We keep preaching on the importance of AI methodology. When writing our thoughts down, it came to us that the steps for running an AI project make a perfect START-UP.AI acrostic.
Here is the acrostic:
Scope,
Translate,
Access,
Recognize,
Test,
Update,
Produce,
Automate &
Improve
And here are the project steps that it stands for:
Scope the project
It is tempting to try and achieve it all, but grand and pretentious projects ate more prune to failures than modes and focused ones. Scoping should cover the business aspects as well as the available data sources.
Translate it from business language to a hypothetical model
Assuming that there is a signal in the data, can you imagine how does a machine learning model can support the business process? What kind of model can support the process?
Access the data
Now get access to the actual data.
Recognize data gaps
Run basic data QA, and spot quality issues in the data.
Test hypotheses
Improve your domain understanding, using the data. Speculate and test your speculations using data.
Update your modeling plans
With all the new insights that you collected from the data, can you have better modeling plans?
Produce a working POC
Transform your data and use it to build a model. Measure the performance of the model and use it for business validation.
Automate it, in production
Automate the training and the prediction of the model in production.
Improve
Over time, examine ways to improve the model