Flow Restraint: Every Extra Step Drops Another Batch

About Face 4, chapter 12, splits what users do into two kinds. One advances the goal directly—pressing the gas and steering on the drive to work. The other serves the tool: opening the garage, warming the engine, waiting at lights—none of which gets you closer to the office. That second kind Cooper calls excise.

In software, excise looks like this: you want to post an update, but first you log in, find the entry, dismiss a coachmark, and clear an upgrade dialog. Posting is the goal; the first four steps are tax. Cooper splits that tax into four kinds:

One test—easy to remember: If you delete this step, can the user still reach the goal? If yes, it's excise. Bring that test to AI-delivered flows.

Ask AI for a signup page and it'll likely give you a five-step wizard: email, phone verification, profile, interest tags, confirm—nothing missing. It's seen too many "proper" signup flows in training data, so it serves the fullest set. Each step looks reasonable alone; together they're a tax corridor.

Cooper was never kind to wizards: they freeze the process into Q&A, and users quickly learn to tap "Next" with their brain offline. His alternative: just do the job, back it with sensible defaults, and let users change later. In the AI era, you have to write that into the brief—it won't think of it alone.

Before you cut, learn the faces. Of the excise sources chapter 12 names, three show up densest in AI output. Open each card for symptoms and the brief you give AI:

Below is an AI-generated membership signup—all five steps. The funnel on the right assumes 78% pass per step: 1,000 people enter, 289 finish five steps. Use the test and audit step by step—cut steps that aren't advancing the goal. Cut right and the funnel rises; cut wrong and it tells you why that one must stay.

Signup is just the easiest example to count. Orders, publishing, registration, tickets—wherever there's a flow, this funnel shows up. Drag the slider to change step count and watch finishers move. A 22% drop-off is demo math only; real numbers vary by product, but the direction never flips: steps and retention move opposite ways.

Cutting steps has theory behind it. About Face 4, chapter 11, covers "flow": Mihaly Csikszentmihalyi's idea that deep focus puts people in a high-output state where time disappears. Cooper's corollary: interaction design's job is to protect that state. Good interaction should be "transparent"—users face the job itself and barely feel the software.

Flow's enemy is interruption. Every extra step, every dialog, every page hop pulls users out of the zone—and getting back takes a warm-up. That's why funnels drop people: drop-off often happens at the interrupt, when someone looks up and thinks "why am I filling this?" and closes the page.

Walkthrough for accepting AI output: run the core flow yourself end to end. At each step ask two things. First: if I delete this step, can the goal still be reached? Second: did this step yank me out of the zone? Only flows that pass both deserve to ship.

Warm up on a food-delivery checkout. One of four steps is classic excise—use the test to pick it.

Source: Original to Xiaoshan Academy's Interaction Engineering series; some principles adapted from About Face 4: The Essentials of Interaction Design.