AI Foundations
Baseline model
A simple reference model used to judge whether a more advanced approach is actually better.
Example
A team compares a new fraud detector with a simple rules-only system before claiming improvement.
Why it matters
It helps teams avoid spending more on a complicated approach that adds little value.
You'll hear it
“Does the new system beat our simple starting approach?”
Can you control it?
Set by the team
The team evaluating a system chooses the baseline and comparison measures. Buyers can still ask what baseline was used and whether the improvement is meaningful.
Common questions
- Can a deliberately weak comparison make progress look bigger?
- Yes. Beating an unsuitable starting point can exaggerate how much useful improvement was achieved.
- What makes a good baseline?
- Use a relevant approach that can be repeated and checked on the same task and examples.
- Can the current process be the baseline?
- Yes. Comparing the model with current human or rules-based performance can be more useful than comparing only with another model.