Evaluation & Quality
Offline evaluation
Testing an AI system on saved datasets before exposing it to live users.
Example
A team scores a model on a fixed dataset before releasing it.
Why it matters
It lets teams investigate weaknesses before ordinary users encounter them.
You'll hear it
“Let's check the saved cases before trying it with live traffic.”
Can you control it?
Set by the team
The people building or running the AI choose this setup. An everyday user generally needs their help to change how this part works.
Common questions
- Can saved tests miss real-world problems?
- Yes. Changing information, user behavior or connected services may not be represented.
- Does offline mean no internet can be used at all?
- Not necessarily. Here it contrasts saved-case testing with measurement on live user interactions.
- Can results improve without users benefiting?
- Yes. The saved cases may reward behavior that matters less in everyday use.