104. Scoping Process For ML Projects

Andrew Ng proposed a scoping process for AI projects like the following.

1. Brainstorm Business Problems (Not AI)
   Asking questions which are NOT AI problems can help you distinguish between the PROBLEM and the SOLUTION. You don’t know whether you should even use AI to solve the client’s problem.
   An example question might be like,
   “What are the top3 things you wish you were working better?”
   

2. Brainstorm AI Solutions
   After defining the problem you can look for ways to solve that with AI.
   

3. Assess Feasibility and value
   Determine the feasibility of the project using metrics such as

   • Human-level-performance
     “Can a human do what you’re trying to do, given the same dataset?”
   • History of projects improvements
     “From the rate of improvements from the past, do think it will keep improving until the expected performance?”
   • Available features for training
     “Do you think the available input x, is useful to map prediction y?”

4. Determine Milestones
   Be specific with the milestone you want to achieve, which balances the satisfaction between the ML engineer’s metric(Ex. Word-Level Precision) and businessman’s metric(Ex. Revenue).
   

5. Budget Resources
   Finally, budget all the resources you’ll need to achieve the milestones you’ve set.