Salesforce Certified AI Associate – Complete Winter '26 Guide

Predictive AI vs Generative AI

Predictive AI uses historical data to forecast outcomes — Einstein Opportunity Scoring, Next Best Action, and Lead Scoring are predictive. Generative AI creates new content from prompts — Einstein Copilot, Sales Email generation, and Case Summaries are generative. The exam tests which type applies to a given use case.

Einstein Trust Layer

The Einstein Trust Layer is Salesforce's data security framework for AI. It automatically masks personally identifiable information (PII) before prompts reach an external LLM, enforces zero data retention by model providers, and provides audit logs of all AI interactions. Know what it does and why it matters for customer data protection.

Salesforce's Trusted AI Principles

The five principles are: Accuracy (outputs should be correct and based on data), Safety (prevent harm), Honesty (be transparent about being AI), Empowerment (humans remain in control), and Inclusivity (AI should not discriminate). Exam questions describe a scenario and ask which principle applies.

Grounding and Hallucinations

Grounding connects AI prompts to real CRM data so responses reflect actual customer records. Without grounding, generative AI may hallucinate — producing confident but incorrect outputs. Prompt templates in Salesforce merge static instructions with live CRM data to reduce hallucinations. Know how grounding mitigates this risk.

Data Quality and AI Bias

AI models are trained on historical data — poor data quality (duplicates, missing values, outdated records) leads to inaccurate predictions. Biased training data produces biased outputs. Responsible AI requires auditing training data, testing for disparate impact across demographics, and ongoing monitoring after deployment. Data governance is a prerequisite for trustworthy AI.

Salesforce's Trusted AI Principles

Know the five principles: Responsible, Accountable, Transparent, Empowering, and Inclusive. Questions about AI ethics and governance will reference these principles — understand what each means in practice.

AI Terminology at a Glance

Distinguish between AI, Machine Learning, Deep Learning, and Generative AI. Know what training data, a model, inference, and a prompt mean without needing mathematical depth.

Einstein Features & Use Cases

Know Einstein Lead Scoring, Opportunity Insights, Einstein Bots, Einstein Search, Einstein Next Best Action, and Einstein Copilot at a feature level — what each does and when to recommend it to a business.

Bias, Fairness & Model Drift

Understand how training data bias affects model outputs, what model drift means over time, and why human oversight and regular model retraining matter for responsible AI deployment.

Generative AI & Grounding

Know what large language models (LLMs) do, what hallucination means and why it is a risk, and how Salesforce Einstein Copilot uses grounding with CRM data to produce more accurate, contextual responses.