Glossary: Salesforce Certified AI Associate

*Note: All terms notated with an asterisk (*) are key to the exam and reviewing the
linked documentation is recommended.

AI Fundamentals (17%):

● Artificial Intelligence: The ability of machines to exhibit human-like capabilities

such as reasoning, learning, planning, and creativity.
● Models: Mathematical representations inside an AI system that determine how it
makes predictions or decisions based on data.
● Neural Networks: AI systems modeled on the neuron connections in the human
brain, comprised of layers of interconnected nodes.
● Node: An interconnected point inside a neural network that transmits signals to
other nodes.
● Weight: Adjustable parameters in a neural network that determine the strength of
connections between nodes.
● Layers: Levels within a neural network where nodes are organized, with data
flowing from input layers to hidden layers to output layers.
● Inputs: Data fed into an AI system that is used for making predictions or
decisions.
● Deep Learning: A type of machine learning that uses neural networks, especially
those with many hidden layers, to identify complex patterns and relationships in
data.
● Natural Language Understanding (NLU): The ability of a computer to
comprehend human speech or text.
● *Natural Language Processing (NLP): The ability of a computer to analyze and
process human language.
● Deep Learning: A type of machine learning that uses neural networks, especially
those with many hidden layers, to identify complex patterns and relationships in
data.
● *Machine Learning: The use of statistical models and algorithms that learn from
data to make predictions without explicit programming.
● *AI Benefits for Support Agents, Managers, and Customers: Increases efficiency,
provides insights, and reduces service representative strain.
● *Predictive AI: Analyzes historical data to predict potential future outcomes.
● *Generative AI: Creates new content like text, images or music based on
patterns in training data.
● Main AI capabilities
 ○ Numeric predictions: Predicting numerical outcomes like sales forecasts or
manufacturing defects.
○ Classifications: Categorizing data based on similarities, like detecting
spam emails or identifying objects in images.
○ Robotic Navigation: Adapting movement and actions based on
environmental feedback, like self-driving cars.
○ Language Processing: Understanding and generating human language,
like chatbots or text summarization.
● Concerns of AI
○ Hallucinations: When AI predictions diverge significantly from grounded
truth.
○ Data security: Protecting training data and maintaining robust models
resistant to attacks and potentially exposing data end-users provide while
using AI models.
○ Plagiarism: Potential for AI systems to replicate copyrighted material from
training data.
○ Sustainability: Environmental impact of computationally intensive AI model
training.
○ Unknown attribution: Output that lacks clarity on its original source due to
an unclear path to what influenced the created output.

AI Capabilities in CRM (8%):

● *Einstein Discovery: The Salesforce AI capability that generates predictive

models and insights from data without needing a data scientist. It helps discover
relevant patterns in all your data.
● *Einstein Prediction: Predict business outcomes, such as churn or lifetime value.
Create custom AI models on any Salesforce field or object with clicks, not code.
● *Einstein Next Best Action: A Salesforce tool that leverages predictive
intelligence to provide personalized, timely recommendations within Salesforce to
guide optimal next actions. It combines predictive models, tangible
recommendations, and automation to turn insights into action.
● *Einstein Language: A Natural Language Processing capability within the
Salesforce Platform that allows extracting insights from unstructured text data
across Salesforce objects. It can understand context and intent within text in
various languages.
● *Einstein Vision: A set of Einstein Platform services that automate image
classification for field service use cases. It can classify images uploaded in
Salesforce based on customized categories. Useful for field agents capturing
images and videos onsite for automatic logging and routing.
 ● *Einstein Computer Vision: The subset of Einstein AI capabilities that employs
deep learning algorithms to identify, categorize, caption, and understand visual
elements like objects, scenes, and activities in images and videos. Enables
automating visual perception tasks.
● *Call Summaries: A Service Cloud Einstein feature that automatically creates
summarized notes from call transcripts between service agents and customers.
Uses speech-to-text, natural language processing, and summarization to capture
key discussion points, requests, and action items.
● *Einstein Lead Scoring: A Sales Cloud Einstein capability that automatically
scores leads based on predictive analysis of historical lead and opportunity data
to determine propensity to convert. Provides visibility into factors influencing the
score to focus sales efforts on hot leads.
● *Einstein Opportunity Scoring Forecasting: Sales Cloud Einstein feature that
automatically scores opportunities to indicate the likelihood of closing a deal
based on predictive analysis of historical opportunity data and machine learning
algorithms. Provides focus on highest potential deals.
● *Einstein Opportunity Forecasting: A Sales Cloud Einstein capability that
predicts the most likely sales revenue outcome based on historical opportunity
data and current pipeline trends. Creates data-driven forecasts to complement
manager predictions.
● *Einstein Case Classification: A Service Cloud Einstein feature that combines
natural language processing, predictive algorithms, and pre-defined routing rules
to automatically categorize cases for faster resolution. Reduces manual case
classification effort.
● *Chatbots: Automated applications that simulate human-like conversational
interactions to respond to routine customer queries and requests, often using
capabilities like natural language processing. Salesforce offers Einstein Bots,
AI-powered chatbots natively integrated with CRM data.

