Checkout on the mobile app broke overnight, and only the last step moved
Conversion on the mobile app through paid search came in at today. That is the furthest any surface sits below its own floor, on the largest session count on the page. Desktop on email is at and mobile web on paid social at . Two surfaces are above their floors: desktop on organic at and the same mobile app on direct traffic at . Underneath it, checkout completion on the app fell to . It held inside its range every day for nineteen days before it, closed the night before last, and is lower again today than it was yesterday.
So what? Something changed overnight and is still deciding the outcome of every session that reaches the last step. The traffic is arriving and the intent is intact, so each hour this runs turns paid-for demand into nothing. A checkout that fails quietly produces no complaint, only a shopper who goes elsewhere.
61.0 vs 68.1%
and lower again today
1.92 vs 2.71% floor
furthest below its own band
8.8 vs 8.9%
the first step is fine
45.9 vs 46.2%
the second step is fine
2.46 vs 2.80%
same traffic, an eighth fewer orders
3.24 vs 2.90% floor
the same app, above its floor
Conversion Rate today against the band each surface is expected to hold
Three surfaces below their own floor, two above.
The scale runs from 0 to 5.
Mobile app on paid search came in at , the furthest below its own floor and on the largest session count. Desktop on email is at and mobile web on paid social at . Desktop on organic at and mobile app on direct at are both above theirs. Each floor is that surface's own, built from the same weekday in the three weeks before.
Checkout Rate by day · mobile app
Nineteen ordinary days, then two below the range.
Checkout completion on the app held inside its range every day for nineteen days before it. It closed the night before last, and today it is — so it is lower again today than it was yesterday. The largest move any ordinary day made across those nineteen days was under two points.
Each step of the checkout, today against a usual day
Two steps barely move. The third does all of it.
| Step | A usual day | Today | Change | Share of the loss |
|---|---|---|---|---|
| Sessions to bag | 8.9% | 8.8% | −0.1 pp | no change |
| Bag to checkout | 46.2% | 45.9% | −0.3 pp | no change |
| Checkout to order | 68.1% | 61.0% | −7.1 pp | the whole of it |
| The three together | 2.80% | 2.46% | −0.34 pp | an eighth of the orders |
Sessions to bag came in at and bag to checkout at , both unchanged. Checkout to order fell to . Multiplied together the app converted of its sessions. Traffic arrived above a usual day rather than below one, so more people reached the last step today than usually do, and fewer got through it.
2.46 vs 2.80%
on traffic that arrived in full
61.0 vs 68.1%
and it has not recovered
1.92 vs 2.71% floor
where the traffic was bought
One step, and only one
Sessions to bag came in at and bag to checkout at , both unchanged, while checkout to order fell to .
Why it matters: Two steps holding and one collapsing puts the fault at the end of the funnel, which is a different team and a different fix from a demand problem.
It happened overnight and has not stopped
Checkout completion held inside its range every day for nineteen days before it, closed the night before last, and is lower again today than it was yesterday.
Why it matters: A single bad day recovers on its own. Two days, the second worse than the first, is something that is still running.
The demand arrived in full
Traffic arrived above a usual day rather than below one, and the same app on direct traffic converted at against its own floor.
Why it matters: Nothing is wrong with the traffic or with the app in general, which is what makes every lost session avoidable rather than explicable.
How many orders has this cost since it started?
Sizes two days of a known rate against a known session count.
Is the last step worse in some categories than others?
A flat loss across categories points somewhere different from a concentrated one.
How is the same step doing on desktop and mobile web?
Says whether this is one device or one build.
Did anything else on the app change at the same hour?
The question this data cannot answer, and the one that has to be asked elsewhere.
Which surfaces were above their own band today?
The control: what a normal surface looks like on the same day.