How to Audit AI for Adverse Impact in Hiring and Beyond
Which TWO practices are recommended when using AI for automated decision-making in hiring?
Quick Answer
The answer is to incorporate human review for high-stakes decisions and to conduct regular auditing for adverse impact. These practices are essential because AI models in hiring can perpetuate historical biases through statistical disparities, and auditing using methods like the four-fifths rule—which flags selection rates for protected groups below 80% of the most-selected group—helps detect such adverse impact before it harms candidates. On the Salesforce AI Associate exam, this concept tests your understanding of ethical AI governance under the “Responsible AI” domain, often appearing as a scenario where an automated system screens resumes and you must choose oversight mechanisms over purely technical fixes. A common trap is selecting “remove all human review for efficiency” or “only audit after complaints,” but the exam emphasizes proactive, continuous auditing paired with human judgment for high-risk decisions. Memory tip: think “Audit and Appeal”—always audit for adverse impact and keep a human in the loop for final calls.
⚠ Common exam trap
Salesforce often tests the misconception that model accuracy alone justifies automated decisions, tempting candidates to pick 'Ignore adverse impact if the model is accurate' (Option E) without recognizing that fairness and ethical compliance are separate, non-negotiable requirements.
Answer choices
Why each option matters
Answer the question above first, then reveal the full breakdown to understand why each option is right or wrong.
Correct answer & explanation
✓
Regularly audit the model for adverse impact.
Regular auditing for adverse impact is a core ethical practice to detect and mitigate bias in AI-driven hiring systems. Audits involve statistical analysis (e.g., the four-fifths rule from the Uniform Guidelines on Employee Selection Procedures) to compare selection rates across protected groups, ensuring the model does not disproportionately disadvantage certain demographics.
Answer analysis
Option-by-option breakdown
For each option: why learners choose it and why it is or isn't the right answer here.
- ✗
Use the AI model as the sole decision-maker.
Why it's wrong here
Sole reliance on AI can be unfair.
- ✓
Regularly audit the model for adverse impact.
Why this is correct
Auditing detects bias.
- ✗
Use all available data including protected attributes.
Why it's wrong here
Using protected attributes may cause bias.
- ✓
Incorporate human review for high-stakes decisions.
Why this is correct
Human oversight ensures fairness.
- ✗
Ignore adverse impact if the model is accurate.
Why it's wrong here
Accuracy does not justify adverse impact.
Go deeper
Related to this question
About these practice questions
Courseiva writes every AI Associate question from scratch — 753 in total, each with an explanation and a wrong-answer breakdown. None are copied from real exams or dumps. Learn why practice questions differ from exam dumps →
Same concept, more angles
2 more ways this is tested on AI Associate
These questions test the same concept from different angles. Work through them to make sure you can recognise it however the exam phrases it.
Variation 1. A social media platform uses an AI model to automatically detect and remove hate speech. The model uses natural language processing and was trained on public posts. Recently, an internal audit reveals that the model removes posts from minority ethnic groups at a rate 3 times higher than from majority groups, even when the content is similar. The model achieves high precision and recall on the test set. The platform's content moderation team is overwhelmed with appeals. The company wants to maintain a safe environment while being fair. Which approach best addresses both goals?
hard- A.Disable the AI moderation and rely solely on user reports.
- ✓ B.Conduct an audit of the training data to identify gaps, then retrain with more representative data including diverse examples of hate speech and non-hate speech.
- C.Add more human moderators to review all flagged content from minority groups.
- D.Adjust the detection threshold only for minority group posts to reduce flags.
Why B: A comprehensive audit and retraining with diverse data addresses the bias at the root. Option A gives special treatment that could be seen as unfair. Option C removes moderation, risking harmful content. Option D does not solve the underlying bias.
Variation 2. An AI system used for recruitment has been found to be biased. Which THREE steps should be taken to address this? (Choose three.)
hard- A.Deploy the model without changes
- ✓ B.Audit the training data for bias
- ✓ C.Retrain the model with a balanced dataset
- D.Remove demographic data from the model
- ✓ E.Monitor outcomes for disparate impact
Why B: Auditing training data, retraining with balanced data, and monitoring outcomes are essential corrective actions.
JA
Written by Johnson Ajibi, MSc IT Security
Senior Network & Security Engineer · founder of Courseiva
This AI Associate practice question is part of Courseiva's free Salesforce certification practice question bank. Courseiva provides original exam-style practice questions with explanations, topic-based practice, mock exams, readiness tracking, and study analytics to help learners prepare for the AI Associate exam.