It's not a soft science.
The problem is that too many people believe you can do research on one group of people, and generalize that universally to all humans, when in fact, the variation in every population of humans is wide enough that you can't say for certain that a given treatment will work for a given set of symptoms—not because the treatments are bad, but because there are differences both in the causes of the same symptoms, and in the workings of the body & brain, between different people.
This doesn't make psychology/psychiatry/psychopharmacology a "soft science"; it just makes it a science that is still in its infancy. Once we have a better understanding of both the various underlying neurological/physical (and, for some, even gastrointestinal, given recent research showing that gut microbiota can affect the mood and brain) causes of various psychological symptoms, and the physical and neurological variations between people, it will be much easier to see, for instance, "ah, we shouldn't use Lorazepam for this patient, because their anxiety is caused by this which is much better treated by CBT and CBD, rather than that which Lorazepam directly addresses".