Retest each winning paywall in every key market before rolling it out

David Barnard… actually haven't raised average revenue per user by the sum total of all those experiments. You're getting 10% here but losing 5% there, and getting 20% here but losing 10% there. What do you feel like was the key to actually stacking those successes? Were you looking back as you tested each of those to see if it had a negative impact on some of your previous tests, to be able to generate that, such a large increase in average revenue per user?

Michal ParizekEssentially executed a bunch of paywall and then price tests, and then whenever we've seen there is a positive results, whenever we've seen there is there's a statistical significant winner, usually if we've retested also in other markets. So usually let's say we've tested in the US or in Europe, and if it works there then we've retested in other, our key markets to make sure that actually it will work there. And it happened a couple of times that something we've only rolled out in some of the markets and not, let's say, all over the world just because we haven't really seen a positive impact. Sometimes also negative.

Michal ParizekSo we make sure that particular change is actually bringing additional incremental revenue in all those key segments are, according to geo typically, and if we haven't seen that then we didn't rolled out. And then in a high level we've monitored the ARPU seven day, typically actually in the RevenueCat charts to essentially make sure that for the new cohorts of users we actually seen that lift in the ARPU. Of course it's sometimes hard to isolate …