Courseiva

CCNA Data For Ai Questions

75 of 113 questions · Page 1/2 · Data For Ai topic · Answers revealed

Practice these questionsExam hubAll questions
1
Multi-Selecthard

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

Select 3 answers
A.Monitor data freshness with Data Check
B.Disable duplicate matching rules for faster load
C.Use Excel for manual data updates
D.Schedule regular data audits
E.Implement validation rules on critical fields
AnswersA, D, E

Data Check alerts on stale or outdated data that could affect model accuracy.

Why this answer

Data Check in Salesforce monitors data freshness by tracking when records were last updated, ensuring that AI models use current data. Stale data can degrade model accuracy, so this practice directly supports data quality for AI.

Exam trap

The trap here is that candidates may think disabling duplicate rules speeds up data loading, but they overlook that duplicate records severely degrade AI model performance by introducing bias and noise.

Practice this question →
2
MCQmedium

A company uses Einstein Prediction Builder to predict customer churn. The data includes account creation date, number of support cases, and average payment delay. After training, the model shows low confidence scores. What is the most likely cause?

A.The training dataset includes fewer than 500 records.
B.The data contains many missing values or outliers for the selected fields.
C.The prediction field is set to a numeric type instead of a picklist.
D.The model was trained on data refreshed daily instead of weekly.
AnswerB

Missing values and outliers degrade model performance, leading to low confidence scores.

Why this answer

Low confidence scores in Einstein Prediction Builder often stem from data quality issues such as missing values or outliers. These anomalies distort the model's ability to learn meaningful patterns, leading to uncertain predictions. Clean, complete data is essential for the model to produce high-confidence scores.

Exam trap

Salesforce often tests the misconception that low confidence is caused by dataset size or refresh frequency, when in reality data quality issues like missing values or outliers are the primary culprit in Einstein Prediction Builder.

How to eliminate wrong answers

Option A is wrong because Einstein Prediction Builder does not require a minimum of 500 records; it can work with smaller datasets, though more data generally improves accuracy. Option C is wrong because the prediction field type (numeric vs. picklist) affects the type of prediction (regression vs. classification), not the confidence score directly. Option D is wrong because the refresh frequency (daily vs. weekly) impacts timeliness, not the inherent confidence of the trained model.

Practice this question →
3
MCQmedium

A company is deploying an AI model that recommends next best actions for sales reps. They notice that the model's recommendations are biased towards high-revenue opportunities. Which data-related action can help reduce this bias?

A.Use a larger neural network model
B.Encrypt the data before training
C.Oversample the underrepresented segments in the training data
D.Remove all low-revenue opportunities from the training data
AnswerC

Oversampling helps balance the representation.

Why this answer

Oversampling underrepresented segments in the training data directly addresses the class imbalance that causes the model to favor high-revenue opportunities. By increasing the frequency of low-revenue examples, the model learns to treat all segments more equally, reducing bias in its recommendations. This is a standard data-level technique for mitigating bias in AI models.

Exam trap

Salesforce often tests the misconception that model architecture changes (like larger networks) can fix data bias, when in fact the root cause is often data imbalance that must be addressed at the data level.

How to eliminate wrong answers

Option A is wrong because using a larger neural network model does not fix data imbalance; it may even amplify bias if the majority class dominates training. Option B is wrong because encrypting data protects privacy but has no effect on model bias or data distribution. Option D is wrong because removing all low-revenue opportunities would worsen the imbalance, making the model even more biased toward high-revenue opportunities.

Practice this question →
4
MCQhard

A company uses Salesforce Data Cloud to unify customer data from multiple sources. After connecting a data stream, they notice that records are missing from the unified profile. What is the most likely cause?

A.The data stream object is not a standard Salesforce object.
B.The data stream is not activated for identity resolution.
C.The data source is not from Salesforce, so it cannot be unified.
D.The reconciliation rule is not configured for the data source.
AnswerD

Reconciliation rules are needed to match records across sources.

Why this answer

Reconciliation rules in Salesforce Data Cloud define how records from different data sources are matched and merged into a unified profile. If a reconciliation rule is not configured for a data source, records from that source may not be properly linked to existing profiles, leading to missing records in the unified view. This is a common configuration step that must be completed after connecting a data stream.

Exam trap

The trap here is that candidates may confuse identity resolution (matching) with reconciliation (merging), assuming that activating identity resolution alone is sufficient to unify profiles, when in fact a reconciliation rule is required to complete the merge process.

How to eliminate wrong answers

Option A is wrong because Data Cloud supports both standard and custom objects as data stream objects; the object type does not inherently cause records to be missing from unified profiles. Option B is wrong because identity resolution activation is required for matching records across sources, but missing records are more directly caused by the lack of a reconciliation rule that defines how to merge matched records. Option C is wrong because Data Cloud is designed to unify data from any source, including non-Salesforce sources, as long as the data stream is properly configured.

Practice this question →
5
MCQeasy

Refer to the exhibit. A dataflow is set up to prepare data for a prediction model. The model is expected to predict close probability for all open opportunities. What is wrong with this dataflow?

A.The output target should be a dataset, not a model.
B.The filter on StageName is too restrictive; it excludes non-won opportunities needed for training.
C.The source should be Lead, not Opportunity.
D.The dataflow is missing a transform node to remove null values.
AnswerB

To predict close probability, the model needs examples of both won and lost deals.

Why this answer

The filter excludes all opportunities that are not 'Closed Won'. The model should be trained on both won and lost opportunities to predict close probability. The filter should be removed or include all stages.

Practice this question →
6
MCQeasy

A Salesforce admin is training an Einstein Bot to answer customer questions. Which data source should the bot use to provide accurate responses?

A.Chatter posts from the product team.
B.Knowledge articles with a published status.
C.Case records from the last 30 days.
D.Lead and contact reports.
AnswerB

Knowledge articles are designed for self-service.

Why this answer

Knowledge articles with a published status are the correct data source because they contain curated, approved, and structured information that Einstein Bot can reliably use to generate accurate responses. The bot leverages natural language processing to match customer questions against these articles, ensuring answers are based on verified content rather than unstructured or transient data.

Exam trap

Salesforce often tests the distinction between structured, authoritative data sources (like Knowledge articles) and unstructured or operational data (like Chatter or Cases), trapping candidates who assume any Salesforce data can be used for AI responses.

How to eliminate wrong answers

Option A is wrong because Chatter posts are informal, unstructured conversations that lack governance and may contain outdated or incorrect information, making them unsuitable for providing accurate, consistent responses. Option C is wrong because Case records from the last 30 days are transactional, often incomplete, and may include unresolved or duplicate issues, which would lead to unreliable answers. Option D is wrong because Lead and contact reports are designed for sales analytics and customer segmentation, not for answering product or service questions, and they lack the detailed, factual content needed for a knowledge base.

Practice this question →
7
Multi-Selectmedium

A company is training a customer service chatbot using historical conversation logs. Which TWO data preparation practices should be followed to ensure data quality?

Select 2 answers
A.Exclude all user identifiers to protect privacy
B.Include answers with varied phrasing to enhance language variety
C.Include only successful interactions that were resolved
D.Filter only English conversations for consistency
E.Use conversation logs with complete transcripts
AnswersB, E

Varied phrasing improves model generalization.

Why this answer

Training a chatbot on varied phrasing (e.g., synonyms, different sentence structures) improves its ability to understand and generate natural language responses. This practice enhances the model's robustness and generalization, preventing overfitting to a narrow set of expressions and ensuring it can handle the diverse ways customers phrase their queries.

Exam trap

Salesforce often tests the distinction between data quality practices (e.g., completeness, diversity, accuracy) and data governance practices (e.g., privacy, security), so candidates mistakenly select privacy-related options like Option A when the question explicitly asks about data quality.

Practice this question →
8
MCQmedium

Refer to the exhibit. A data file for click-through model training has the above content. Which data quality issue is most critical to address before training?

A.The header row is missing a column name for the last field
B.Missing value in the Conversions column for the third row
C.Inconsistent date formats across rows
D.Clicks column is an integer but may need scaling
AnswerB

Missing target values cannot be used for supervised learning and must be handled.

Why this answer

Missing values in the Conversions column directly impact the supervised learning target variable. If the label (conversion) is missing for a training instance, the model cannot learn the correct mapping from features to outcome, leading to biased or incomplete training. This is a critical data quality issue that must be addressed before training, typically via imputation or row removal.

Exam trap

Salesforce often tests the distinction between data quality issues that prevent training (like missing target values) versus issues that are merely preprocessing concerns (like scaling or date formatting), leading candidates to overthink minor formatting problems.

How to eliminate wrong answers

Option A is wrong because the header row missing a column name for the last field is a metadata issue, not a data quality issue; the model can still parse the data as long as the values are present and correctly ordered. Option C is wrong because inconsistent date formats across rows, while potentially problematic for feature engineering, do not directly prevent model training; date parsing can be handled during preprocessing. Option D is wrong because the Clicks column being an integer does not inherently require scaling; scaling is a preprocessing step applied to features to improve convergence, not a data quality issue that must be addressed before training.

Practice this question →
9
MCQhard

A company is using Einstein Discovery to predict customer churn. The model was created six months ago and has been making predictions. Recently, the model's accuracy has dropped significantly. The data scientist confirms that the data schema has not changed. What is the most likely reason for the drop in accuracy?

A.The data source is not being refreshed daily
B.The model's features have become irrelevant
C.The model is suffering from concept drift
D.The model needs to be retrained weekly instead of monthly
AnswerC

Concept drift happens when the statistical properties of the target variable change over time.

Why this answer

Concept drift occurs when the statistical properties of the target variable change over time, causing the model's predictions to become less accurate even though the data schema remains unchanged. In Einstein Discovery, models are trained on historical data, and if the underlying patterns of customer churn evolve (e.g., due to market shifts or new competitor behavior), the model's learned relationships become stale. Since the data schema is confirmed unchanged, concept drift is the most likely cause of the accuracy drop.

Exam trap

Salesforce often tests the distinction between data schema changes (which would affect feature availability) and concept drift (which affects the relationship between features and the target), leading candidates to incorrectly choose options about data freshness or feature relevance when the real issue is a shift in the underlying data distribution.

