What is a Bias Assessment?

A Bias Assessment helps you understand whether your AI system displays discrimination that could unfairly disadvantage certain groups of people. It is one of the most important assessments you can run, especially for systems that make or influence decisions affecting individuals.

In our platform, the Bias Assessment is divided into two parts: a Qualitative Bias Assessment and a Quantitative Bias Assessment. Every Bias Assessment starts with the qualitative stage, and depending on the findings, your team may proceed to the quantitative stage for deeper analysis.

Why bias matters

AI systems can be used to make decisions about people - from hiring and lending to customer service and content recommendations. When these systems display performance that differs between groups of people, certain users might be unfairly treated or disadvantaged.

Our Bias Assessment helps you understand what analysis is already in place and reveals whether your system exhibits acceptable performance across different groups. This is essential for regulatory compliance, ethical AI practices, and maintaining trust with the people your systems affect.

Qualitative Bias Assessment

The qualitative stage is where every Bias Assessment begins. This is a structured set of questions that evaluates your system's design, implementation, and deployment with respect to fairness and inclusion.

The qualitative assessment is organized into sections that cover:

• Impact - Questions about how your system affects different groups of people and what decisions it influences
• Web Accessibility - Whether your system follows accessibility guidelines and is usable by people with different needs
• Third Party - Whether external components or vendors introduce additional bias risk

Your team answers these questions based on their knowledge of the system. The responses give us a qualitative picture of where bias risks might exist and whether deeper quantitative analysis is needed.

Quantitative Bias Assessment

If the qualitative stage indicates that quantitative analysis is needed, your team can proceed to the Quantitative Bias Assessment. This is where we analyze your system's actual outputs to measure bias mathematically.

For the quantitative assessment, you provide your system's data in a structured format. We then analyze the data across the demographic groups you define - measuring whether the system's predictions or decisions differ unfairly between groups.

Here is how it works:

1. You upload your data with model predictions, labels, and protected attribute columns

2. You map the columns in our platform so we know which fields represent predictions, groups, and outcomes

3. We run statistical analysis and compute fairness metrics across each group comparison

4. Results are displayed as visual charts showing where differences exist and how significant they are

The metrics we compute give you a clear, measurable view of whether your system's behavior differs across groups. This goes beyond intuition - it gives you data-backed evidence of fairness or unfairness.

What you get from a Bias Assessment

After completing a Bias Assessment, you have a clear record of:

• The qualitative risk factors identified during the structured review
• Quantitative metrics showing how your system performs across different demographic groups
• Visual breakdowns that make it easy to communicate findings to stakeholders
• A clear audit trail showing that your organization has evaluated fairness as part of its governance process

Results are stored on the Asset in your inventory and can feed into downstream activities like Bias Mitigation, compliance reporting, and stakeholder reviews.