Why "Almost Winning" Converts Better Than Winning

It has the shape of a useful fact. Almost winning beats winning. It is counterintuitive enough to feel like insight, short enough to survive being passed along, and it arrives with research attached.

The research is real. Slot-machine studies have repeatedly found that outcomes landing just short of a jackpot — two matching symbols and a third a position away — produce more continued play than ordinary losses do. Some of that work found near misses activating reward-related brain activity in patterns resembling actual wins. It is a well-documented effect with a substantial literature behind it.

The problem is what happens when that finding is lifted into advertising, because the machines in those studies have a property that almost nobody carries across: the player cannot influence the result. Not slightly. Not at all. The reels are decided before they stop.

That is not an incidental detail of the experiment. There is a strong argument that it is the whole mechanism.

In a game of pure chance, a near miss is informative in a way that is completely false — and that falseness is the point.

Two matching symbols and a third just off the line looks like evidence that you are getting closer. It is not. Each spin is independent and the previous one contains no information about the next. But the display is built to resemble a skill outcome, and people read it as one, because reading outcomes as feedback about progress is what people do.

So the slot near miss works by producing an illusion of approaching competence in a system where competence is impossible. Strip the illusion away — tell the player plainly that the result was decided before the reels moved — and the motivating quality goes with it.

Now transfer that to a playable ad, where the player genuinely does influence the outcome. The illusion is no longer needed, because the thing it was imitating is actually present. And the mechanism that made the slot effect strong is the very thing you have removed.

This is why the advice tends to produce disappointing results in practice. Teams build an ad the player cannot win, expecting the slot finding to apply, and get an ad that feels rigged — because it is.

If closeness is not the active ingredient, something else is. The best candidate is the explanation the player forms about why they lost.

People do not experience a loss as a bare result. They immediately, automatically and usually without noticing produce an account of it — a reason. That account is what determines whether they want another attempt, and closeness matters only because it makes certain accounts more available than others.

There are roughly four explanations available, and they lead to completely different places.

A near miss is useful because it makes the first two accounts more available than the last two. A loss by a hair reads as luck or as a fixable error. A loss by a mile reads as being outmatched.

But this is a correlation between closeness and explanation, not an identity. It is entirely possible to build a near miss that produces the fourth account — and engineered near misses do it reliably, because a player who loses by exactly one move every single time eventually notices the pattern.

The honest answer is that it depends on what you want the player to do next, and the two options are not competing for the same job.

A win produces satisfaction. Satisfaction is a closed state — pleasant, well-disposed toward the product, and finished. The player has had the experience and is not carrying anything onward. That is a fine outcome if what you need is a good association, and a poor one if what you need is an action in the next thirty seconds.

A narrow loss with a legible cause produces an intention. The player wants to try again, and the only available route to trying again is the product. That is an outcome pointed directly at the thing you are paying for.

So the claim is defensible in a narrow form: a near miss converts better than a win for the specific act of installing, immediately. It is not a claim that near misses make people like your product more. They probably do the opposite, slightly.

Worth noticing that the left column is not a failure. A great deal of advertising is trying to produce exactly that state, and campaigns judged on anything other than same-session installs may well prefer it.

The reason the industry converged on the right column is not that it is better. It is that install counts are what most campaigns are measured against, and the right column is what moves that number.

The distinction that matters is between designing a game where narrow losses happen and designing a game that produces them on purpose. The first is pacing. The second is a claim about the player's agency that is not true.

There is an argument for engineered near misses that is rarely stated but obviously operating: it works, the numbers are real, and the player's irritation is somebody else's problem.

Taken one campaign at a time, that is roughly correct. The costs do not land where the benefit is booked.

What that reasoning misses is that the detection is collective. A person does not learn that one advertiser rigs its ads — they learn that playable ads are rigged. The scepticism generalises to the format, and it arrives at the next advertiser, who may have built something completely honest and now has to overcome a suspicion they did not create.

This is an ordinary commons problem and it has the ordinary trajectory. Early adopters of the tactic do well. As it spreads, the average player becomes harder to convince. Eventually the format performs worse for everyone, including the people who never used the tactic, and the response is usually to reach for something more aggressive.

We make a tool for building these ads, so our interest here is transparent: we would prefer a format people still trust in five years. But the argument does not depend on that interest. It is just what happens to a signal that gets exploited faster than it regenerates.

It is worth being clear about the state of the evidence, because this is a claim that circulates with far more confidence than it has earned.

The gambling research is solid within its own domain. Whether it transfers to a short interactive advertisement is not established, and the transfer is not a small step — it crosses from a game of pure chance to a game of skill, from repeated play to a single exposure, and from a self-selected gambling population to whoever happened to be scrolling.

The advertising evidence is thinner still. What circulates is internal campaign results, compared across teams measuring different things on different products, which is not a body of evidence so much as a collection of anecdotes with numbers attached. The claim has been repeated long enough to sound established without ever having been tested in the format it is applied to.

What we would stand behind is the narrower mechanism: the player's explanation for their loss drives what they do next, and that explanation is designable. Whether narrowness is the best way to shape it is an open question, and the assumption that it is has been repeated far more often than it has been tested.