How to eliminate wrong answers

Option A is wrong because the data source not being refreshed daily would cause predictions to be based on outdated records, but the question states the model's accuracy dropped significantly and the schema hasn't changed—concept drift is a more fundamental issue than refresh frequency. Option B is wrong because features becoming irrelevant is a form of feature drift, but the question specifies the data schema hasn't changed, meaning the same features are still available; concept drift refers to the relationship between features and the target changing, not the features themselves. Option D is wrong because retraining weekly instead of monthly might help with drift, but the core reason for the drop is that the model's learned patterns no longer match current behavior—simply increasing retraining frequency without addressing the drift source is a band-aid, not the root cause.

Practice this question →
10
Multi-Selectmedium

A data analyst is evaluating data quality for an Einstein model. Which TWO dimensions are most critical for model accuracy?

Select 2 answers
A.Uniqueness
B.Accuracy
C.Consistency
D.Completeness
E.Timeliness
AnswersB, D

Incorrect values directly degrade model predictions.

Why this answer

Completeness (no missing values) and accuracy (correct values) are fundamental to model performance.

Practice this question →
11
Multi-Selectmedium

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

Select 3 answers
A.Data Stream object definition
B.Data Transform
C.Data Source connection
D.Data Model mapping
E.Data Action
AnswersA, C, D

Defines the stream's schema and source type.

Why this answer

A is correct because a Data Stream object definition is required to specify the schema and fields for the data being ingested from Salesforce into Data Cloud. Without this definition, Data Cloud cannot interpret the structure of the incoming records, making it impossible to map or transform the data.

Exam trap

Salesforce often tests the distinction between mandatory configuration steps and optional enhancements, so the trap here is that candidates mistake Data Transform or Data Action as required because they are commonly used in data pipelines, but they are not prerequisites for establishing the data stream itself.

Practice this question →
12
MCQhard

A company uses Einstein Forecasting for revenue prediction. The historical data shows seasonal spikes every quarter. The model consistently underestimates peak periods. What is the best data preparation step to improve accuracy?

A.Increase the forecast horizon to 12 months.
B.Add a 'quarter' index field (1-4) to the dataset.
C.Remove the spike data points as outliers.
D.Use only the last 6 months of data to reduce noise.
AnswerB

Providing explicit seasonality indicators helps the model learn periodic behavior.

Why this answer

Einstein Forecasting can detect seasonality if the data contains enough history and a seasonality marker. Adding a 'quarter' feature explicitly helps the model capture recurring patterns.

Practice this question →
13
MCQeasy

Refer to the exhibit. A developer runs a SOQL query. What does the output indicate?

A.The query returned 10 records in total.
B.The query is still processing.
C.The output is incomplete.
D.The query failed.
AnswerA

totalSize shows the number of records returned, and done=true means the query finished.

Why this answer

The SOQL query output shows '10 records returned' with no error or partial result indicator, confirming that the query completed successfully and returned exactly 10 records. In Salesforce SOQL, the query result includes a 'totalSize' field that reflects the total number of records matching the query criteria, and here it matches the number of records displayed, indicating a complete and successful retrieval.

Exam trap

Salesforce often tests the misconception that a small result set might be incomplete or that the query is still running, but the presence of a record count matching the displayed records and no error or pagination indicator confirms a complete and successful query.

How to eliminate wrong answers

Option B is wrong because SOQL queries are synchronous and either complete or fail; there is no 'still processing' state in the output—if processing were ongoing, the query would not return a result set. Option C is wrong because the output explicitly states '10 records returned' and shows all records, with no truncation or 'more records available' indicator; SOQL uses query locators for large result sets, but here the count matches the displayed records, so the output is complete. Option D is wrong because a failed query would return an error message or exception, not a list of records; the presence of a result set with a record count confirms success.

Practice this question →
14
MCQmedium

A large enterprise needs to integrate data from Salesforce CRM, an external ERP, and marketing automation to train an AI model for cross-sell recommendations. Which data storage strategy is most aligned with Salesforce's AI capabilities?

A.Use only Salesforce CRM data and ignore external sources
B.Store each source separately in Data Cloud and train models on each
C.Export all data to an external data lake and build a custom model
D.Use Salesforce Data Cloud to unify the datasets
AnswerD

Data Cloud provides harmonization, governance, and native Einstein integration.

Why this answer

Salesforce Data Cloud is designed to unify data from multiple sources into a single platform for AI and analytics. Exporting to a data lake adds complexity, using only Salesforce objects limits data scope, and storing flat files lacks governance.

Practice this question →
15
MCQmedium

A dataset contains a 'date' column. Which feature engineering technique would best capture both long-term trends and seasonal patterns?

A.Extract year, month, day as separate features.
B.Use only the day of week.
C.Create cyclic features (sin/cos of month, day).
D.Drop the date column.
AnswerC

Cyclic encoding preserves the periodic nature of time.

Why this answer

Cyclic features using sine and cosine transformations preserve the circular nature of temporal data (e.g., month 12 and month 1 are adjacent, not far apart). This allows a model to learn both long-term trends (via the year component) and seasonal patterns (via the cyclic encoding of month and day) without imposing a false linear ordering. In contrast, simple numeric extraction treats time as linear, which can misrepresent seasonal cycles.

Exam trap

Salesforce often tests whether candidates recognize that simple numeric extraction (e.g., month as 1–12) fails to model cyclical continuity, leading them to mistakenly choose Option A over the correct cyclic encoding.

How to eliminate wrong answers

Option A is wrong because extracting year, month, and day as separate numeric features introduces a linear ordering that fails to capture the cyclical relationship between months (e.g., December and January are treated as far apart). Option B is wrong because using only the day of week ignores long-term trends and seasonal patterns across months or years, capturing only weekly periodicity. Option D is wrong because dropping the date column discards all temporal information, making it impossible for the model to learn any time-based patterns.

Practice this question →
16
MCQmedium

A data engineer is troubleshooting a predictive model that stopped updating. The data flow from Data Cloud shows 'Data Transform Failed' with error: 'Field Amount cannot be null'. What is the most likely cause?

A.The data transform includes a filter that removes records with null Amount.
B.The source object has a validation rule.
C.The data flow schedule is incorrect.
D.The target field in the model requires a non-null value but source data has nulls.
AnswerD

This directly matches the error: the transform requires non-null input.

Why this answer

The error 'Field Amount cannot be null' indicates that the target field in the predictive model is configured to require a non-null value. When the data flow attempts to write records with null Amount values into that field, the transform fails. This is a common schema constraint mismatch where the source data contains nulls that violate the target field's nullability requirement.

Exam trap

Salesforce often tests the distinction between source-side constraints (validation rules) and target-side constraints (field nullability in the model schema), leading candidates to incorrectly choose Option B when the error actually originates from the target field requirement.

How to eliminate wrong answers

Option A is wrong because a filter that removes records with null Amount would prevent nulls from reaching the target, not cause a 'cannot be null' error. Option B is wrong because validation rules apply at the source object level during record creation or update, not during a data flow transform that reads data. Option C is wrong because an incorrect schedule would cause the data flow to run at the wrong time or not at all, not produce a specific transform error about a null field.

Practice this question →
17
MCQhard

A retail company uses Einstein Next Best Action with customer data from Data Cloud. The recommendations are not personalized. The admin checks the data quality dashboard and finds that the 'Customer_Profile' object has 40% records with missing 'PreferredChannel' field. What is the best course of action?

A.Remove the field from the model.
B.Impute the missing values using the mode of the field.
C.Increase the data refresh frequency.
D.Train the model with only records that have non-null PreferredChannel.
AnswerB

Imputation is a standard data cleaning technique that maintains dataset size and field utility.

Why this answer

Imputing missing values using the mode (most frequent value) of the 'PreferredChannel' field is a standard data preprocessing technique that preserves the dataset size and statistical distribution. In Einstein Next Best Action, missing categorical data can degrade model personalization, and mode imputation is a simple, effective way to handle this without losing records or altering the model structure.

Exam trap

The trap here is that candidates might think removing the field or filtering out incomplete records is simpler, but Salesforce often tests the understanding that imputation is a standard, non-destructive method to handle missing data in AI models, especially when the missing rate is high.

How to eliminate wrong answers

Option A is wrong because removing the field entirely discards potentially valuable signal from the 'PreferredChannel' feature, which could reduce model accuracy and personalization. Option C is wrong because increasing data refresh frequency does not address the root cause of missing data; it only updates the data more often without fixing the quality issue. Option D is wrong because training the model only on records with non-null 'PreferredChannel' reduces the training dataset size by 40%, which can lead to biased or less robust models and loss of valuable customer information.

Practice this question →
18
Multi-Selecthard

Which THREE factors should be considered when evaluating the quality of a dataset for an AI model?

Select 3 answers
A.Total number of records available for training.
B.Presence of outliers that may skew the model.
C.Number of distinct labels in the outcome field.
D.Percentage of missing values in key fields.
E.Number of duplicate records in the dataset.
AnswersB, D, E

Outliers can distort the model's understanding.

Why this answer

Outliers can disproportionately influence model training, especially in algorithms like linear regression or k-means clustering, leading to biased predictions. Evaluating the presence and impact of outliers is critical for ensuring the model generalizes well to unseen data.

Exam trap

Salesforce often tests the misconception that dataset size (option A) is a primary quality metric, whereas the exam emphasizes that completeness, consistency, and absence of bias (e.g., missing values, duplicates, outliers) are more critical for model reliability.

Practice this question →
19
MCQhard

You are a data scientist at a retail company. The company uses Einstein Discovery to analyze customer purchase patterns. The model is built on a dataset of 50,000 transactions. The model's R-squared is 0.85, but the predictions for new customers are consistently off by a large margin. The data includes features like 'Customer Age', 'Income', 'Previous Purchases', and 'Product Category'. The model was trained on data from the past two years. However, six months ago, the company launched a new loyalty program that significantly changed purchasing behavior. You suspect the model is not generalizing to new customers. What should you do to validate your hypothesis?

A.Create a holdout set of transactions from the last six months and compare model performance on it vs. older data
B.Exclude new customers from the dataset entirely
C.Increase the training data size to include older transactions
D.Remove the 'Product Category' feature to simplify the model
AnswerA

If performance is worse on recent data, concept drift is confirmed.

