Analytics

The Analytics page shows recent Nudges, delivery, and clicks. Default order is recency. You can sort by volume when you need it. Feed and Campaigns default to the last 30 days.

What the numbers mean

Sent is accepted delivery, not “everyone saw it”. Browsers can still drop or delay a notification.
Clicks are useful. They are not proof that the Nudge created the visit. People who would have come back anyway also click.
Qualified visits are subscribers who arrived and stayed: either they were on the page long enough to be reading it, or they opened a second page. It is a stricter bar than a click, and it is the outcome the effect card is scored on.

The comparison group

The Did it work? card is the only number on this page that answers the question the others cannot: did the Nudge create the visit, or would the visit have happened anyway. It can answer it because a group of subscribers deliberately gets nothing.

How the group is built:

About one subscriber in twenty is held aside permanently, for as long as they are subscribed.
Every live send holds aside a second group of the same size, drawn for that send.
Welcomes and test sends are never held out. A welcome that never arrives is a broken onboarding, not a measurement.
Audiences under 50 people are never held out. Taking a comparison group out of a small list costs you real reach and still gives you nothing readable.

Yes, this means a small share of your subscribers hears from you less than the rest. That is the price of the lift number being a measurement rather than a claim, and it is the whole reason the card is allowed to say "yes" out loud.

What the likely range means

The card reports the share who visited in each group, the difference between them, and a likely range around that difference. The range is a Newcombe-Wilson interval at 95 percent: the true effect very likely sits inside it, and it narrows as more people are measured.

The range, not the headline number, is the result. A lift of 4 points with a range from 1 to 7 is a finding. The same 4 points with a range from -3 to 11 is noise that happens to be pointing up this week.

So the card only says nudges worked when the whole range clears zero. Below that it says too early to say or no difference yet, whatever the point estimate happens to read. That is not the card being cautious about a result it has. It is the card refusing to report a result it does not have.

Why a claim gets withheld

Under the verdict, the card names what it is still short of: people in the comparison group, outcomes recorded, or subscribers sent to. Each has a floor, and all of them have to be met before any verdict beyond "too early" is allowed. A window that is still open is named too, because a period that has not finished can move.

There is no override for this and no setting that turns it off. Do not publish a lift claim from a thin sample, and the product will not hand you one to publish.

Measured, or estimated

Two kinds of number live on these pages and they do not carry the same weight.

Measured against the comparison group: the lift on the effect card, the extra visits it adds up to, and the extra opt-outs it cost. These have a group that got nothing to compare with.
Model estimates: Send fit, the score in the composer for how well one nudge fits your audience right now, and the "fewer unsubscribes" figure on the restraint receipts. These are the product's own reasoning about what would probably have happened. The receipts label the churn number as a conservative guess rather than a measured result, and that label is accurate.

Estimates are useful for deciding. Measurements are what you take into a meeting. When a number matters to somebody outside your team, check which of the two it is first.

Feed and campaigns

The home feed and campaign list are time-bounded. Use the date window. Load more stays inside that window.