The side effects *in* a serum are generated at the time you research that serum. If you researched a serum that had side-effects, that recipe, in terms of the ingredients used, the quantities, the methods of production, etc. is flawed. You didn't know enough about it at the time to create an unflawed recipe. But if you keep using that flawed recipe, it will keep producing flawed batches, obviously.
I mean, it is plain, real-world logic we are talking about. If you didn't know how to hard-boil an egg, and you programmed machines to follow a recipe that, in the end, only made a runny egg, then despite the fact you now know better, so long as you follow the same process it will continue to produce the same level of runny egg.
Re-research the serum recipe to create a recipe that has no side effect. At 0 percent chance of side effect that is guaranteed. But 9 times out of ten you could have created a serum recipe with no side effects when you had a 10% chance of a side effect too.
You always want to be balancing out your investment compared to the rewards you expect. It sometimes makes a lot more sense to simply spend multiple time-periods on re-rolling a serum to get a side-effect free version, than to spend twice as many time periods researching a lower chance of side effect per roll.