Comment by lesinski

A lot of big organizations have absorbed a Lean Startup POV -- don't debate, just ship an MVP and measure. But in practice, we're usually measuring a proxy metric, it's a short-term test, the audience may be unrepresentative and because of all this, the result doesn't generalize.

What do you think an experiment needs before it can actually be called learning? And what kinds of product questions should not be done as experiments at all?