Over the years software professionals have a lot of tools, methodologies, and best practices established for planning and executing software projects.
There are subtle differences when it comes to AI projects.
AI projects are iterative in nature. They need data analysis and preparation, model building (training and testing), deployment, and optimization — which could lead to further data analysis. This cycle is nicely shown here:
One approach to handle this in-buil iterative nature of AI projects is to form an AI program. That consists of multiple AI projects. Each project going through a full series of steps as outlined. And at the end of each project (iteration), there is a clearly defined goal (OKR).
Then each project iteration could be further broken down into specific goals for each step. In many organizations, these steps are carried out by individual teams. Each team could have it’s own clear goal defined and exit criteria for completion of their work. And then these steps come together at the end of the iteration towards a final step for ensuring the overall goals of the project iteration are met.
This is shown here in a very simple form:
Typically each project iteration goal is to ‘evaluate performance, learn and adapt the future iteration. So that every future iteration reduces the ‘error’ or deviation from the expected outcome and actual results. Something like this:
Some good resources for AI program planning are:
Microsoft project template for data science projects. Google Cloud process for CI/ CD for ML OPS. AI / ML project organization guide.
But overall, after years of experience, what I can safely say is planning AI programs needs a very tight collaboration between project sponsors (executives) and the program planning (leadership) team. To ensure the AI programs ser set up for success with appropriate OKRs. More on that in the next article!
Originally published at https://medium.com on October 1, 2021.
