I think it makes perfect sense if only for practical reasons: to realistically implement minimax in a game like this there is a need to reduce by any means the number of moves to analyze. Considering that a single unit in the middle of plains capable of moving 2 hexes has 19 possibilities per turn, it becomes clear that there is no way to perform a deep analysis of even the simplest of configurations (say 2 units vs. 2 units) in a short amount of time.
One way to simplify the tree of possibilities might be to restrict the search to "relevant moves" for any given unit, which could be for instance:
-attacks on enemy units
-moving to the hex offering the highest defense
-moving to the furthest hex from any enemy unit (a sort of retreat move)
This could bring down the range of possibilities per turn for a given unit from 20ish to 5ish. I think this is somewhat similar to the chess engine technique of selective extensions (which in chess is only applied deep in the tree search, but this ain't chess).
Combining this with hotspot detection (restricting the tree search to groups of a few closeby units) might yield decent performance. But I'm not convinced that all these sacrifices would result in a much better AI that what is currently implemented...
