Practice 03.05 · Push notifications

Measure impact after the tap

An open shows that the copy attracted attention; it does not prove value. Connect the send to the destination screen and target action, then add a comparison and negative signals.

Practice outcomeA report for one scenario covering sends, deliveries, opens, useful actions, a no-push comparison group, and trust impact.

90 minutes for the first report

In plain language

A tap is the middle of the journey, not the result

A push can collect many opens with loud copy while people do nothing after the tap or turn notifications off. The report must show the whole chain and answer one question: did the useful action start happening more often because of the message?

What you will need

Event chainSend, delivery, open, the destination screen, and one confirmed useful action.

ComparisonA small random group without the message, or at least a comparable period before launch.

Negative signalsNotification opt-outs, complaints, uninstalls, and repeat opens without the target action.

Terms in plain language

When to use it

The team reports delivery and open rate but does not know whether users completed the promised action — or whether it would have happened without the message.

In plain language

Where to assemble the chain

The send platform and product analytics must join by scenario and user. If the join is incomplete, state that as a limitation.

Send platform

Take the number of eligible users, sends, deliveries, and opens for one campaign ID.

Without an automatic export, export the campaign summary and record the exact send window.

Product analytics

Count destination-screen opens and the target action within a pre-chosen window after the push.

If the events are not joined, add campaign_id to the deep link and the screen event before the next launch.

Trust and control

Put opt-outs alongside, plus the results of the group that matched the scenario but randomly received no message.

Without a control group, compare with a similar week and state plainly that causal conclusions are limited.

Set up measurement for one scenario

Name one useful action

Before launch, define the post-tap result and the measurement window. Do not use a generic “app visit.”

Where to do it: In the scenario specification and the report title.
What the result looks like: Read at least half of the new analysis within 24 hours of delivery.

Join the events into one chain

Use a single campaign_id and trace the journey from the send platform to the screen and the action in product analytics.

Where to do it: In the push parameters, the deep link, and the app events.
What the result looks like: Send → delivery → open → screen view → half the material read, all with campaign_id=topic_update_01.

Add a comparable group

Randomly hold back a small share of the eligible audience. It shows the baseline level of the action.

Where to do it: In the launch settings and a separate row of the report.
What the result looks like: 10% of the audience gets no push; the action: 3.2% versus 4.9% among recipients.

Put negative signals alongside

Watch opt-outs and complaints in the same window. A lift in the action does not count as a win if the scenario visibly erodes trust.

Where to do it: In the last rows of the report.
What the result looks like: Opt-outs at 1% against the usual 0.3% — the scenario needs rework despite the lift.

Make one of three decisions

Keep, change, or switch off the scenario. Record the reason and the date of the next check.

Where to do it: In the report conclusion and the communications log.
What the result looks like: Keep the audience, cut the frequency to a single send, and repeat the comparison in two weeks.

Practical examples

Open rate 20%, action 4.9%

Of 8,600 delivered pushes, 1,720 were opened, but 420 completed the useful action. The control group came in at 3.2% — there is an incremental effect, but far smaller than the open rate.

The lift in actions did not cover the opt-outs

The target action grew, but opt-outs hit 1% against the usual 0.3%. The decision — reduce the frequency and repeat the test rather than scale right away.

The finished artifact

Funnel of one push scenario

The numbers must belong to one audience, one window, and one scenario. Show the comparison next to them, not only the absolute result.

Step	Users	Conversion	What it means
Matched the scenario	10,000	—	Eligible audience
Received the push	8,600	86%	Check permissions and tokens
Opened the message	1,720	20%	The copy caught attention
Opened the target screen	1,410	82%	The deep link mostly works
Completed the action	420	30%	Final conversion: 4.9% of deliveries
Turned notifications off	86	1%	Compare with the usual level

Control group without the push: 3.2% completed the useful action. The +1.7 percentage point difference makes the scenario a candidate to continue — if opt-outs stayed within the usual range.

Do not optimize for taps alone

Loud copy can collect curiosity opens while eroding trust. Judge the scenario by the whole chain and the negative signals.

Report checklist

Every metric must belong to the same scenario, audience, and measurement window.

The push has one target action and a pre-defined window. Send, open, screen, and action are joined by a single campaign ID. There is a comparable group without the message, or the comparison limitation is stated plainly. Opt-outs and complaints are shown next to the useful action. The report ends with a decision: keep, change, or switch off the scenario.

Which conclusions the team needs

Technical funnel

Delivery, opens, and arrival on the right screen show where the route breaks.

Incremental effect

The difference in the target action between push recipients and a comparable group without the message.

Trust

Opt-outs, complaints, and later retention do not degrade beyond the acceptable level.

A good report lets you switch off useless pushes as confidently as you scale the genuinely useful ones.