Purpose

ConveYour assesses material fairness and bias risks in AI-enabled workflows before release and throughout use. The goal is not to claim that an AI system is bias-free. It is to identify credible ways a workflow could produce unfair, discriminatory, misleading, or otherwise harmful outcomes, then reduce those risks and document the work.

This approach is informed by the practical structure of the NIST AI Risk Management Framework: identify the context and risks, test them, document results, and manage issues over time. NIST AI RMF

Where bias risk can arise

Bias risk does not live only in the underlying AI model. It may arise through:

• The customer workflow and the decision the AI is asked to support
• Data provided to the feature, including incomplete, historical, or proxy information
• Prompts, instructions, examples, filters, and scoring criteria set by ConveYour or the customer
• The way an output is displayed, interpreted, or acted upon by a user
• Differences in performance across language, disability, culture, job type, or other relevant contexts
• A user's overreliance on an AI output without enough review

For employment-related workflows, ConveYour treats use cases that could influence access to an opportunity or an employment action as higher risk.

Risk assessment before release

For each material AI-enabled workflow, ConveYour records a practical risk assessment that includes:

• The feature's intended use and foreseeable misuse
• The users affected and the potential impact on them
• Inputs the feature receives and outputs it produces
• Whether sensitive personal information, protected characteristics, or likely proxies may be present
• The prompts, instructions, rules, and product configuration that shape the output
• A review of whether the workflow could unfairly favor, exclude, stereotype, or disadvantage people