Here's a template for an ML opportunity charter. This is done once the opportunity map leads us to decide which one is the first model to build; learn more here.
Project title
A clear working name.
Manufacturing problem
What currently happens, and why is it a problem?
Decision to support
What specific decision would the model help someone make?
User
Who would use the prediction?
Unit of prediction
One batch, panel, specimen, image, machine-hour or other unit.
Input data
What information would be available before the decision?
Target
What exactly should be predicted?
Current baseline
How is the decision currently made?
Success measures
Examples:
- Mean absolute prediction error
- Percentage of defects detected
- Reduction in laboratory trials
- Reduction in manual review
- Reduction in scrap or energy
- Time saved per batch
Cost of an incorrect prediction
What happens after a false pass, false reject or inaccurate numerical prediction?
Human oversight
Who reviews the model output, and who makes the final decision?
Data limitations
What information is currently missing, inconsistent or unreliable?
First pilot
What small, low-risk experiment could test the concept?