Why this answer

Creating a holdout set of transactions from the last six months directly tests whether the model's performance has degraded due to the loyalty program's impact on purchasing behavior. By comparing the R-squared or other metrics on this recent holdout set versus older data, you can quantify the drop in predictive accuracy and confirm that the model fails to generalize to the new data distribution. This approach is a standard method for detecting concept drift in machine learning models, especially when external changes (like a loyalty program) alter the underlying patterns.

Exam trap

Salesforce often tests the misconception that improving model performance (e.g., by adding more data or simplifying features) is the correct response to poor generalization, rather than first validating the hypothesis of concept drift through a time-based holdout evaluation.

How to eliminate wrong answers

Option B is wrong because excluding new customers entirely would remove the very data needed to detect the generalization failure, and it does not validate the hypothesis about model performance on new customers. Option C is wrong because increasing training data with older transactions would only reinforce the model's bias toward pre-loyalty-program patterns, making it even less adaptable to the new behavior. Option D is wrong because removing the 'Product Category' feature simplifies the model but does not address the root cause of concept drift; it may reduce accuracy further and does not test whether the loyalty program caused the shift.

Practice this question →
20
MCQhard

A company is building a text classification model for customer support tickets. They have a dataset of 10,000 tickets. The team decides to use active learning for labeling. Which approach best aligns with active learning principles?

A.Randomly select 2,000 tickets and label them manually.
B.Train a preliminary model and prioritize labeling tickets with low prediction confidence.
C.Use a pre-trained model to label all tickets automatically.
D.Have subject matter experts label all 10,000 tickets.
AnswerB

Active learning focuses on uncertain samples.

Why this answer

Active learning iteratively selects the most informative unlabeled data points for labeling, typically those with low prediction confidence from a preliminary model. This minimizes labeling effort while maximizing model performance, which is the core principle of active learning.

Exam trap

Salesforce often tests the distinction between active learning and passive learning (random sampling) or semi-supervised learning, and the trap here is assuming that any automated labeling (like using a pre-trained model) qualifies as active learning, when in fact active learning requires iterative human feedback based on model uncertainty.

How to eliminate wrong answers

Option A is wrong because random selection ignores model uncertainty, wasting labeling effort on data that may not improve the model. Option C is wrong because using a pre-trained model to auto-label all tickets bypasses the human-in-the-loop feedback essential for active learning and may propagate errors. Option D is wrong because labeling all 10,000 tickets defeats the purpose of active learning, which is to reduce labeling cost by focusing only on informative samples.

Practice this question →
21
MCQeasy

A marketer wants to use Einstein Segment Creation to build a segment for a campaign. Which data source can be used?

A.Standard report snapshots.
B.Data Cloud unified profile data.
C.External web analytics.
D.Einstein Activity Capture data.
AnswerB

Unified profiles contain the data needed for segmentation.

Why this answer

Einstein Segment Creation works with Data Cloud unified profiles.

Practice this question →
22
Multi-Selecteasy

Which TWO are common data quality issues that can negatively impact AI model performance?

Select 2 answers
A.Missing values in critical fields
B.Low model accuracy during validation
C.Insufficient storage space for data
D.Inconsistent data governance policies
E.Duplicate records in the dataset
AnswersA, E

Missing data is a common quality issue.

Why this answer

Missing values in critical fields (Option A) are a common data quality issue because many AI models, particularly those relying on statistical or gradient-based optimization, cannot handle null or NaN inputs without imputation or removal. If missing values are not addressed, the model may learn biased patterns or fail to converge, leading to degraded predictive performance.

Exam trap

Salesforce often tests the distinction between data quality issues (problems with the data itself) and model performance issues or infrastructure constraints, so candidates mistakenly select options like low accuracy or insufficient storage as data quality problems.

Practice this question →
23
MCQmedium

A data scientist needs to prepare data for Einstein Discovery. The dataset includes a field 'Customer_Status__c' with values 'Active', 'Inactive', and 'Churned'. How should this field be treated?

A.Create separate boolean fields for each value to improve model accuracy.
B.Remove the field because text fields cannot be used in Einstein Discovery.
C.Keep as a text field and let Einstein Discovery handle it as a categorical predictor.
D.Convert to numeric values 1, 2, 3 to preserve order.
AnswerC

Einstein Discovery automatically treats text fields as categorical predictors.

Why this answer

Einstein Discovery natively supports text fields as categorical predictors, automatically encoding them for model training. The platform handles string values like 'Active', 'Inactive', and 'Churned' without requiring manual transformation, preserving the semantic meaning and cardinality of the data.

Exam trap

The trap here is that candidates assume text fields must be converted to numbers or one-hot encoded for machine learning, but Einstein Discovery abstracts this preprocessing, and manual conversion can introduce ordinal bias or unnecessary complexity.

How to eliminate wrong answers

Option A is wrong because creating separate boolean fields for each value (one-hot encoding) is unnecessary and can introduce multicollinearity or increase feature dimensionality without benefit, as Einstein Discovery's internal preprocessing already handles categorical encoding optimally. Option B is wrong because text fields are fully supported in Einstein Discovery as categorical predictors; the platform does not require numeric-only inputs and can process string values directly. Option D is wrong because converting to numeric values 1, 2, 3 implies an ordinal relationship that does not exist among 'Active', 'Inactive', and 'Churned', which would mislead the model into treating the categories as ordered, degrading prediction accuracy.

Practice this question →
24
MCQeasy

A team is building a pipeline to train a model daily. The source data arrives in CSV files but needs to be converted to Parquet for efficiency. Which pipeline step should perform this conversion?

A.Feature engineering step
B.Model deployment step
C.Data validation step
D.Data ingestion step
AnswerD

Ingestion can transform data into a more efficient format.

Why this answer

The data ingestion step is responsible for bringing raw data into the pipeline, including format conversions like CSV to Parquet. Converting to Parquet at ingestion improves storage efficiency and query performance for downstream processing, as Parquet uses columnar storage and compression.

Exam trap

Salesforce often tests the distinction between data ingestion (raw data handling) and data validation (quality checks), leading candidates to confuse format conversion with validation steps.

How to eliminate wrong answers

Option A is wrong because feature engineering transforms existing data into features for model training, not raw format conversion. Option B is wrong because model deployment serves the trained model for inference, not data preprocessing. Option C is wrong because data validation checks data quality and schema compliance, but does not perform format conversion.

Practice this question →
25
Multi-Selecthard

Which THREE actions are recommended when preparing data for Einstein Next Best Action? (Choose 3)

Select 3 answers
A.Provide data on which actions were offered and whether they were accepted
B.Include at least 10 different action types per strategy
C.Record rejections (actions not taken) as negative examples
D.Use only historical data from the last 30 days
E.Retrain the model weekly with fresh interaction data
AnswersA, C, E

This is essential for reinforcement learning.

Why this answer

Einstein Next Best Action (NBA) requires historical interaction data showing which actions were offered and whether they were accepted to train the predictive model. This feedback loop enables the AI to learn which actions are most effective for specific customer contexts, directly improving recommendation accuracy.

Exam trap

Salesforce often tests the misconception that more action types or recent data alone improve model performance, when in fact the key requirements are balanced positive/negative examples, sufficient historical depth, and regular retraining with fresh interaction data.

Practice this question →
26
Multi-Selectmedium

A data analyst is troubleshooting Einstein Article Recommendations that are not showing up on the site. Which TWO checks should be performed first? (Choose 2)

Select 2 answers
A.Ensure at least 100 articles are in the knowledge base
B.Confirm that article authors have the correct profile permissions
C.Check that article view events are being captured in the data
D.Increase the recommendation frequency from daily to hourly
E.Verify that the recommendation model is published and active
AnswersC, E

Without view data, the model has no basis to recommend.

Why this answer

Einstein Article Recommendations rely on user interaction data, specifically article view events, to generate personalized recommendations. If these events are not being captured, the model has no input to learn from, and recommendations will not appear. Checking event capture is a fundamental first step in troubleshooting data pipeline issues.

Exam trap

Salesforce often tests the misconception that increasing data volume or frequency (Options A and D) will fix recommendation issues, when in fact the core problem is usually missing event data or an inactive model.

Practice this question →
27
Multi-Selecteasy

Which THREE types of data sources are commonly integrated into Salesforce Data Cloud for AI use cases?

Select 3 answers
A.Third-party demographic data
B.Web and mobile app engagement data
C.CRM transaction records
D.Model training logs
E.Data transformation scripts
AnswersA, B, C

External data enhances AI models.

Why this answer

Salesforce Data Cloud can ingest third-party demographic data from external sources (e.g., data enrichment providers) to enrich customer profiles. This data, when combined with first-party data, enables AI models to generate more accurate predictions and segmentations. Data Cloud’s Data Streams and Data Lake objects support structured ingestion of such external datasets.

Exam trap

Salesforce often tests the distinction between data sources (raw inputs) and data processing artifacts (logs, scripts), leading candidates to mistakenly select model training logs or transformation scripts as valid data sources.

Practice this question →
28
Multi-Selecthard

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

Select 3 answers
A.Completeness
B.Consistency
C.Uniqueness
D.Timeliness
E.Volume
AnswersA, B, D

Ensures no missing values that could bias the model.

Why this answer

Completeness, timeliness, and consistency are fundamental data quality dimensions. Volume is not a quality dimension; uniqueness is related to consistency but not always required.

Practice this question →
29
MCQeasy

An administrator is configuring a Salesforce AI model that uses historical sales data. The data includes fields like 'Amount', 'Close_Date', and 'Lead_Source'. What is the primary purpose of data preprocessing in this context?

A.To generate visualizations for business stakeholders
B.To increase the storage capacity of the database
C.To enforce data access permissions for different user roles
D.To clean and transform data into a format suitable for model training
AnswerD

Preprocessing ensures data quality and format.

Why this answer

Data preprocessing is essential for AI models because raw historical sales data often contains missing values, inconsistent formats, and noise. Cleaning (e.g., handling nulls in 'Amount') and transforming (e.g., encoding 'Lead_Source' into numerical features) ensure the model can learn patterns effectively, directly impacting training accuracy and convergence.

