Offer decline rates are a design signal worth reading

Offer reporting is built around conversion. What percentage bought, at what price, in what context. Everything that did not convert is aggregated into a single residual and ignored.

That residual contains most of the information. A player who saw an offer and did not buy has told you something specific, and the shape of who declines what is a better guide to catalog design than the conversion numbers are.

Declines are not one thing

Bundling all non-purchases together is the mistake that makes the number useless. There are at least three distinct behaviours in there.

A player who saw the offer, engaged with it — opened it, looked at the contents, considered — and then closed it has evaluated and rejected. That is a real signal about the offer.

A player who dismissed it within a second has not evaluated anything. They were interrupted and closed the interruption. This says something about timing and placement, and nothing about the offer's contents or price.

A player who never saw it — it appeared while they were mid-action, or in a screen they skip — has not participated at all. Counting them as a decline makes a placement problem look like an offer problem.

These three get very different responses, and a single conversion percentage merges them into a number that cannot distinguish any of them. Separating engaged declines from instant dismissals is usually a small instrumentation change and it makes everything downstream readable.

What a considered decline tells you

The engaged decline is the valuable one. This player wanted to know what was on offer, found out, and said no.

If a specific offer is declined heavily after engagement, the problem is the offer — the contents are not wanted at that price by these players. That is actionable and specific.

If a player declines many different offers after engaging with each, the problem is not any individual offer. This is someone who keeps checking and never finds anything worth having, which usually means the catalog does not contain what they want. That is a much more important finding than any single offer's performance, and it is invisible unless you look per player rather than per offer.

If declines concentrate at a particular point in progression, the offers being shown there are mismatched to what players need at that stage. A player working on one problem being offered help with a different one declines regardless of price.

Each of these points at a different fix, and none of them is "lower the price", which is what an undifferentiated conversion number tends to suggest.

The decline sequence matters more than the rate

A single decline is weak evidence. A sequence is strong.

The pattern worth tracking is what a player does across several consecutive offers. Someone who declines four in a row and then stops opening the store has disengaged from the store entirely, and everything shown to them afterward is noise that costs attention and gives nothing back.

That sequence is a better trigger for changing approach than any individual offer's numbers. It says the current strategy for this player is not working, and continuing it has a cost — repeated irrelevant offers train players to ignore the surface they appear on, which then applies to the offer that would have converted.

The inverse sequence is equally informative. A player who declines several and then buys has been shopping, and what they eventually bought tells you what the earlier offers should have been. Comparing the purchased item to the declined ones for the same player is one of the cheapest sources of insight available, and it requires no new data.

See how we use decline behaviour to change what gets shown →

Using declines to decide what to stop showing

The practical output of reading declines is usually subtraction rather than addition.

An offer with a high engaged-decline rate across every population is not converting anyone and is occupying a slot. Removing it improves the store for everyone by making room for something with a chance.

An offer with a high decline rate in one group and a good rate in another should not be shown to the first group. This is the most common finding and the easiest to act on: the offer is fine, the audience is wrong.

A player with a long decline sequence should see fewer offers, not more. This is counterintuitive and consistently correct. Reducing volume for someone who has declined repeatedly preserves the possibility that they engage with a future one. Increasing volume guarantees they stop looking.

The general reframe is that a decline is a player answering a question you asked. Conversion tells you which questions got a yes. Declines tell you which questions were worth asking at all, and to whom — and a catalog tuned only on the yeses keeps asking the same wrong questions to the same people indefinitely.

Talk to us about reading offer response data →