Measure Web Push as a sequence of permission, eligibility, delivery, response, and downstream outcome. Without the definition above, two dashboards can use the same label for different calculations.
Permission metrics
Explanation-to-prompt rate
People who select the custom enable action divided by people who saw the explanation. This tests the clarity and relevance of the permission proposition.
Prompt grant rate
Granted native permission outcomes divided by native prompts shown. A dismissed prompt should not be silently grouped with a denial if the browser exposes them separately.
Stored subscription rate
Subscriptions successfully saved divided by granted outcomes. A gap can reveal service-worker, key, network, or backend validation problems; the subscription lifecycle guide defines the browser and server states behind that rate.
Delivery metrics
Eligible subscriptions
Stored subscriptions remaining after permission, preference, frequency, quiet-hour, recent-send, and endpoint-health rules are applied.
Attempted deliveries
Unique campaign and subscription pairs submitted to the sending queue. Retries for the same pair should not inflate this count.
Provider accepted
Attempts accepted by the push service. Acceptance means the provider received the encrypted message, not that a device displayed it or a person saw it.
Permanent failure rate
Permanent endpoint failures divided by attempted deliveries. Use the Web Push status-code action table to separate endpoint retirement from bounded retry.
Normal Web Push delivery does not provide a universal proof that the notification was displayed or viewed by a person.
Response metrics
Click-through rate can use attempted, accepted, or delivered as the denominator depending on available data. Name it explicitly. Also measure unique clickers, repeat clicks, notification closes where supported, and the delay between send and click.
Downstream outcomes
The click is rarely the business objective. Measure the action promised by the message: article read, report downloaded, order reviewed, reminder completed, product saved, or account issue resolved. Use campaign identifiers that remain useful without exposing subscriber secrets in URLs.
Trust and fatigue metrics
- Explicit unsubscribes per accepted delivery
- Browser permission loss over time
- Inactive subscriptions by last click or visit
- Outcome rate by recent-send count
- Suppression counts produced by the frequency-cap policy and quiet hours
- Permanent endpoint failures by browser family and subscription age
A compact measurement record
For each campaign, store the audience definition, eligibility time, schedule, attempted count, provider outcomes, retry count, permanent failures, clicks, opt-outs, destination outcomes, and the version of the message that was sent. This makes a result explainable later.
Questions a useful report can answer
- Where did eligible subscribers get excluded?
- Which provider errors are temporary and which are permanent?
- Did the message create the promised downstream action?
- How did recent notification volume affect opt-outs?
- Are inactive endpoints being removed?
- Can the campaign be reproduced from its stored audience and message rules?
Use one event identity across the funnel
Create an internal campaign ID and one send-attempt ID per subscription. Carry a non-secret campaign token into the click destination, then join it to downstream events under a stated attribution window. Do not put a raw endpoint, subscription key, email address, or account secret in the URL. Deduplicate retries so the same intended message does not become several attempts in a campaign total.
Separate monitoring from experimentation
Operational monitoring asks whether the system is healthy: queue latency, acceptance responses, permanent failures, and worker errors. Experimentation asks whether one eligible message policy creates more useful outcomes than another. For an experiment, define the primary outcome and control group before sending. Report the group sizes and uncertainty instead of declaring a winner from a small click difference.
Publish metric definitions beside the chart
PushNotifications.org's delivery model keeps attempts, accepted requests, failures, and clicks separate; use the definitions on this page to keep each dashboard label equally precise.
A dashboard should reveal the numerator, denominator, time zone, attribution window, deduplication rule, and known blind spots. This is especially important for “delivery.” The HTTP Web Push protocol defines push-service request behavior, while the Push API defines delivery to the worker. Neither creates universal evidence that a person noticed the notification.