“the only experiments that we are sad about is an experiment that is inconclusive Right then we feel we we wasted our time or we didn't set up the experiment correctly or the hypothesis was not right... If we lost then well then we just learn something”

Treat experiments as a portfolio — inconclusive is the only real failure

Score the experimentation program on velocity × win-rate × average win. Losses still ship learnings; wins ship features. The only failure mode is an inconclusive test — you spent the bandwidth and produced no decision. Track inconclusive percentage as a portfolio health metric and explicitly attack it (better hypotheses, better instrumentation, larger sample) rather than chasing higher win-rate vanity.