The Averages Trap: Why Cohort Blending Hides Burning Markets
Blended D30 can stay flat while the market receiving your next UA dollar deteriorates. An average is useful for reporting; it is not automatically the right number for allocation.
One of the easiest ways to make a bad scaling decision is to trust a blended retention number.
Take a simple, illustrative example. A game’s Tier-1 cohort is 20% of users and retains 12% at D30. The other 80% retains at 7%. The blended number is 8%.
Later, Tier-1 retention falls to 6%. The other cohort improves to 8.5%. The blended number is still 8%.
The arithmetic is fine. The business read is not.
If the next UA dollar is going into Tier-1 markets, 8% blended D30 is not enough evidence to scale. I want the cohort receiving that dollar split out in the MMP — by source, geography, platform, and product version.
The same thing happens with monetisation. Blended ARPDAU can look stable while high-value payer behaviour is deteriorating. Mix can hide the movement that matters.
An average is useful for reporting. It is not automatically the right number for allocation.
Before approving the next step, ask: what changed inside the average, and does that change the decision?