Traditional quality metrics can tell us whether an AI system meets defined requirements, but they do not always tell us whether its outputs should be trusted in real-world decision-making.

An AI system may pass tests, achieve strong accuracy, and still create uncertainty around explainability, consistency, human oversight, or the consequences of an incorrect result.

This session explores how traditional quality measurement can be extended with trust-oriented metrics and controls.

We will look at:
-- why quality and trust are related but not identical;
-- where traditional quality metrics stop being sufficient;
-- how trust requirements change with the impact and uncertainty of AI-enabled decisions;
-- practical approaches for measuring confidence, human verification, consistency, and control effectiveness;
-- examples from AI-assisted code generation and business applications.
The session introduces a practical way to think about trust as an architecture of complementary measurements and safeguards rather than a single score.

First presented: Software Quality Days 2026, Vienna

Target audience: quality professionals, AI and digital-transformation leaders, governance and risk specialists, architects, product leaders, and managers responsible for AI-enabled systems.

Preferred duration: 30–45 minutes, including Q&A.
Suitable for mixed business and technical audiences; no deep machine-learning background is required.