How long is too long?Choosing a retargetingwindow.

Why recency predicts return

Habit is the engine behind most app use. A person who opened an app this morning has a routine that includes it, and the routine carries them forward. Each day without a session weakens it. The longer the gap, the more the app has to compete against everything else that has since filled the same slot in the person's day.

That is why time since the last session is the strongest single signal for retargeting. It stands in for several things at once: how embedded the habit was, how much the person remembers, and how likely a prompt is to land on a moment when they are open to it. Two people with the same purchase history can be worlds apart if one used the app yesterday and the other three months ago.

The same logic also shows why recency does not act as a straight line. Very recent users return on their own. Very old users have mostly moved on. The most responsive people sit in the middle, in the stretch where the habit has loosened but the memory remains.

The usage cycle sets the window

What counts as "recent" depends on how the app is meant to be used. A fixed number of days has no meaning until you know the app's natural rhythm.

• A daily game. Players open it every day or most days. Two days of silence is already unusual, and a week is a real lapse. The window opens early and the useful span is short, because the habit fades quickly and the game is easily replaced.
• A weekly grocery app. The natural cycle is the weekly shop. A person who skips one week has not lapsed. A person who has skipped several has. The window opens after the usual rhythm has clearly broken, and never earlier.
• A yearly travel app. People may book once or twice a year, and months of silence are normal. Measuring lapse in days would flag nearly everyone. The useful signal is a gap relative to the person's own booking season, and the window is measured in months.

The principle is to measure silence against the app's own cycle. A gap that is ordinary for one product is a warning for another.

Too early and too late

A window has two edges, and each fails in its own way.

If it opens too early, you reach people who would have come back unaided. Imagine a thousand weekly shoppers, offered here only as an illustration. Some skip a single week for ordinary reasons and return the next. An ad shown in the first quiet week reaches many of these people, gets credited when they return, and changes nothing. The money is spent on a return that was already underway, and the attribution report flatters it.

If it closes too late, or never opens, you reach people whose habit is gone. An ad that arrives months after the last session meets a person who barely recalls the app. They may tap, but the chance of a lasting return is low, and the cost of reaching them is the same as for someone more responsive. The honest framing is that late reach is not worthless, only more expensive per return and better suited to a different message.

Finding the window from the return curve

You do not have to guess. Your own data already contains the answer. Take people who went quiet in the past and ask: of those who eventually came back without any advertising, how long did it take? Plot the share who have returned against the days since their last session. This is the app's return curve.

The curve usually climbs steeply at first, as the people who were merely busy come back, and then flattens as the remainder drift further away. The point where it begins to flatten is the early edge of the window. Before it, many people are still returning on their own. After it, each further day of silence is a stronger sign that a prompt is needed.

A holdout then tests the edge. Run the same message at several points on the curve, each with its own holdout, and compare the lift. The window that shows the most additional returns per unit of spend is the one to use. Re-read the curve from time to time, because a redesign, a seasonal peak or a new competitor can move it.

Different windows for different value tiers

One window for everyone leaves money behind. A person who spent heavily is worth more to recover, so the sensible response is to reach them sooner, with a stronger message, and to keep trying for longer. A person who never made a purchase is worth less, so a light touch, later and capped tightly, matches what a return is worth.

In practice this means grouping the audience by value as well as recency and setting the opening edge, the creative and the spending ceiling for each group. Value can mean past purchases, subscription status, progress in a game or any measure that predicts what a returning person will do. The bidder then pays more for a high-value lapsed person and less for a low-value one, each within their own window.

Frequency inside the window

Being inside the window is not a licence to show the ad repeatedly. Frequency has its own shape. The first exposure does the most work, because it brings the app back to mind. Additional exposures add less each time and eventually begin to irritate, which can push a person further away and teach them to ignore the app's name.

A frequency cap sets the ceiling: so many exposures per person within a set period. A good cap is set against the window, so that a person near the early edge sees a light cadence and a person deeper in sees ads spaced further apart. Spacing matters as much as the total. Several impressions in one session are worth far less than the same number spread across days, because the person has not had time to act on the first.

Vary the message too. A person who has seen the same banner five times has learnt everything it can say. A change in offer, format or the screen it opens gives repeated exposure something new to offer.

When to stop

Every window needs an end. Two signals say stop. The first is the person returning: once they have a session, suppress them immediately, because an ad shown to an active user is wasted and inflates the result. The second is the curve of diminishing returns: past a certain point in the silence, a holdout shows the lift has fallen close to nothing and the spend no longer earns its place.

Ending the campaign for a person does not end the relationship. It frees the budget for earlier stages, where each unit of spend does more. Some advertisers pass the quiet remainder to first-party channels such as email, which cost little to keep running. Others return to the audience after a long pause with a fresh reason, such as a new release, and treat it as a new campaign with its own window and its own test.