to use this field as a predictor. What should the admin do to make this field suitable for the model?

Accepted Answer

Convert the text field to a Date field using a formula or data transformation.. Converting the text field to a Date field ensures Einstein Discovery can interpret the values as dates, enabling time-based analysis and proper feature engineering. Other options either fail to address the data type mismatch or introduce unnecessary complexity without solving the core issue.

Answer

Use a formula field to extract the year and month as separate numeric fields.

Answer

Create a new numeric field that represents the number of days since a reference date.

Answer

Leave the field as text and let Einstein Discovery automatically infer the date format.

Question 11

A data analyst at a retail company is preparing a dataset for an Einstein Discovery model to predict customer churn. The dataset includes a field 'CustomerTenure' with values in months. The analyst notices that some records have missing values for this field. What is the recommended approach to handle missing values in this numeric field?

Accepted Answer

Replace missing values with the mean or median of the field.. Imputing missing numeric values with the mean or median is a standard data preparation step that maintains data integrity and allows the model to leverage all records. It avoids the bias and data loss associated with deletion or arbitrary zero replacement.

Answer

Replace missing values with zero to indicate no tenure.

Answer

Delete all records with missing values to ensure data completeness.

Answer

Leave the missing values as blank and let Einstein Discovery handle them automatically.

Question 12

A data engineer is preparing a dataset for an Einstein Discovery model to predict loan default. The dataset includes a 'CreditScore' field with values ranging from 300 to 850, and an 'Income' field with values from 20,000 to 500,000. The engineer notices that the model's performance is poor and suspects that the scale difference is affecting the model. Which data preparation technique should be applied to address this issue?

Accepted Answer

Normalize both fields to a 0-1 range using min-max scaling.. Min-max scaling brings both fields to a comparable range, allowing the model to treat them equally. This is crucial when features have different units and magnitudes, as it prevents one from overshadowing the other. For Einstein Discovery, scaling numeric inputs can enhance model accuracy and training speed, especially for algorithms that rely on distance calculations.

Answer

Remove outliers from both fields to reduce the range difference.

Answer

Create a composite field by multiplying CreditScore and Income.

Answer

Apply one-hot encoding to both fields to convert them into binary indicators.

Question 13

A Salesforce admin is configuring an Einstein Prediction Builder model to predict whether a customer will churn. The training dataset includes a field 'CustomerSince' with dates, but the model fails to train. What is the most likely reason?

Accepted Answer

The dataset does not contain enough historical records to train the model.. Einstein Prediction Builder needs a sufficient volume of historical data to train a model. A common cause of training failure is an inadequate number of records. While field types and nulls can affect performance, they do not prevent training. The date field is supported, so the most likely reason is insufficient data.

Answer

The 'CustomerSince' field contains null values, which are not allowed.

Answer

Einstein Prediction Builder requires all fields to be numeric.

Answer

The 'CustomerSince' field is a date field, which is not supported for prediction.

Question 14

A Salesforce data engineer is preparing a dataset for an Einstein Discovery model that predicts customer lifetime value. The dataset includes fields such as 'TotalPurchaseAmount', 'NumberOfPurchases', 'LastPurchaseDate', and 'CustomerSinceDate'. The engineer wants to create additional features to improve model performance. Which TWO feature engineering techniques are appropriate for this scenario? (Choose two.)

Accepted Answer

Compute the number of days since the last purchase using LastPurchaseDate and the current date.. Calculating average purchase amount and computing days since last purchase are both effective feature engineering techniques that derive new, predictive variables from existing data. They enhance the model's ability to capture customer behavior patterns, leading to better predictions of lifetime value.

Answer

Replace missing values in TotalPurchaseAmount with the overall average.

Answer

Convert the LastPurchaseDate field to a text string to use as a categorical feature.

Answer

Create a binary flag indicating whether the customer has made a purchase in the last 30 days.

Question 15

A data analyst is working with a dataset in Salesforce Data Cloud to train a propensity model. They notice that the 'Annual_Revenue__c' field has values ranging from 0 to 1,000,000,000, while 'Number_of_Employees__c' ranges from 1 to 10,000. They want to ensure both features contribute equally to the model. Which technique should they apply?

Accepted Answer

Normalize both fields to a 0-1 range using min-max scaling.. Min-max scaling transforms features to a common range, such as 0 to 1, which prevents features with larger magnitudes from dominating the model. This is particularly useful when the algorithm is sensitive to feature scales, like k-nearest neighbors or neural networks. By applying min-max scaling, both annual revenue and number of employees contribute proportionally to the model's predictions.

Answer

Standardize both fields to have a mean of 0 and standard deviation of 1.

Answer

Apply one-hot encoding to both numeric fields.

Answer

Convert both fields to categorical bins based on quartiles.