You have to be prepared to deal with a lot of uncertainty if you want to work on Artificial Intelligence projects.
You can estimate and plan a traditional software project, but most importantly, you know that sooner or later, with more or less difficulty, you will end up getting the project built. It’s a deterministic process.
In most of the Artificial Intelligence projects you really don’t know if your idea of a project will be successful until you start evaluating the model and using the algorithms to make decisions. You really don’t know how many labeled items you need in order to get to the percentage of accuracy you want and maybe you don’t even know how much accuracy do you need to get a useful model.
The approach should be more similar to a research project, splitting the project into measurable small steps and defining a success criteria for each step, in order to stop the development of the project if you can’t meet the goals you established in the first place. And always have a plan B ready to deploy if you don’t get the desired results.