There’s no consensus yet on the “right” approach to AI projects. Nor will you find agreed, proven estimating models or benchmarks specifically for AI project costs and durations.
There’s plenty of guidance on how long regular IT development should take, but much of it tends to focus on activity to elicit requirements from users, or build technical designs and write code. There’s much less help for the more unstructured work at the heart of AI work — such as crunching data, evaluating algorithms and processing training data.
For now, estimating the non-traditional parts of AI work often relies more on personal experience and expectations than something more objective.
You can read this article in full in “About Data Science” by Medium