Exam trap

Salesforce often tests the distinction between data preprocessing and other data management tasks; the trap here is that candidates confuse preprocessing with reporting (visualizations) or security (permissions), when the core goal is to prepare data for model ingestion.

How to eliminate wrong answers

Option A is wrong because generating visualizations is a downstream analytics task, not the primary purpose of preprocessing for model training. Option B is wrong because preprocessing does not increase storage capacity; it may reduce data size through cleaning but does not affect database storage limits. Option C is wrong because enforcing data access permissions is a security and governance concern, handled by Salesforce's sharing and permission settings, not by data preprocessing steps.

Practice this question →
30
MCQhard

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?

A.Increase the weight of the majority class in the loss function.
B.Use SMOTE to generate synthetic fraud samples to balance the dataset.
C.Train multiple models on different random subsets and average predictions.
D.Use a simpler model to avoid overfitting on the majority class.
AnswerB

SMOTE creates synthetic instances of the minority class, allowing the model to learn fraud patterns effectively and improve recall.

Why this answer

With extreme imbalance, oversampling the minority class (e.g., SMOTE) generates synthetic fraud examples, helping the model learn fraud patterns and improve recall without discarding legitimate data.

Practice this question →
31
MCQeasy

A Salesforce admin wants to use Einstein Prediction Builder to predict case resolution time. What type of data is most critical for training this model?

A.Customer satisfaction survey responses
B.Historical case records including resolution time
C.Product inventory levels
D.Employee work schedules
AnswerB

Historical data is essential for training.

Why this answer

Einstein Prediction Builder requires historical data with known outcomes to train a supervised machine learning model. Historical case records containing actual resolution times provide the labeled examples needed for the model to learn patterns and predict future case resolution times. Without this ground truth data, the model cannot be trained to make accurate predictions.

Exam trap

The trap here is that candidates may confuse factors that influence resolution time (like employee schedules or inventory) with the actual labeled outcome data required to train a supervised prediction model.

How to eliminate wrong answers

Option A is wrong because customer satisfaction survey responses measure post-resolution sentiment, not the actual resolution time, and they lack the precise timestamp data required for regression-based time prediction. Option C is wrong because product inventory levels are unrelated to case resolution time; they might be relevant for supply chain predictions but not for service case duration. Option D is wrong because employee work schedules, while potentially influencing resolution time, are not the historical outcome data needed to train the model — the model needs actual resolution times from past cases, not staffing inputs.

Practice this question →
32
MCQmedium

A company is preparing customer data for a predictive model. They notice that many records have missing values for the 'annual income' field. Which approach is best to handle this issue while minimizing bias?

A.Remove all records with missing values.
B.Use model-based imputation considering other features.
C.Replace missing values with the mean.
D.Set missing values to zero.
AnswerB

Model-based imputation leverages other features to predict missing values, preserving relationships and minimizing bias.

Why this answer

Model-based imputation (Option B) is best because it uses relationships between features (e.g., education, job role) to predict missing 'annual income' values, preserving data distribution and minimizing bias. This approach avoids the distortion caused by simple mean/zero imputation and retains sample size better than deletion.

Exam trap

Salesforce often tests the misconception that mean imputation is a safe default, but the trap here is that it ignores feature dependencies and can artificially shrink variance, leading to overconfident model predictions and biased coefficients.

How to eliminate wrong answers

Option A is wrong because removing all records with missing values can introduce selection bias and reduce sample size, potentially discarding valuable patterns in the data. Option C is wrong because replacing missing values with the mean ignores feature correlations, artificially compresses variance, and can bias relationships in the predictive model. Option D is wrong because setting missing values to zero is arbitrary and unrealistic for income data, likely creating a skewed distribution and misleading model coefficients.

Practice this question →
33
MCQhard

A data pipeline fails intermittently when processing large CSV files. The error log shows 'OutOfMemoryError'. Which configuration change is most likely to resolve this?

A.Use a smaller file size limit.
B.Increase the number of worker threads.
C.Switch to XML format.
D.Increase the heap memory for the processing application.
AnswerD

Increasing heap memory provides more space for large file processing.

Why this answer

The OutOfMemoryError indicates that the Java Virtual Machine (JVM) heap space is exhausted while processing large CSV files. Increasing the heap memory (e.g., using -Xmx flag) allocates more memory to the application, allowing it to handle larger datasets without crashing. This directly addresses the root cause of insufficient memory for the data pipeline's processing workload.

Exam trap

Salesforce often tests the misconception that increasing parallelism (worker threads) solves memory issues, but in reality, more threads increase memory pressure and can trigger OutOfMemoryError faster.

How to eliminate wrong answers

Option A is wrong because using a smaller file size limit is a workaround that avoids the problem rather than solving it, and it may not be feasible if large files are required by the business. Option B is wrong because increasing worker threads typically increases memory consumption and contention, which would worsen the OutOfMemoryError, not resolve it. Option C is wrong because switching to XML format would likely increase memory usage due to verbose markup and parsing overhead, making the error more likely, not less.

Practice this question →
34
MCQhard

A company has international customers and wants Einstein Prediction Builder to forecast deal closure probability. The data includes fields like 'region', 'product line', and 'deal amount'. What is a best practice to ensure the model works for all regions?

A.One-hot encode the region field using 50+ dummy variables.
B.Remove the region field to avoid bias.
C.Use region as a numeric rank based on past conversion rates.
D.Group regions into broader categories like 'Americas', 'EMEA', 'APAC'.
AnswerD

Grouping reduces noise and improves generalizability while maintaining regional distinction.

Why this answer

Grouping regions into broader categories like 'Americas', 'EMEA', and 'APAC' reduces high cardinality and sparsity in categorical features, which improves model stability and prevents overfitting in Einstein Prediction Builder. This approach ensures each region group has sufficient training data to learn meaningful patterns, enabling the model to generalize better across all regions without introducing bias from rare categories.

Exam trap

Salesforce often tests the misconception that more granular data (like one-hot encoding with many categories) always improves model accuracy, when in fact it can harm performance due to sparsity and overfitting in prediction builder tools.

How to eliminate wrong answers

Option A is wrong because one-hot encoding a region field with 50+ dummy variables introduces high cardinality and sparsity, which can cause the model to overfit to rare categories and degrade prediction performance in Einstein Prediction Builder. Option B is wrong because removing the region field entirely discards valuable geographic information that can significantly influence deal closure probability, leading to a less accurate model. Option C is wrong because using region as a numeric rank based on past conversion rates introduces ordinal bias and assumes a linear relationship that may not exist, which can misrepresent the true categorical nature of the data and reduce model interpretability.

Practice this question →
35
MCQeasy

A company wants to train an AI model to predict customer churn using historical data that contains many missing values. What is the best practice for handling missing data?

A.Use only features without missing values.
B.Ignore missing values as they do not affect AI training.
C.Impute missing values using mean or median.
D.Remove all records with missing values.
AnswerC

Imputation preserves data and reduces bias.

Why this answer

Imputing missing values using mean or median is a standard practice that preserves the dataset size and statistical properties, allowing the AI model to learn from all available features without introducing bias from data removal. This approach is particularly effective for numerical features in customer churn prediction, where missing values are often random and imputation maintains the distribution for algorithms like logistic regression or gradient boosting.

Exam trap

Salesforce often tests the misconception that removing missing data is safe, but the trap here is that candidates overlook how data removal can shrink the dataset and introduce bias, while imputation is a more balanced and widely accepted practice in AI workflows.

How to eliminate wrong answers

Option A is wrong because discarding features with missing values can remove valuable predictors, reducing model accuracy and ignoring the fact that missingness itself may carry predictive signal. Option B is wrong because ignoring missing values causes most AI algorithms to fail or produce incorrect results, as they cannot process null or NaN entries, leading to runtime errors or biased learning. Option D is wrong because removing all records with missing values can drastically reduce the dataset size, introduce selection bias, and discard potentially useful patterns in the remaining data.

Practice this question →
36
MCQeasy

In Salesforce CRM Analytics (formerly Einstein Analytics), what is the primary purpose of a dataset?

A.To prepare data for AI and analytics
B.To run SQL queries directly
C.To store raw, unprocessed records
D.To create dashboards only
AnswerA

Datasets are the building blocks for AI modeling, dashboards, and analytical queries.

Why this answer

In Salesforce CRM Analytics, a dataset is the foundational data structure that transforms raw data into an optimized, columnar format for analytics and AI features like Einstein Discovery. It is created by extracting, cleaning, and aggregating data from sources such as Salesforce objects or external connectors, enabling efficient querying, dashboarding, and machine learning model training. This makes option A correct because the primary purpose is to prepare data specifically for AI and analytics workloads.

Exam trap

Salesforce often tests the misconception that datasets are simply raw storage containers, but the trap here is that candidates overlook the 'preparation for AI' aspect and choose 'store raw records' because they confuse datasets with database tables or data lakes.

How to eliminate wrong answers

Option B is wrong because datasets do not support direct SQL query execution; instead, they use SAQL (Salesforce Analytics Query Language) or lens-based exploration for querying. Option C is wrong because datasets store processed, flattened, and indexed data, not raw, unprocessed records—raw data is typically held in dataflows or external systems before transformation. Option D is wrong because while datasets can be used to build dashboards, their primary purpose is broader, encompassing AI, analytics, and data preparation, not just dashboard creation.

Practice this question →
37
MCQmedium

What is the most likely cause of the error?

A.Authentication failure
B.Data quality threshold violation
C.Data schema mismatch
D.Network timeout
AnswerB

Null values exceed acceptable threshold.

Why this answer

The error mentions a high percentage of null values in a critical field, which violates a data quality threshold. Option A is wrong because schema mismatch would show field type inconsistencies. Option C is wrong because authentication failure would show a different error.

Option D is wrong because network timeout would mention connection issues.

Practice this question →
38
Multi-Selecthard

Which TWO techniques are commonly used to handle missing values in a dataset for AI training?

Select 2 answers
A.L1 regularization
B.Deletion of rows with missing values
C.One-hot encoding
D.Min-max normalization
E.Imputation with mean or median
AnswersB, E

Simple but valid method.

