Hey, Daniel! It is a great article that gives strong background on the AI projects pricing. We have been working on a very similar article recently. However, you have covered almost all. Let me give some comments though.
From my experience, there two types of clients that work with the Machine Learning consultants. The first is a customer without ML experts on board and looking to resolve a business problem. In this case, I advise against hourly rate, since we — as consultants — should take the risk of the project on us.
The second group is big corporations with experts on board, but looking for plug-n-play power in their research. This group would benefit more with an hourly rate, especially that the projects are usually connected with the travel.
To complement this article, I’ll recommend reading also our guide Questions to ask a Machine Learning Consultant to make a responsible choice. Rates charged by consultants are quite high, but that’s justified because this work is challenging and full of risks. As you said, the slippery proportion of the project can be estimated by the value of PI :)
One more thing to consider is also checking if the consultancy company have a dedicated DevOps / Backend team. As you said, in many projects, offer more time is spent on the server and interface than on the actual ML model. The philosophy we use is to separate these two worlds — of Software Engineering
Marek — Founder and Head of Artificial Intelligence @ Sigmoidal.io