Pilots are tightly scoped interventions that we expect not to work (remember, we have to explicitly prove efficacy as a defense against confirmation bias), so we use small populations, focus on speed to market, and do them in an operationally dirty way.