Why this answer

Deleting rows with missing values is a straightforward technique to handle missing data, especially when the missingness is random and the dataset is large enough that removing a few rows does not significantly impact model performance. This approach avoids introducing bias from imputation methods but can lead to loss of valuable information if too many rows are removed.

Exam trap

Salesforce often tests the distinction between data preprocessing techniques (like handling missing values) and model regularization or feature engineering, so candidates may confuse L1 regularization or one-hot encoding as methods for missing data when they serve entirely different purposes.

Practice this question →
39
MCQeasy

When training an Einstein Discovery model, which data type is not supported as a predictor field?

A.Multi-select picklist
B.Numeric
C.Picklist
D.Date
AnswerA

Multi-select picklists have multiple values per record and cannot be used directly as predictors.

Why this answer

Multi-select picklists are not supported as predictors in Einstein Discovery. Numeric, picklist, and date fields are supported.

Practice this question →
40
MCQeasy

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?

A.Remove all customers with missing purchase history from the training set.
B.Assign a random purchase frequency to each new customer to add variety.
C.Impute missing purchase history with the average purchase frequency across all customers.
D.Use only customers with complete purchase history to train a more accurate model.
AnswerC

Imputation provides a baseline signal for new customers, enabling the model to make reasonable recommendations.

Why this answer

Imputing missing purchase data with a sensible default (e.g., average purchase frequency) gives the model signal for new customers, improving recommendations without discarding data.

Practice this question →
41
MCQmedium

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?

A.Create a Data Transform to merge duplicate records based on fuzzy matching on name and address fields
B.Increase the data stream frequency to get more recent data
C.Change the primary key in the data model to use a different identifier
D.Use a Calculated Insight to aggregate customer behavior over time
AnswerA

Directly addresses the duplicate issue and creates a unified view.

Why this answer

The best course of action because creating a Data Transform with fuzzy matching merges duplicates into a single clean record, improving data quality for the model. Option B is flawed because increasing frequency does not fix existing duplicates. Option C aggregates but doesn't resolve the duplication.

Option D changes the primary key but duplicates remain.

Practice this question →
42
MCQmedium

Refer to the exhibit. A data scientist tries to query the dataset but receives an error. Which of the following is the most likely cause?

A.The requested fields are not included in the policy.
B.The condition filters out records with amount=5000.
C.The data scientist is not listed in the allowedUsers array.
D.The policy format is invalid JSON.
AnswerA

If the query requests a field not listed (e.g., customer_name), it would be denied.

Why this answer

'data_scientist' is in allowedUsers, so they are allowed. Option B is not in policy, C the policy filters amounts >0 and <10000, so 5000 is included, D the fields are in the policy, so they should be accessible.

Practice this question →
43
Multi-Selecthard

Before training an Einstein Prediction model, a data analyst must perform data quality checks. Which THREE checks are most critical?

Select 3 answers
A.Confirm that label distribution matches the target baseline
B.Remove duplicate records that could cause data leakage
C.Verify consistent data types across records (e.g., all dates as Date)
D.Ensure all features follow a normal distribution
E.Check for missing values in key fields
AnswersB, C, E

Duplicates can over-represent certain patterns.

Why this answer

Duplicate records can cause data leakage by allowing the model to see the same or highly similar data in both training and validation splits, leading to overfitting and inflated performance metrics. Removing duplicates ensures that the model generalizes to unseen data rather than memorizing repeated instances.

Exam trap

Salesforce often tests the misconception that all features must be normally distributed, which is a requirement for some statistical tests but not for machine learning models like those in Einstein Prediction Builder, which can handle non-normal data via tree-based or ensemble methods.

Practice this question →
44
MCQeasy

For a real-time AI application that requires low-latency access to customer interaction data, which storage solution is most appropriate?

A.Flat files on a network drive.
B.In-memory data store.
C.Relational database with complex joins.
D.Data lake with batch processing.
AnswerB

In-memory storage offers microsecond latency, ideal for real-time AI.

Why this answer

In-memory data stores (e.g., Redis, Memcached) store data in RAM rather than on disk, providing sub-millisecond read/write latencies essential for real-time AI applications that need immediate access to customer interaction data. This eliminates disk I/O bottlenecks and enables high-throughput, low-latency data retrieval for time-sensitive inference or decision-making.

Exam trap

Salesforce often tests the misconception that relational databases are always the best for structured data, but the trap here is that candidates overlook the strict latency requirement and choose a relational database (Option C) without considering that complex joins and disk-based storage make it too slow for real-time AI workloads.

How to eliminate wrong answers

Option A is wrong because flat files on a network drive introduce high latency due to network overhead and disk I/O, and they lack the indexing and concurrency control needed for real-time access. Option C is wrong because relational databases with complex joins incur significant query processing overhead and disk-based storage, making them unsuitable for low-latency requirements despite ACID compliance. Option D is wrong because data lakes with batch processing are designed for high-throughput, periodic analytics (e.g., hourly/daily) and cannot provide the sub-second response times required for real-time AI interactions.

Practice this question →
45
Multi-Selectmedium

Which TWO data preparation steps are required before using Einstein Discovery for sales forecasting? (Choose 2)

Select 2 answers
A.Convert all text fields to numeric using one-hot encoding
B.Remove duplicate records
C.Include a date or timestamp field for time series analysis
D.Ensure all predictor fields have no missing values
E.Normalize numeric fields to a 0-1 scale
AnswersC, D

For forecasting, a date field is needed to order records.

Why this answer

Einstein Discovery requires a date or timestamp field to perform time series analysis, which is essential for identifying trends, seasonality, and patterns in historical sales data. Without this field, the model cannot properly order observations or forecast future values based on temporal dependencies.

Exam trap

Salesforce often tests the misconception that manual data preprocessing steps like normalization or one-hot encoding are required, when in fact Einstein Discovery automates these steps, and the key prerequisite is ensuring a proper date/timestamp field exists for time-based analysis.

Practice this question →
46
Multi-Selectmedium

A company is implementing Einstein Prediction Builder to predict whether a support case will escalate. Which TWO data preparation steps should the admin take to improve model accuracy?

Select 2 answers
A.Include as many fields as possible to provide more context
B.Ensure missing values are handled appropriately (e.g., imputed or excluded)
C.Encrypt all fields containing personally identifiable information
D.Exclude cases that were closed without escalation
E.Remove fields that have a one-to-one relationship with the outcome
AnswersB, E

Missing values can bias the model; proper handling improves accuracy.

Why this answer

Correct: Removing redundant fields (like record IDs) and handling missing values are crucial for model accuracy. Option A is wrong because more fields can introduce noise. Option C is wrong because data encryption is about security, not accuracy.

Option D is wrong because all cases should be included to represent the full pattern.

Practice this question →
47
Multi-Selectmedium

Which TWO of the following are common dimensions of data quality that must be addressed for AI training?

Select 2 answers
A.Storage efficiency
B.Accuracy of values
C.Encryption strength
D.Consistency with external benchmarks
E.Completeness of records
AnswersB, E

Accuracy ensures data correctly represents real-world entities.

Why this answer

Accuracy of values (Option B) is a fundamental dimension of data quality because AI models learn patterns from training data; if the data contains incorrect values, the model will learn and propagate those errors, leading to unreliable predictions. For example, in a dataset of customer ages, a single erroneous entry of '200' can skew the model's understanding of age distributions, directly impacting model performance.

Exam trap

Salesforce often tests the distinction between data quality dimensions (accuracy, completeness, consistency) and operational or security attributes (storage efficiency, encryption strength), tricking candidates into selecting options that sound technical but are irrelevant to data quality for AI training.

Practice this question →
48
MCQhard

Refer to the exhibit. A developer runs this SOQL query to prepare data for Einstein Lead Scoring. The query returns an error. What is the most likely issue?

A.The alias 'TotalAmount' is not allowed in the HAVING clause.
B.The query misses a GROUP BY clause.
C.The SUM(Amount) cannot be used in the HAVING clause.
D.The WHERE clause condition is invalid.
AnswerA

In SOQL, HAVING must use the full aggregate expression, not an alias.

Why this answer

The HAVING clause references alias TotalAmount, but SOQL does not allow aliases in HAVING; the aggregated expression must be repeated.

Practice this question →
49
MCQmedium

A multinational corporation uses Salesforce AI to analyze customer feedback across multiple languages. They have 10,000 English reviews, 2,000 Spanish reviews, and 500 French reviews. The sentiment model performs well on English (F1=0.85) but poorly on French (F1=0.40). The data scientist wants to improve French sentiment performance without collecting new data. What should they do?

A.Translate all French reviews to English and train only on English data.
B.Use a multilingual pre-trained model without any additional French data.
C.Remove French data and use only English and Spanish to avoid imbalance.
D.Apply data augmentation to the French reviews using back-translation (translate to another language and back) to create more training examples.
AnswerD

Back-translation generates realistic paraphrases, augmenting the French dataset and improving model performance.

Why this answer

Data augmentation techniques like back-translation generate synthetic French samples, effectively increasing the minority language's representation and helping the model learn better.

Practice this question →
50
Multi-Selectmedium

Which THREE factors should be considered when selecting features for a predictive model in Salesforce?

Select 3 answers
A.Volume of data available for each feature
B.Correlation between features to avoid multicollinearity
C.Relevance of the feature to the target variable
D.Compliance with data privacy regulations
E.Business interpretability of the feature
AnswersB, C, E

Multicollinearity can harm model stability.

Why this answer

Multicollinearity occurs when two or more features are highly correlated, which can destabilize model coefficients and reduce interpretability. In Salesforce's predictive models, such as those built with Einstein Discovery, correlated features can inflate variance and lead to unreliable predictions. Avoiding multicollinearity ensures that the model's feature importance estimates are trustworthy and that the model generalizes well to new data.

Exam trap

Salesforce often tests the distinction between feature selection criteria (predictive power, correlation, interpretability) and broader data management concerns (privacy, volume), leading candidates to mistakenly include compliance or data volume as direct feature selection factors.

Practice this question →
51
MCQeasy

A company is building a chatbot using Einstein Bot's AI capabilities. They want to train intent recognition using historical chat transcripts. The transcripts contain many typos (e.g., 'hellp' instead of 'help') and slang (e.g., 'gonna' instead of 'going to'). The initial model performs poorly, misclassifying many intents. What data cleaning step is most important?

