The stages of a project.
DEFINING THE PROBLEM
What decision or task is this for? What does success look like?
Most failed projects failed here, by starting with the technology rather than the problem.
GATHERING DATA
Usually the largest cost. Data must be relevant, sufficient, and correctly labelled where labels are needed.
PREPARING DATA
Cleaning, removing errors, handling missing values, ensuring consistency.
Frequently most of the work.
CHOOSING AN APPROACH
Whether to train something, fine-tune something existing, or use a general model with good prompting.
For most business problems, the last option is now the right starting point.
BUILDING AND TESTING
Training, then measuring performance on data the system has not seen.
DEPLOYING
Putting it where it will be used, with monitoring.
MONITORING
Performance degrades as the world changes. A model trained on last year's conditions drifts.
THE STAGE PEOPLE SKIP
Deciding how you will know it is working after deployment.
