Because they still produce changes in things like molecular shape, structure, affinity, size and so on. As long as these changes don’t substantially deleteriously affect an existing adaptive function, such changes can accumulate, and potentially produce new adaptive functions.
They can ever so slightly alter the active site of some enzyme, without negatively affecting it’s function in a substantial way, but having the inadvertent effect of making that active site better able to accommodate another substrate a bit different from the one it was already active on. That’s how for example the lactate dehydrogenases (LDH) enzymes evolved from malate dehydrogenases (MDH).
See: Boucher JI, Jacobowitz JR, Beckett BC, Classen S, Theobald DL. An
atomic-resolution view of neofunctionalization in the evolution of apicomplexan
lactate dehydrogenases. Elife. 2014 Jun 25;3. DOI: 10.7554/eLife.02304