I'm looking at working on a machine learning project for a company, where they are interested in paying in instalments at certain milestones in the project. My initial thoughts are that how to define these milestones?
It's development of a recommender system and deployment on their site. It might take up to 12 months in worst case scenario.
I am thinking to ask for payment at these points:
- before starting the project (0 months)
- when I demo a prototype (low accuracy but bare bones functionality) (3 months)
- when I get the prototype running on live data (higher accuracy) (6 months)
- when it passes A/B testing and user evaluation (9 months)
- when it's fully deployed (12 months)
What I am not sure of is, is it practical to define milestones like this? I am also wary of defining acceptance criteria like percentage accuracy since it's really hard to work out what would be acceptable.
I am also looking into this standard for ideas for milestones: https://en.wikipedia.org/wiki/Cross-industry_standard_process_for_data_mining
which would suggest Business Understanding, Data Understanding, Data Preparation, Modelling and Evaluation. However these are more fluid "phases" rather than milestones.
I'd like to find out if there's an established best way to scope and plan projects like this and how to define acceptability at each stage?
Thank you!
