How do you measure efficiency of data scientists?

I think it depends highly on the company's size and structure. There is a quite interesting article about how to use Agile in DS: https://towardsdatascience.com/applying-agile-methodology-to...

For us it was powerful to use certain Agile principles, e.g. splitting the work into weekly "sprints". The goal of the initial sprints is the development of the baseline model. In further milestones, we develop more sophisticated models. At each step, we measure metrics values and compare them with the baseline. It helps us to monitor the progress. After a few iterations, it's often easy to understand at what point you are now and how much time is needed to gain the required metrics.