Designing the review lane for tired people

The review lane is the least glamorous screen in our product and the one people spend the most hours in. Our rule while designing it was simple: a reviewer should be right in forty seconds, at 5 pm, on a Friday.

Start from the question, not the document

Our first prototype showed the whole document with the held fields highlighted. It was accurate and exhausting. Reviewers scrolled, hunted, and second-guessed fields we had not asked about.

The current screen shows one question at a time: the field name, the value we read, the reason we held it, and a zoomed crop of the page region it came from. The rest of the document is one key press away, not on screen by default.

Three decisions that mattered

1. Show the reason. “Confidence 0.62” is not a reason. “The 15th digit is smudged” is. Reasons turn a guess into a glance.
2. Make confirming the cheap path. Enter confirms. Editing is deliberate. Most held fields are right, and the interface should not punish the common case.
3. Never lose work. Every keystroke saves. Sessions time out gracefully. A reviewer pulled into a meeting returns to the same field.

What we cut

We removed side-by-side comparisons unless a rule asked for them, a progress bar that made people rush, and a leaderboard that one customer requested and then asked us to remove after a week. Speed matters, but a review lane that rewards speed alone will teach your team to confirm without reading.

What we measure

We watch time to resolve, but also the reversal rate: how often a confirmed value is later corrected downstream. If time falls and reversals rise, we made the screen worse, not better.

The best review screen is mostly empty, because the system did its job upstream. Everything else is making sure the few things that remain are easy to get right.