The step they skip is the one after they win

You have watched this happen and probably blamed the wrong thing. Someone finishes the part they came to do — they get the export, the answer, the generated draft — and then they don’t do the last small step: they don’t save it, don’t name it, don’t confirm, don’t hit publish. The step was one tap. It wasn’t hard. They just… didn’t.

It’s tempting to file this under laziness or bad copy. It’s neither. It has a name in cognitive psychology — the post-completion error — and it’s one of the most reliable failures in all of human-machine interaction. The classic examples are almost comically universal: leaving the original document on the copier glass after you’ve walked off with your copies; leaving your card in the older ATMs that gave you the cash first; driving away from the pump with the gas cap dangling. In every case the main goal was met, and the trailing step — the one that served the goal but wasn’t the goal — evaporated.

The mechanism is about how attention is powered. Your intention to do something is held in working memory, energized by the goal. When the main goal is achieved, that energy drains — the whole task feels done — and any subtask still waiting in the queue loses the thing that was keeping it lit. The card, the original, the save step: each was only ever there in service of the real aim, so when the aim is satisfied, they wink out. It isn’t forgetting a fact. It’s the goal that was holding the reminder up letting go.

Once you see it, the fix stops being “remind them harder” — a bigger toast notification fired after the goal is fighting a brain that has already closed the file. Three better moves, roughly in order of how much I’d trust them:

Put the housekeeping before the payoff. The strongest fix is structural: reorder so the trailing step isn’t trailing. This is why newer ATMs make you take your card before they dispense the cash — they moved the drop-prone step in front of the goal that kills attention. If your users must name a file, pick a destination, or accept terms, ask while they’re still climbing toward the result, not in the exhale after it.

Remove the step, or do it for them. A post-completion step that the system can just perform — autosave, a sensible default destination, an implicit confirm — is a step that can’t be skipped because it no longer depends on a drained human. The most reliable trailing action is the one the user doesn’t have to take.

If it must trail, make it a wall, not a whisper. When you genuinely cannot reorder or automate — the step is legally or destructively load-bearing — don’t rely on a passive nudge. Block forward motion until it’s done, and phrase it as the last rung of the same goal (“Publish to finish”) rather than a new, separate chore that a satisfied brain has no energy left to start.

The general lesson is quietly humbling for anyone who designs flows: the dangerous moment in your product isn’t the hard part. It’s the instant right after the user wins, when you have their success and have already lost their attention. Design as if that moment is exactly as fragile as it is.

The science, to look up: the post-completion error (Byrne & Bovair, 1997, “A working memory model of a common procedural error”); goal- and activation-based accounts of intention in working memory; related everyday-slip work by Donald Norman and James Reason on capture and description errors. This is distinct from simply forgetting — it’s the loss of a subgoal once the superordinate goal is satisfied.