A familiar sequence.
The campaign clears its D7 ROAS target. Budget goes up. A few weeks later the cohort payback estimate has slipped, sometimes by months, and nobody can point to the thing that changed. Creative is the same, CPI is in band, D7 is still on target.
Nothing is broken. The target was measuring the wrong part of the curve.
D7 sees about a tenth of the answer
D7 ROAS is the revenue a cohort produces in its first seven days, divided by what it cost to acquire. For a direct-response channel it is a fast, honest read of whether the spend is working at all.
What it is not is a read of payback, because on a free-to-play title payback is decided by the tail.
Take the free calculator's published defaults: D1 retention of 38%, D7 of 17%, D30 of 9%, a cost per install of $1.10 and ARPDAU of $0.12. Each install returns $1.49 of net revenue in its first year. $0.14 of that arrives in the first seven days. D7 ROAS is scoring the cohort on under a tenth of its first-year revenue, and the other nine tenths are the part that decides when it pays back.
So the question is not whether D7 is a good metric. It is what else has to be true for a D7 number to mean what the room thinks it means.
Move D30 by a point
Hold everything at those defaults. D7 ROAS is 0.13 and payback lands on day 207. The calculator shows payback and return from day 30 onwards; D7 ROAS is the one number here it does not display, and at these inputs it is 0.13 in every case below except the last.
Now change one input: D30 retention from 9% to 8%. D1 and D7 are untouched, so D7 ROAS is still 0.13 to two decimal places. Payback moves to day 275.
Open that comparison and put D30 back to 9 to see it snap back. At 7% it no longer pays back inside the year at all, and D7 ROAS is still 0.13.
A single point of D30 is a small move. It is well inside the difference between two audiences, two creatives or two weeks of the same campaign, and D7 is blind to all of it.
The campaign that beat target
It gets worse when the early read improves.
Scaling into new audiences can buy players who engage hard early and leave sooner. Model that as D7 up two points to 19% and D30 down one to 8%, with nothing else changed.
Open it here. D7 ROAS rises from 0.13 to 0.14, so on the dashboard this campaign is beating the one it replaced. Payback moves from day 207 to day 363.
The better-looking cohort is the worse buy by five months, and the metric the budget follows says the opposite.
There is a second, less obvious effect in the same numbers. On a curve fitted through those three points, better early retention does not lower the D30 you need. It raises it. At D7 of 17%, holding payback at day 207 takes D30 of about 9%. At D7 of 19% it takes about 9.3%, because a curve that starts higher and lands in the same place at day 30 is falling faster when it gets there.
Put the tail in the target
The fix is not to abandon D7. It is fast, and speed matters in UA. The fix is to stop letting it stand alone.
Pair every D7 target with the D30 it assumes. A D7 ROAS target was set against a curve, whether or not anyone wrote the curve down. Write it down. At the defaults above, payback inside 180 days needs D30 of roughly 9.7%, inside 270 days roughly 8%, and inside a year roughly 7.4%. That is the floor the D7 target silently depends on.
Read D7 as a leading indicator of the curve, not of the return. When D7 comes in on target, the next question is whether D14 and D30 are tracking the curve the target assumed. If early retention improved and later retention did not, that is a warning, not a win.
Judge scaled spend on the latest point you can see. New audiences change the shape of the curve, and the shape is what payback depends on. The first cohort's D30 from a new audience is worth more than its D7 ROAS, even though it arrives three weeks later.
None of this needs new data. It needs the target to say what it is assuming about the part of the curve it cannot see yet.
Running it
Kadar ranks which assumption is driving your payback day, so you can see how much of it rests on the tail rather than the first week, and solves the inverse: name a payback day and it returns what one input would have to change to get there, on your own numbers.
Run your own numbers in the free calculator. No sign up.