A.Use a spell-checker only for words that appear infrequently.
B.Keep the raw text as is because it reflects real user behavior.
C.Normalize text by applying spell-correction and replacing slang with standard terms.
D.Remove all messages that contain typos or slang to clean the dataset.
AnswerC

Normalization reduces noise and variability, enabling the model to focus on meaningful patterns.

Why this answer

Normalizing text by correcting common typos and expanding slang reduces vocabulary sparsity and helps the model learn consistent word associations, improving intent recognition.

Practice this question →
52
MCQeasy

A machine learning team is preparing a dataset for a supervised learning task. They have 100,000 labeled samples. Which data preparation step is essential before splitting into train/test sets?

A.Normalize all features to the same scale.
B.Remove all outliers from the dataset.
C.Shuffle the dataset randomly.
D.Visualize the data distribution for each feature.
AnswerC

Shuffling prevents biased splits.

Why this answer

Shuffling the dataset randomly before splitting into train/test sets ensures that the data distribution is similar across both subsets. Without shuffling, the split might inadvertently separate ordered or grouped data (e.g., time-series or batches), leading to biased model evaluation. This step is essential for supervised learning to prevent data leakage and ensure the test set is representative of the overall population.

Exam trap

Salesforce often tests the misconception that normalization or outlier removal must be done before splitting, but the trap here is that candidates overlook the fundamental need to randomize the data order to avoid temporal or structural bias in the train/test split.

How to eliminate wrong answers

Option A is wrong because normalizing features to the same scale is a preprocessing step typically applied after splitting the data, using statistics (e.g., mean and standard deviation) computed only from the training set to avoid data leakage into the test set. Option B is wrong because removing all outliers before splitting can introduce bias and reduce the dataset's representativeness; outlier handling should be done with care, often after splitting, and may be domain-specific. Option D is wrong because visualizing data distributions is an exploratory step that helps understand the data but is not essential before splitting; it can be performed after splitting to avoid influencing the split decisions.

Practice this question →
53
MCQeasy

Which method is most suitable for ingesting streaming data from IoT sensors into a data lake?

A.Copying data via FTP.
B.Batch ingestion every 24 hours.
C.Manual upload via web interface.
D.Real-time streaming with Apache Kafka.
AnswerD

Kafka provides high-throughput, fault-tolerant streaming for IoT data.

Why this answer

Apache Kafka is the most suitable option because it is a distributed streaming platform designed for high-throughput, fault-tolerant, real-time data ingestion. IoT sensors generate continuous, high-velocity data streams, and Kafka's publish-subscribe model allows data to be ingested into a data lake with low latency, ensuring near-real-time availability for analytics.

Exam trap

Salesforce often tests the distinction between batch and real-time processing, and the trap here is that candidates may choose batch ingestion (Option B) thinking it is simpler or sufficient, overlooking the fundamental requirement for low-latency streaming in IoT sensor data ingestion.

How to eliminate wrong answers

Option A is wrong because FTP (File Transfer Protocol) is a batch-oriented file transfer protocol that lacks real-time streaming capabilities, introduces latency, and does not handle continuous data streams from IoT sensors efficiently. Option B is wrong because batch ingestion every 24 hours introduces unacceptable latency for streaming IoT data, which often requires immediate processing for time-sensitive applications like anomaly detection or predictive maintenance. Option C is wrong because manual upload via a web interface is impractical for high-frequency sensor data, as it requires human intervention, cannot scale, and introduces significant delays and errors.

Practice this question →
54
MCQmedium

A data scientist notices that the model accuracy drops significantly after retraining with new data. Upon inspection, they find that many records have missing values for a key feature. Which data quality improvement should be prioritized first?

A.Implement imputation for missing feature values.
B.Normalize the feature range.
C.Reduce the number of features.
D.Remove duplicate records.
AnswerA

Imputation addresses missing data, a common cause of accuracy drop.

Why this answer

The core issue is that missing values in a key feature introduce noise and bias, directly degrading model performance. Imputation (option A) is the most direct and impactful first step because it preserves the dataset size and feature set, allowing the model to learn from complete patterns. Without addressing missing data first, other quality improvements like normalization or feature reduction would be applied to corrupted data, failing to resolve the root cause.

Exam trap

Salesforce often tests the misconception that data quality improvements like normalization or feature reduction are universal fixes, when in fact the most urgent step is always to handle missing data, as it directly undermines model training and inference.

How to eliminate wrong answers

Option B is wrong because normalizing the feature range (e.g., scaling to 0-1) does not address missing values; it only adjusts the distribution of existing values, leaving the model to train on incomplete records. Option C is wrong because reducing the number of features may discard the key feature entirely, which could be critical for prediction, and does not fix the missing data problem in the remaining features. Option D is wrong because removing duplicate records addresses redundancy, not missing values; duplicates are not the cause of the accuracy drop, and removing them could even reduce valuable training data.

Practice this question →
55
Multi-Selecthard

Which THREE of the following are best practices for feature engineering in Einstein Studio?

Select 3 answers
A.Remove all records with missing values
B.Apply normalization to numerical features
C.Use raw data directly without any transformation
D.Use domain knowledge to create derived features
E.Use one-hot encoding for categorical variables
AnswersB, D, E

Normalization ensures features are on a similar scale.

Why this answer

Options B, D, and E are correct. Normalization scales features, domain knowledge creates meaningful derived features, and one-hot encoding handles categorical variables. Option A is wrong because raw data often needs processing.

Option C is wrong because removing missing values can lose information; imputation is often better.

Practice this question →
56
Multi-Selecteasy

A company is ingesting data from multiple sources into Data Cloud for Einstein. Which THREE data preparation steps should be performed?

Select 3 answers
A.Normalization
B.Field mapping
C.Encryption
D.Data labeling
E.Deduplication
AnswersA, B, E

Ensures consistent data formats across sources.

Why this answer

Normalization is correct because Data Cloud requires data from multiple sources to be transformed into a consistent format, such as standardizing date formats, units, or naming conventions, to ensure the data can be unified and analyzed effectively. This step is critical for Einstein AI models to process data without inconsistencies that could skew predictions or insights.

Exam trap

Salesforce often tests the distinction between data preparation steps (normalization, field mapping, deduplication) and data security or ML-specific tasks (encryption, data labeling) to see if candidates confuse operational data engineering with security or model training processes.

Practice this question →
57
Multi-Selecteasy

A company is preparing customer data to train a custom AI model for sentiment analysis. Which two data preparation best practices should they follow? (Choose two.)

Select 2 answers
A.Use only data from the last month.
B.Ensure data is representative of all customer demographics.
C.Remove all records with missing values.
D.Label data manually by a single annotator.
E.Anonymize personally identifiable information (PII) before training.
AnswersB, E

Representative data prevents model bias and improves generalization across customer segments.

Why this answer

Ensuring representative data and anonymizing PII are critical for model fairness and privacy. Removing all records with missing values can discard useful information; using only recent data may introduce bias; single-annotator labeling can cause subjective bias.

Practice this question →
58
MCQhard

A data scientist notices that an Einstein model for predicting customer churn has unusually high accuracy on training data but performs poorly on validation data. Which data issue is the most likely cause?

A.The dataset has an imbalanced class distribution
B.The dataset contains many missing values
C.The model was trained on stale data from a different season
D.A field containing future information (e.g., 'churn_date') was included in features
AnswerD

Data leakage from a field that reveals the outcome causes overfitting and high train accuracy.

Why this answer

Including a field like 'churn_date' in the feature set introduces target leakage, where the model has access to information that would not be available at prediction time. This causes the model to appear highly accurate on training data (since it can directly 'see' the outcome) but fails to generalize to validation data where such future information is absent. In Salesforce Einstein, features must be strictly historical or static to avoid this data leakage issue.

Exam trap

Salesforce often tests the concept of data leakage by presenting it as a scenario where the model performs well on training data but poorly on validation data, and the trap is that candidates may confuse this with overfitting or class imbalance, rather than recognizing the inclusion of a future or target-related field as the root cause.

How to eliminate wrong answers

Option A is wrong because imbalanced class distribution typically causes the model to predict the majority class, leading to high accuracy on training data but poor performance on validation data only if the imbalance is extreme and not handled; however, the question describes 'unusually high accuracy' on training data, which is more characteristic of overfitting or leakage, not class imbalance. Option B is wrong because missing values generally degrade model performance across both training and validation sets, not causing a stark contrast between high training accuracy and low validation accuracy. Option C is wrong because stale data from a different season would cause poor performance on both training and validation data if the validation data is from the same season, or poor performance on validation data if it is from a different season, but it would not explain unusually high training accuracy.

Practice this question →
59
MCQmedium

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?

A.Add a binary feature indicating whether each quarter was during the pandemic.
B.Remove the data points corresponding to the pandemic year from the training set.
C.Normalize the entire dataset using Z-scores to reduce the impact of the outlier.
D.Include the outlier data and increase the model capacity to capture the anomaly.
AnswerB

Removing the outlier helps the model focus on typical patterns, improving generalization to future non-pandemic quarters.

Why this answer

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.

Exam trap

Salesforce often tests the misconception that you should keep all data and adjust the model (e.g., via normalization or capacity increase) rather than removing non-representative outliers, leading candidates to pick options like C or D.

How to eliminate wrong answers

Option A is wrong because adding a binary pandemic feature does not remove the outlier's influence; the model may still overfit to the anomalous drop and fail to generalize, as the feature only labels the outlier without correcting the skewed distribution. Option C is wrong because Z-score normalization scales the data but does not eliminate the outlier's impact on the model's learned weights; the extreme value still distorts the mean and variance, leading to biased forecasts. Option D is wrong because increasing model capacity to capture the anomaly encourages overfitting to the pandemic year's unique pattern, which will not recur, thus worsening generalization and maintaining the overestimation error.

Practice this question →
60
MCQeasy

Which data transformation is most appropriate for converting categorical variables into numerical format for a machine learning model?

A.Normalization.
B.One-hot encoding.
C.Principal component analysis.
D.Standardization.
AnswerB

One-hot encoding creates binary columns for each category, making them usable in models.

