Select Two (Multi-Select) Questions

Uniqueness

Why wrong: Important for deduplication but not as critical for model accuracy.

Consistency

Why wrong: Important but less critical than completeness and accuracy.

Timeliness

Why wrong: Timeliness matters for relevance, but not directly for accuracy.

Remove all records with missing values in any field.

Why wrong: Einstein Discovery can handle missing values.

Select exactly 10 predictor fields manually.

Why wrong: Einstein can automatically select features.

Create a separate dataset for training and validation.

Why wrong: Einstein automatically splits data.

Include as many fields as possible to provide more context

Why wrong: Including irrelevant fields can reduce model accuracy.

Encrypt all fields containing personally identifiable information

Why wrong: Encryption is a security measure, not a data preparation step for model accuracy.

Exclude cases that were closed without escalation

Why wrong: Excluding non-escalated cases would remove the negative examples needed for training.

It is only available for lead scoring models.

Why wrong: Prediction Builder can predict outcomes on any object, not just leads.

It only supports predictions on the Opportunity object.

Why wrong: Prediction Builder supports multiple objects, including custom ones.

The prediction model automatically retrains every 24 hours.

Why wrong: Training frequency is adjustable, not fixed to 24 hours.

Sentiment analysis of customer emails

Why wrong: Sentiment analysis is typically in Einstein Service, not a core GPT for Sales feature.

Automated lead enrichment

Why wrong: Lead enrichment is a data processing task, not generative AI.

Generating Apex code

Why wrong: Apex code generation is a developer-focused feature, not sales-specific.

Use AI to replace human jobs entirely.

Why wrong: AI should augment, not replace, humans.

Ensure models achieve 100% accuracy before deployment.

Why wrong: Perfection is unrealistic; continuous improvement is key.

Fully automate decision-making without human review.

Why wrong: Ethical AI requires human oversight.

Create a custom scoring model specific to the business.

Why wrong: Standard model is used by default; custom is optional.

Deactivate any existing scoring rules.

Why wrong: Scoring rules can coexist with Einstein Opportunity Scoring.

Assign the Einstein Opportunity Scoring permission set to users.

Why wrong: This is needed for viewing scores, not for activation.

Increasing dataset size by adding noise.

Why wrong: Adding noise without care can degrade data quality.

Normalizing all features.

Why wrong: Normalization is important for some models but not always critical for data quality.

Using only labeled data.

Why wrong: This depends on the problem; unlabeled data can be useful for unsupervised learning or semi-supervised approaches.

Apply regularization.

Why wrong: Regularization helps with overfitting but is not specific to pair correlation.

Include both to capture more information.

Why wrong: High correlation can lead to unstable coefficient estimates.

Increase the sample size.

Why wrong: More data does not fix multicollinearity.

Remove outliers beyond 3 standard deviations

Why wrong: Outliers may contain signal; removal should be justified, not automatic.

Convert numbers to string labels

Why wrong: Loses order and magnitude, degrading performance.

Apply one-hot encoding

Why wrong: One-hot encoding is for categorical features, not numeric.

User profile data

Why wrong: User profiles can enhance recommendations but are not required.

Sales reports

Why wrong: Sales reports are aggregated and not directly used in recommendation training.

External product prices

Why wrong: Prices are not required for collaborative filtering or content-based filtering.

Data compression

Why wrong: Compression saves storage but does not address privacy.

Color coding fields in the dataset

Why wrong: Color coding is a UI feature, not a privacy control.

Indexing speed for real-time recommendations

Why wrong: Indexing speed is a performance concern, not privacy.

Label all available data immediately

Why wrong: Labeling all data can be resource-heavy and may reduce quality.

Label only training data

Why wrong: Validation and test data also need labeling.

Use automated labeling tools exclusively

Why wrong: Automated tools may not catch all nuances; human oversight is needed.

Uniqueness

Why wrong: While important for certain contexts, uniqueness is not a core quality dimension typically required for all AI models.

Volume

Why wrong: Volume refers to quantity, not quality; a quality dimension evaluates characteristics.

Increase data stream ingestion frequency

Why wrong: Does not address existing quality issues.

Implement data retention policies

Why wrong: Retention policies manage how long data is kept, not quality.

Use Calculated Insights to detect anomalies

Why wrong: Anomaly detection identifies issues but does not resolve them.

Remove all records with missing values

Why wrong: May discard valuable data; consider imputation instead.

Use raw data directly without any transformation

Why wrong: Raw data may not be suitable for models due to scale and encoding.

Multicollinearity

Why wrong: Multicollinearity affects regression coefficients, not data quality.

Data volume

Why wrong: Volume is about quantity, not quality.

High dimensionality

Why wrong: Dimensionality is a modeling concern, not a data quality issue.

Store data indefinitely for future use

Why wrong: Indefinite storage violates data minimization principles.

Encrypt data in transit and at rest

Why wrong: Encryption is a security measure, not a direct privacy consideration.

Use only data from the last month.

Why wrong: Using only recent data may miss long-term patterns and reduce sample size, harming model robustness.

Remove all records with missing values.

Why wrong: Removing all missing values may discard valuable data; imputation or handling missingness is preferred.

Label data manually by a single annotator.

Why wrong: Single-annotator labeling introduces individual bias; best practice is multiple annotators with consensus.

Accuracy

Why wrong: Not a core ethical principle.

Profitability

Why wrong: Not an ethical principle.