Docs / Guias
Este artigo é exibido em inglês.
How EngageNudge decides
Every nudge goes through the same checks in the same order, whether you sent it from the composer, a feed sent it, or the scheduler did. This article is that order, start to finish, plus what you can overrule and what you cannot.
When support tells you the product recommended against something, this is the page behind that sentence.
The order things are checked in
1. When you compose. Before a recommendation appears, four things are read together:
- Audience. Who actually matches this campaign: topics, interests, device filters. If that is nobody, nothing after it matters.
- Send pressure. How many nudges these subscribers have already had in the recent window, against the caps in your attention policy.
- The daily budget. How many campaigns this site has already sent today.
- The confidence floor. How well the content fits, and whether that clears the minimum you set for routine sends.
Out of those comes the recommendation and the estimate in the confirmation dialog: how many can receive this now, how many cannot, and why. If nobody can, the send is refused there rather than starting a run that cancels itself minutes later.
2. Billing. Your plan's audience cap, the send pause, and whether the DPA is accepted. These are not judgment calls and they are not part of the recommendation. They allow the run or they do not.
3. Each subscriber, during fanout. The run then walks the audience one person at a time:
- Frequency. Their daily cap, their weekly cap, and the minimum spacing since the last push to them.
- Engagement state. Somebody who has stopped clicking carries a tighter cap of their own. Somebody dormant is left out of broad sends. Both belong to the person, not to the campaign.
4. The comparison group. A small share of subscribers is held aside so the effect card has something to measure against. Analytics covers how that group is built and why it exists.
5. At the moment of delivery. Timing is checked last, against the clock when the message would actually go out rather than when you pressed send, so a scheduled or delayed run is judged on the time it lands. Both your site's window and each subscriber's own window are read here. Fatigue and attention covers how those windows behave.
Anyone held at any step shows up on the card afterwards, with the step that held them and the setting behind it.
What an override moves, and what it does not
Send anyway exists because restraint is a default, not a lock. The product will argue with you, and it will never stop a send silently.
An override relaxes exactly these, and nothing else:
- your daily and weekly caps, and the minimum spacing between pushes
- the tighter caps carried by subscribers who have cooled off
- the exclusion of dormant subscribers from broad sends
- your site's own delivery window
It never touches:
- a subscriber's own window. They set it themselves in the preference centre. Nothing on your side moves it, including a breaking story.
- consent. Unsubscribes and dead endpoints are final.
- your plan and fair-use limits, the DPA, and the send pause
- a feed's own daily cap and minimum delay
Afterwards the card says what the override cost, in the form "Sent to 120, of whom 85 were past the recommended limit", and the send is filed as an override in your receipts. Nobody has to take your word for it later, including you.
Measured, or estimated
The product keeps these apart, and it is worth doing the same when you report upward.
Measured against the group that got nothing: the lift on the effect card, the extra visits behind it, and the extra opt-outs it cost. A claim is only made when the likely range clears zero and the sample floors are met.
Estimated by a model: Send fit in the composer, and the "fewer unsubscribes" figure on the receipts. These are the product's reasoning about what would probably have happened. Worth having while you decide, worth labelling when somebody outside your team reads it. The receipts call the churn figure a conservative guess, because that is what it is.
Why hold anything back at all
The short version: a rested audience clicks more.
The longer version is that the two sides of one extra send are not the same size. The upside is a few more clicks this week. The downside is either a permanent unsubscribe or the quieter version, where somebody stops opening anything from you and stays on the list looking like an asset. The first cost is recoverable next week. The second one is not, and it does not show up in a click-through rate.
So the product spends your sends where they are worth the most and declines where they are not. Then it shows you the bill for that: What we skipped on each nudge, and Held back this week across the whole programme, each with its reason. A tool that only shows what it sent is asking to be trusted. This one shows what it withheld, so you can judge whether the restraint earned its keep, and overrule it when it did not.