Why this answer

One-hot encoding is the correct transformation because it converts categorical variables into a binary vector representation, where each category becomes a separate column with a 1 or 0. This allows machine learning models to interpret categorical data without implying any ordinal relationship, which is essential for algorithms that rely on numerical distances or linear algebra.

Exam trap

Salesforce often tests the distinction between data preprocessing techniques (normalization, standardization) and encoding methods, trapping candidates who confuse scaling with categorical conversion.

How to eliminate wrong answers

Option A is wrong because normalization scales numerical features to a range (e.g., 0 to 1) and is used for continuous data, not for converting categorical variables into numbers. Option C is wrong because principal component analysis (PCA) is a dimensionality reduction technique that transforms existing numerical features into uncorrelated components, not a method for encoding categorical data. Option D is wrong because standardization centers data around a mean of 0 and standard deviation of 1, which is applied to numerical features and would not create meaningful representations for categorical variables.

Practice this question →
61
MCQhard

Refer to the exhibit. A data pipeline fails during the DataTransformation stage. What is the most likely root cause?

A.The pipeline has a network connectivity issue.
B.The data type for 'income' is incorrect.
C.A transformation step references the 'age' column, but it is not present in the input data.
D.The 'age' column contains null values.
AnswerC

The error clearly states 'age' column not found.

Why this answer

The error occurs during the DataTransformation stage, which processes data after it has been successfully ingested. If a transformation step references the 'age' column but that column is missing from the input data, the pipeline will fail with a column-not-found error. This is a common schema mismatch issue in data pipelines, distinct from connectivity or data quality problems.

Exam trap

Salesforce often tests the distinction between pipeline stages (ingestion vs. transformation) and the specific type of error (missing column vs. data quality issue) to see if candidates understand that a missing column causes an immediate failure, while nulls or type mismatches may be handled differently depending on the pipeline configuration.

How to eliminate wrong answers

Option A is wrong because a network connectivity issue would typically cause the pipeline to fail during the data ingestion or extraction stage, not during the DataTransformation stage. Option B is wrong because an incorrect data type for 'income' would cause a type conversion error, but the question specifically states the failure is during transformation, and the error would be related to type mismatch, not a missing column. Option D is wrong because null values in the 'age' column would not cause a pipeline failure during transformation unless the transformation logic explicitly fails on nulls; most pipelines handle nulls gracefully or can be configured to skip or impute them.

Practice this question →
62
MCQeasy

A fraud detection model is being trained on transaction data where only 1% of transactions are fraudulent. The current model predicts 'non-fraud' for all transactions, achieving 99% accuracy. Which technique should be applied to improve model performance?

A.Remove the minority class to have balanced data
B.Set a lower classification threshold for fraud
C.Add more features like transaction location
D.Oversample the minority class or undersample the majority class
AnswerD

Resampling techniques create a more balanced training set, improving recall for fraud.

Why this answer

Oversampling or undersampling addresses class imbalance, allowing the model to learn minority patterns. Using more features alone doesn't fix imbalance, setting a lower threshold may help but is less common than resampling, and removing minority class is counterproductive.

Practice this question →
63
MCQeasy

For an AI project, data must be stored in a way that supports both training and real-time inference. Which storage solution meets this requirement?

A.Data warehouse (e.g., Snowflake)
B.Relational database (e.g., PostgreSQL)
C.Data lake (e.g., Amazon S3 or Azure Data Lake)
D.In-memory cache (e.g., Redis)
AnswerC

Data lakes store raw and processed data for various purposes.

Why this answer

A data lake (e.g., Amazon S3 or Azure Data Lake) is the correct choice because it can store vast amounts of raw, unstructured, and structured data in its native format, making it ideal for training AI models on diverse datasets. At the same time, data lakes support real-time inference by enabling direct access to data via APIs or streaming services (e.g., AWS Lambda or Azure Functions) without the latency of transforming data into a schema-on-write structure. This dual capability—handling both batch processing for training and low-latency reads for inference—is a key requirement that other storage solutions cannot fulfill as effectively.

Exam trap

Salesforce often tests the misconception that a data warehouse or relational database is sufficient for AI workloads because candidates overlook the need for raw, unstructured data storage and the flexibility of schema-on-read, instead focusing only on structured query performance.

How to eliminate wrong answers

Option A is wrong because a data warehouse (e.g., Snowflake) is optimized for structured, aggregated data and analytical queries, not for storing raw, unstructured data needed for AI training, and its schema-on-write approach introduces latency unsuitable for real-time inference. Option B is wrong because a relational database (e.g., PostgreSQL) enforces strict schemas and ACID transactions, which limit the flexibility to store diverse data types (e.g., images, text) required for AI training, and its row-based storage is inefficient for high-throughput, low-latency inference workloads. Option D is wrong because an in-memory cache (e.g., Redis) is designed for ephemeral, high-speed data access but lacks persistent storage and the capacity to hold large-scale training datasets, making it unsuitable for long-term data storage required for AI model training.

Practice this question →
64
MCQhard

A data scientist is building a predictive model for customer churn using Salesforce data. The dataset has 20 features, and the target variable is highly imbalanced (5% churn, 95% non-churn). Which technique should be applied to handle the class imbalance before training?

A.Apply Principal Component Analysis (PCA) for dimensionality reduction
B.Create interaction features between existing variables
C.Use accuracy as the evaluation metric
D.Use Synthetic Minority Over-sampling Technique (SMOTE)
AnswerD

SMOTE creates synthetic examples of the minority class.

Why this answer

SMOTE (Synthetic Minority Over-sampling Technique) is the correct choice because it generates synthetic samples for the minority class (churn) by interpolating between existing minority instances, effectively balancing the dataset without simply duplicating data. This prevents the model from being biased toward the majority class (non-churn) and improves recall for the churn class, which is critical in imbalanced classification problems.

Exam trap

Salesforce often tests the misconception that any data preprocessing technique (like PCA or feature engineering) can fix class imbalance, when in fact only resampling methods (SMOTE, ADASYN) or cost-sensitive learning directly address the skewed target distribution.

How to eliminate wrong answers

Option A is wrong because PCA is a dimensionality reduction technique that does not address class imbalance; it reduces feature space but does not alter the distribution of the target variable. Option B is wrong because creating interaction features may capture non-linear relationships but does not solve the imbalance problem; it can even exacerbate overfitting if the minority class remains underrepresented. Option C is wrong because accuracy is a misleading metric for imbalanced datasets—a model predicting all non-churn would achieve 95% accuracy but fail to identify any churn cases; metrics like precision, recall, F1-score, or AUC-ROC are appropriate instead.

Practice this question →
65
MCQmedium

After applying a log transformation to a numeric feature, an Einstein model’s performance dropped significantly. What is the most likely cause?

A.The data volume was reduced by the transformation
B.The feature was normally distributed after transformation
C.The feature contained zero or negative values
D.The transformation introduced multicollinearity with other features
AnswerC

Log of non-positive values is undefined, causing missing or infinity values.

Why this answer

Log transformation is undefined for zero or negative values because log(0) is negative infinity and log of a negative number is not a real number. In Salesforce Einstein, numeric features with such invalid transformed values can cause the model to fail or produce erratic results, leading to a significant drop in performance. This is the most likely cause given the symptom described.

Exam trap

Salesforce often tests the misconception that log transformation always improves model performance, but the trap here is that candidates overlook the mathematical constraint that log is undefined for non-positive values, causing them to choose a less relevant option like data volume reduction or multicollinearity.

How to eliminate wrong answers

Option A is wrong because log transformation does not reduce data volume; it merely applies a mathematical function to each value, preserving the number of records. Option B is wrong because making a feature normally distributed is typically beneficial for many models, not detrimental; a normal distribution after transformation would likely improve, not degrade, performance. Option D is wrong because log transformation is applied to a single feature and does not introduce multicollinearity, which is a relationship between two or more independent variables; it cannot create collinearity with other features on its own.

Practice this question →
66
MCQhard

A company is using customer support tickets to train a model for auto-classifying issues. The dataset includes fields like 'Case Title', 'Description', 'Product', and 'Customer Name'. Which privacy concern is most critical to address before training?

A.Anonymize personal identifiable information (PII) in Description and Title
B.Encrypt session tokens used in the support system
C.Remove all case numbers to prevent data leakage
D.Ensure all customers have opted in before using their data
AnswerA

PII in text must be removed to comply with privacy regulations and prevent bias.

Why this answer

Anonymizing PII in the text fields is critical to avoid exposing customer information in model artifacts or predictions. Session tokens are irrelevant, and case numbers are not PII. Opt-in is a legal requirement but not directly about data preparation for AI.

Practice this question →
67
MCQeasy

Refer to the exhibit. A data analyst runs a profile on a dataset and sees these statistics. Based on best practices, which action should be taken first?

A.Impute the 500 missing values with the mean
B.Remove the 200 duplicate records
C.Remove the 50 outliers in the Amount field
D.Skip all preprocessing and train the model directly
AnswerB

Duplicates can artificially inflate certain patterns and cause data leakage.

Why this answer

Duplicate records introduce bias and redundancy, leading to overfitting or skewed model performance. Removing duplicates is a standard first step in data preprocessing to ensure data integrity before handling missing values or outliers. In the context of the AI Associate exam, best practices prioritize deduplication early in the data cleaning pipeline.

Exam trap

Salesforce often tests the order of preprocessing steps, trapping candidates who jump to imputation or outlier removal without first cleaning duplicates, which is the foundational step in data preparation.

How to eliminate wrong answers

Option A is wrong because imputing missing values with the mean should only be considered after duplicates are removed, as duplicates can inflate the mean and distort imputation. Option C is wrong because removing outliers should be done after addressing duplicates and missing values, and only after understanding the domain context; premature outlier removal can discard legitimate data. Option D is wrong because skipping all preprocessing ignores fundamental data quality issues (missing values, duplicates, outliers) that degrade model accuracy and reliability, violating best practices for AI workflows.

Practice this question →
68
Multi-Selecteasy

A data scientist is preparing numeric features for a regression model. Which TWO transformations are commonly applied to improve model performance?

Select 2 answers
A.Normalize to a 0-1 range
B.Remove outliers beyond 3 standard deviations
C.Convert numbers to string labels
D.Apply one-hot encoding
E.Standardize to mean 0 and variance 1
AnswersA, E

Scales features to a common range, helpful for distance-based models.

