View a markdown version of this page

RAIRC04-BP01 Identify baseline performance targets - Responsible AI Lens

RAIRC04-BP01 Identify baseline performance targets

Set specific performance goals for your AI system before you build it. These goals become the pass or fail criteria that determine whether your system is ready to release. Good targets are based on real data, not guesswork, and assist you to make clear decisions about when your system is working well enough to release.

Level of risk exposed if this best practice is not established: High

Implementation considerations

  1. Research existing performance benchmarks in your domain by collecting data on how current solutions perform and what users need from your system. Look at industry standards, competitor performance, and user satisfaction data to understand the performance bar for your specific use case.

  2. Collect baseline data from existing systems, user studies, or pilot tests that show what performance levels are achievable and what users will accept for each of your metrics. Real baseline data assists you to set targets that are challenging but realistic instead of impossible or too simple.

  3. Set specific performance targets for your metrics by deciding what performance levels are acceptable for each measurement. This approach transforms your measurement capabilities into clear pass or fail criteria that guide your development and deployment decisions.

  4. Plan how you'll track and update your performance targets as you learn more about your system and users by building feedback loops that capture real-world performance data after deployment. Create processes for adjusting targets when you discover your initial goals were too high, too low, or missed important performance dimensions. Flexible target management assists you to improve your system over time while maintaining the discipline of clear performance goals.

Resources

Related documents: