12,000 institutions checked daily: how do you avoid drowning in false alarms?

Multiple testingHard

Problem. You're building alerts on refresh success for every institution: flag any institution whose success rate today is significantly below its own normal. With about 12,000 institutions, the first version, which flags anything with p below 0.01, raises around 150 alerts a day, and the on-call team says almost all are nothing. How do you fix it?

Before you reveal: say your answer out loud, as if you were in the real interview — get your reasoning across clearly first. There is no single correct answer: reading what the interviewer is really after and defending your own thinking is what makes an answer strong.