Ethical Considerations of AI (39%):

● *Salesforce’s Trusted AI Principles: Guidelines established by Salesforce

focused on developing responsible and ethical AI, consisting of: Responsible -
Prioritize human rights, protect data, maintain scientific rigor, prevent misuse;
Accountable - Uphold responsibility towards customers, partners, society; seek
continual feedback and improvement; Transparent - Ensure explainability and
clarity behind AI recommendations and processes; Empowering - Augment
human intelligence, don't fully replace it; Inclusive - Ensure AI benefits people
inclusively, reflecting diverse values.
● *5 Guidelines for Responsible AI Development: Accuracy - Deliver verifiable,
transparent results with explanations; Safety - Reduce bias, toxicity, and misuse;
 protect privacy and security; Honesty - Respect data provenance; clarify when
content is AI-generated; Empowerment - Enhance human abilities; don't fully
automate where human judgment is needed; Sustainability - Balance innovation
pace and energy usage.
● *Ethical AI Practice Maturity Model: Ad Hoc - Informal advocacy for ethical
practices; Organized & Repeatable - Formal programs and training established;
Managed & Sustainable - Ethical checkpoints integrated across product lifecycle;
Optimized & Innovative - Mature ethical practices fully integrated with privacy,
legal, design.
● *Biases- What Are They: Systematic errors or unfair assumptions in data or
models that lead to discriminatory or prejudicial outcomes.
● *Biases - Why You Don’t Want Them: Biased models output skewed, inaccurate
predictions and unfair decisions; Biased AI systems exclude or disadvantage
certain groups; Biased algorithms undermine trust in AI.
● *Biases - How to Avoid: Have clean, high-quality data without embedded societal
biases; Validate data by checking for sampling bias and underrepresentation;
Don't use parameters that could introduce unintentional bias like gender,
ethnicity, ZIP code (demographic data); Continuously monitor models for
developing biases and adjust accordingly.

Data for AI (36%):

● *Data Quality: Having accurate, complete, consistent and relevant data. This
comprises dimensions like:
● *Data Dimensions:
○ Age: When data was last updated.
○ Completeness: Whether all expected data is present.
○ Accuracy: Extent to which data is correct.
○ Consistency: Standardized formatting of data.
○ Duplication: Presence of duplicate records.
○ Usage: How effectively data is being leveraged.
● How to Make Data Good and How to Clean Data:
○ *Remove Duplicates: Delete any duplicated data records.
○ *Fix Incorrect Data: Fix Incorrect Data: Correct any erroneous data values
like typos.
○ *Fill in missing data: Populate blank data fields where possible.
○ Salesforce Tools to help
■ *Validation Rules: Ensure data adheres to specified formats and
values.
■ Duplication Rules: Prevent duplicate record creation.
■ Required fields: Make key fields mandatory for more complete data.
 ■ Third-party app exchange apps can also help:
● Data Management Strategies
○ Naming: Establish conventions for object, field, and record naming.
○ Formatting: Define consistent data formats.
○ Workflow: Set processes for data lifecycle and maintenance.
○ Quality: Institute standards around accuracy and completeness.
○ Privacy/Security and Permissions: Determine appropriate data access
levels to prevent data breaches and only the correct users access to data.
○ Monitoring: Perform periodic data reviews and audits.
● Training Data: The historical, labeled datasets used to train machine learning
models to find patterns and make predictions. Quality training data is essential
for developing effective AI.