Why do AIs often struggle to find mate?
Pikafish and the vast majority of board-game AIs focus solely on playing-strength test results. Most positions in which their mate-finding looks poor are positions where one side already has an overwhelming advantage. Neural networks distinguish poorly between positions with such large advantages, which leads to this behavior. It has almost no effect on playing strength; mate-finding ability and playing strength often pull in opposite directions.
Mate-finding in a position with an overwhelming advantage does not represent playing strength.
Concluding that an engine is weak because it struggles to find mate in a few positions makes at least two mistakes: mate-finding ability ≠ playing strength, and the sample is too small to have statistical significance.
Of course, a small minority of cases may involve an equal or slightly better position in which the engine misses a mate. Such cases do exist and are real engine blind spots.
Related: Is a stronger engine better in every position?, How do you test an engine?.