AI Associate Advanced Practice Exam: Hard Questions 2025

A healthcare provider wants to deploy generative AI summaries for Service Cloud cases. Agents will paste patient-provided text into prompts. Compliance requires minimizing the chance of exposing protected health information (PHI) beyond what is needed, ensuring agents are warned when prompts contain sensitive data, and maintaining an auditable trail of how AI outputs were generated. Which approach best balances responsible AI, governance, and operational practicality?

A retail bank is piloting AI-assisted email drafting for customer service. The pilot shows higher resolution speed but a compliance review finds the AI occasionally makes definitive statements about fees that depend on account terms. The bank wants to reduce regulatory risk without losing productivity. What is the most effective remediation strategy?

A company wants Einstein Copilot to answer questions like, "Which open opportunities in my territory are most likely to slip this month and why?" The response must reference current pipeline fields, recent activity, and any risk signals captured in Salesforce. Users also want the copilot to optionally execute an action to create follow-up tasks. Which architecture best supports accurate, actionable responses inside Salesforce?

A service organization deploys generative AI to draft knowledge articles from closed cases. After rollout, authors report that drafts frequently include details from unrelated cases that the author cannot access. Investigation shows the drafting feature was configured to use a broad corpus of case data. What is the best corrective action?

A sales team uses an AI feature to summarize call notes stored as Activities. Many summaries omit critical next steps because those are often recorded in a custom object related to the Opportunity rather than in the Activity itself. The team wants more complete summaries without requiring reps to duplicate data entry. What is the best solution approach?

A company enables AI-generated replies in Service Cloud. Agents complain that the copilot sometimes suggests responses that conflict with the latest policy updates posted in a new Policy object. The policy records are added daily and must be reflected immediately. What is the most effective troubleshooting and remediation path?

A marketing team wants to build an AI model to predict churn risk for subscription customers. They discover that key fields like 'Reason for Cancellation' and 'Competitor Mentioned' are often blank or inconsistently populated across business units. Leadership wants quick results but also wants the model to be reliable and explainable. What is the best next step?

A global company wants to use AI to recommend next-best actions based on customer interactions across multiple Salesforce orgs and an external commerce platform. Data is duplicated and customer identifiers differ by system, causing the AI to recommend actions for the wrong person in some edge cases. What is the most critical data remediation to address first?

An admin is asked to make AI-generated case summaries available to agents, but only after the summaries are reviewed and approved by a supervisor. Additionally, if the case is reassigned, the system must prevent previously generated summaries from being shown to unauthorized users. Which implementation best meets these requirements?

A company deploys a generative AI assistant to draft proposals using CRM data and a library of approved clauses. Users report that the assistant is accurate but slow, and sometimes times out when proposals include many products and regional terms. The business wants to improve responsiveness while keeping content grounded and compliant. What is the best architectural adjustment?