Feedforward

The word arrived as a complaint. In 2002 Djajadiningrat, Overbeeke and Wensveen published a paper whose title asks Donald Norman how, exactly, a person is supposed to know what a control does, and their answer was that an interface owes two things and the literature only talked about one. Feedback reports what happened. Feedforward communicates the purpose of an action before it is performed. Eleven years later Vermeulen, Luyten, van den Hoven and Coninx pinned it to Norman’s own diagram: feedback is the bridge over the gulf of evaluation, and feedforward is the bridge over the gulf of execution, the gap between having an intention and knowing which act will satisfy it.

The reason the word is worth learning is that it separates three things a crowded neighbourhood keeps mashing together. An affordance is what the thing lets you do, a property of the relationship between an object and a person. A signifier is the mark that tells you where and how to do it, the underline under a link, the handle on a drawer. Feedforward is neither: it is a statement about the outcome. “You can press this” is an affordance, “press here” is a signifier, and “pressing this deletes 14 files and cannot be undone” is feedforward. The test is the tense. If the sentence the interface is saying uses the future, and names a consequence rather than a mechanism, it is feedforward.

It pays for itself most obviously where the act is expensive: destructive commands, bulk operations on a selection, anything irreversible, anything that costs money, anything that grants a permission. In all of those the general rule is that specificity is the whole value. “This may affect other items” is worse than nothing, because it teaches the reader that the warnings here are noise; “removes 14 projects, every build log, and the deploy keys” is a sentence somebody can actually act on. The same logic explains the pattern’s smallest instances, where feedforward is not text at all: a drop indicator drawing the line the row will land on, a drag preview carrying what is being moved, a price total updating while the option is only hovered. Each of those is a claim about the result, made before the release.

Two duties come with it. The first is reach: an announcement that only exists on hover does not exist for a touch screen, for a keyboard, or for a screen reader, so the consequence has to be stated somewhere durable as well, on focus, in the confirmation step, or in the label itself. The second is honesty, which is where the pattern meets its abusers. A cost revealed only after the commit is drip pricing, and a preview that omits the part somebody would object to is the same lie with better manners. In the four parts of a microinteraction, feedforward is the odd one out: it is not the trigger, the rules, the feedback or the loop, but the thing said before the trigger is pulled. That is exactly why it needs its own name, and why so many interfaces have none of it.