Why this answer

Normalizing features to a 0-1 range (min-max scaling) ensures that all numeric features contribute equally to the model, preventing features with larger magnitudes from dominating the gradient descent optimization. This is especially important for distance-based algorithms like k-nearest neighbors or neural networks, where feature scale directly impacts convergence speed and model accuracy.

Exam trap

Salesforce often tests the distinction between data cleaning (e.g., outlier removal) and feature transformation (e.g., scaling), leading candidates to mistakenly select outlier removal as a transformation that improves model performance.

Practice this question →
69
MCQmedium

An admin created a data stream to bring external customer data into Data Cloud for Einstein. The data stream fails with error 'Schema mismatch: expected 10 fields, got 8'. What is the likely cause?

A.The data flow has a filter that drops fields.
B.The target object has validation rules.
C.The source file has extra columns.
D.The data stream definition expects more fields than the source provides.
AnswerD

Directly matches the error: expected 10, got 8.

Why this answer

The error 'Schema mismatch: expected 10 fields, got 8' indicates that the data stream definition in Data Cloud is configured to map 10 fields from the source, but the actual source file or API response only provides 8 fields. This mismatch occurs when the schema defined in the data stream does not match the source schema, typically because the source has fewer columns than expected. Option D correctly identifies this as the likely cause.

Exam trap

Salesforce often tests the distinction between schema-level errors (field count mismatch) and data-level errors (validation rules, filters), leading candidates to confuse data flow operations with data stream schema definitions.

How to eliminate wrong answers

Option A is wrong because a filter in a data flow drops rows (records), not fields (columns), and the error explicitly mentions a field count mismatch, not a row count issue. Option B is wrong because validation rules on the target object would cause record-level failures during data insertion, not a schema mismatch error during the data stream definition or ingestion phase. Option C is wrong because extra columns in the source file would cause the error to report more fields than expected (e.g., 'expected 10, got 12'), not fewer.

Practice this question →
70
MCQmedium

During data transformation, a data scientist applies one-hot encoding to a categorical feature with 50 unique values. The resulting dataset has 50 new columns. What is a potential drawback of this transformation?

A.Reduction in training time
B.Increased interpretability of the model
C.High cardinality leading to sparse data and overfitting
D.Loss of ordinal information in categories
AnswerC

High cardinality creates many sparse columns, risking overfitting.

Why this answer

One-hot encoding a categorical feature with 50 unique values creates 50 binary columns, each representing one category. This high cardinality leads to a very sparse matrix (most entries are 0), which can cause the model to overfit by learning noise from rare categories, especially when the dataset is not large enough to support such dimensionality.

Exam trap

Salesforce often tests the misconception that one-hot encoding always improves model performance by preserving all information, when in fact high cardinality introduces sparsity and overfitting risks that can degrade model accuracy.

How to eliminate wrong answers

Option A is wrong because one-hot encoding increases the number of features, which typically increases training time due to higher dimensionality, not reduces it. Option B is wrong because adding 50 new binary columns reduces interpretability; the model becomes more complex and harder to explain, especially with many dummy variables. Option D is wrong because one-hot encoding is designed for nominal (unordered) categories; ordinal information is not lost because it was never present — the feature had no inherent order, so no information is lost.

Practice this question →
71
Multi-Selectmedium

Which TWO data sources can be used with Einstein Prediction Builder?

Select 2 answers
A.Files uploaded to Salesforce Files.
B.Data Cloud objects using the harmonized data model.
C.Standard Salesforce objects like Account and Opportunity.
D.Chatter feed posts.
E.Dashboard and report snapshots.
AnswersB, C

Data Cloud objects are supported.

Why this answer

Einstein Prediction Builder requires structured data that can be mapped to a prediction objective. Data Cloud objects using the harmonized data model provide a unified, standardized schema that Prediction Builder can consume directly, enabling predictions across multiple Salesforce and external data sources. Standard Salesforce objects like Account and Opportunity are also supported because they contain the fields and relationships needed to train predictive models.

Exam trap

Salesforce often tests the misconception that any data in Salesforce (like files or Chatter posts) can be used directly with Einstein Prediction Builder, when in fact only structured, field-level data from objects or harmonized Data Cloud objects is supported.

Practice this question →
72
MCQmedium

A company uses Einstein Prediction Builder to predict customer churn. The model's accuracy is low. The admin reviews the training data and notices that only 2% of records are churned. What should the admin do to improve the model?

A.Remove the churned records.
B.Increase the amount of training data.
C.Use oversampling techniques.
D.Change the prediction field.
AnswerC

Oversampling balances the classes and improves model sensitivity.

Why this answer

When a dataset has severe class imbalance (only 2% churned records), the model becomes biased toward predicting the majority class (non-churned), leading to low accuracy despite high apparent performance. Oversampling techniques, such as SMOTE or random oversampling, artificially increase the number of churned records in the training set to balance the classes, allowing Einstein Prediction Builder to learn patterns for the minority class more effectively.

Exam trap

Salesforce often tests the misconception that adding more data always improves model performance, but here the trap is that candidates overlook class imbalance and choose 'Increase the amount of training data' (Option B) without realizing that more data with the same imbalance does not solve the problem.

How to eliminate wrong answers

Option A is wrong because removing the churned records would eliminate the minority class entirely, making it impossible for the model to learn to predict churn, and would result in a model that always predicts non-churn. Option B is wrong because simply increasing the amount of training data without addressing the class imbalance will likely maintain the same 2% churn ratio, providing more majority-class examples but not improving minority-class learning. Option D is wrong because changing the prediction field would alter the target variable itself, which does not fix the underlying class imbalance issue and would require redefining the business problem.

Practice this question →
73
MCQmedium

You are an admin at a financial services firm. The firm wants to use Einstein Next Best Action to offer personalized product recommendations to customers on its service portal. The data includes customer profiles, transaction history, and support case history. The Einstein Next Best Action strategy is configured with a recommendation that shows a 'Savings Account' offer to customers who have a checking account. However, the recommendation is not appearing for any customers. You check the Data Flow and see that the 'Account' object data is flowing correctly. The recommendation's filter condition is: AND( Has_Checking_Account__c = true, Age__c > 18 ). You verify that many customers meet these conditions. What is the most likely reason the recommendation is not appearing?

A.The 'Account' object is not supported by Einstein Next Best Action
B.The recommendation is not activated or published
C.The customer data is not being refreshed in real time
D.The filter condition is syntactically incorrect
AnswerB

Recommendations must be activated and published to be served to customers.

Why this answer

The most likely reason the recommendation is not appearing is that it has not been activated or published. In Einstein Next Best Action, recommendations must be explicitly activated or published to become available for serving to customers; configuration alone does not make them live. Since the data flow is correct and the filter conditions are valid, the missing activation step is the typical cause of a recommendation not showing.

Exam trap

The trap here is that candidates may focus on data flow or filter syntax issues, but the real test is understanding that activation is a required step in Einstein Next Best Action to make recommendations live.

How to eliminate wrong answers

Option A is wrong because the 'Account' object is fully supported by Einstein Next Best Action, as it is a standard Salesforce object that can be used in recommendation strategies. Option C is wrong because Einstein Next Best Action does not require real-time data refresh; it works with batch-synced data, and the Data Flow showing correct data indicates the data is available. Option D is wrong because the filter condition AND( Has_Checking_Account__c = true, Age__c > 18 ) is syntactically correct in Salesforce formula syntax and would not cause the recommendation to fail silently.

Practice this question →
74
Multi-Selecteasy

Which TWO considerations are important when labeling data for a supervised learning model?

Select 2 answers
A.Maintaining consistent guidelines.
B.Labeler expertise.
C.Using automated labeling for all tasks.
D.Ignoring inter-labeler agreement.
E.Labeling only a small sample.
AnswersA, B

Clear guidelines ensure labelers apply the same criteria, reducing variability.

Why this answer

Maintaining consistent guidelines (A) is critical because supervised learning models learn patterns from labeled data; inconsistent labels introduce noise and confuse the model, degrading its accuracy. Labeler expertise (B) ensures that domain-specific nuances are correctly captured, which is especially important for tasks like medical imaging or legal document classification where errors have high cost.

Exam trap

Salesforce often tests the misconception that automated labeling is a complete substitute for human labeling, when in reality it requires careful validation and is typically used to augment, not replace, human effort.

Practice this question →
75
MCQmedium

A data scientist is preparing data for Einstein Discovery. The dataset has 10,000 records with 5 predictors and one outcome. The outcome is binary (1/0). What is the minimum number of positive outcomes typically required for a reliable model?

A.250
B.500
C.100
D.50
AnswerA

50 per predictor * 5 predictors = 250 positive outcomes.

Why this answer

For binary classification with 10,000 records and 5 predictors, a common rule of thumb in predictive modeling (including Einstein Discovery) is to have at least 10 events per predictor variable (EPV). With 5 predictors, you need at least 50 positive outcomes, but to ensure model stability and reliable training, a minimum of 250 positive outcomes (5% of 10,000) is typically required. This aligns with best practices for avoiding overfitting and achieving adequate statistical power.

Exam trap

Salesforce often tests the 10 events per predictor variable (EPV) rule, but the trap here is that candidates mistakenly apply the EPV rule directly (50 for 5 predictors) without considering the additional requirement for a minimum of 250 positive outcomes to ensure model reliability in Einstein Discovery.

How to eliminate wrong answers

Option B (500) is wrong because it overestimates the minimum requirement; while 500 positive outcomes would certainly be sufficient, the question asks for the minimum typically required, which is lower at 250. Option C (100) is wrong because it underestimates the requirement; with 5 predictors, 100 positive outcomes would only provide 20 events per predictor, which is below the recommended 10 EPV rule for reliable models. Option D (50) is wrong because it represents the bare minimum for 5 predictors under the 10 EPV rule, but in practice, Einstein Discovery and general best practices require a higher minimum (often 250 or 5% of records) to ensure model convergence and avoid instability.

Practice this question →
Page 1 of 2 · 113 questions totalNext →

Ready to test yourself?

Try a timed practice session using only Data For Ai questions.

Start 20-question session