Why AI dashboards look like slop

Dashboards That Work, part one: why AI dashboards look like slop.

Ask an AI tool for a dashboard and you get the same screen: a row of stat cards, one chart, a table and a sidebar. It looks clean, and people still call it slop. Savva tested this with a design grader he's building, which checks public design rules such as contrast, text size and spacing. Dashboards built with AI passed seventy percent of the rules on average. Real status pages and analytics tools passed sixty-six to sixty-nine percent, and finance apps fifty-four. By the rules, the AI screens are as good or better. His conclusion: the rules check hygiene, and slop is about what the screen is for.

Here's the kind of screen an AI gives Wickerloaf, the fictional bakery from this course. The owner takes orders ahead for pickup, and she opens this at half past six every morning. Six cards, the same size, each in its own color. What should she do first?

The first tell: there's no answer at the top. Everyone opens a dashboard with a question. Is anything broken? Which of these needs me right now? This screen opens with totals instead. Forty-two orders this week is just a number. Savva points to Stephen Few, who wrote Information Dashboard Design, on a total shown on its own: compared to what? Is this good or bad? The fix starts before any design. Write one sentence: who opens this, what they want to know, and what they'll do next. For Wickerloaf, the owner opens it to find out whether any of this morning's orders can't be filled, and then she calls those customers. Now the top of the screen can say it. Two orders need a call, because the seeded rye came out three loaves short. Everything else either helps her believe that answer, or it waits.

The second tell: everything is equally loud. Every card the same size, the same weight, and its own color. It's what people mean when they say a dashboard looks like a Christmas tree. Blur your eyes and every block weighs the same. He quotes Nielsen Norman Group's guide to visual hierarchy: if everything is contrasted, then nothing stands out. So the answer gets the size, the place where the eye lands first, and the color. Everything else gets quieter.

Savva reads every dashboard at three speeds. A glance of about five seconds, to see if anything's wrong. A scan of about thirty seconds, to see what and where. And a few minutes of study, to understand why, and decide what to do. AI dashboards only serve the study. Everything is there, and it's no help when you have five seconds. Ben Shneiderman put the same order into his mantra for data screens: overview first, zoom and filter, then details on demand.

The third tell: it's never calm. A screen with something yellow on it every day teaches you to ignore yellow. Savva points to a Nielsen Norman Group study of smart home alerts from this year: with too many alerts, people stop reading them, and miss the one that matters. So when nothing needs the owner, the screen should look almost empty, and say so. Nothing needs you this morning. It says when it last checked. And the detail, all fourteen orders, is one tap away. That's what Nielsen Norman Group calls progressive disclosure.

The fourth tell is the worst one: a zero that lies. A morning with no orders, and an order system that didn't answer, look exactly the same if the screen shows zero for both. Nielsen Norman Group's first usability heuristic is visibility of system status: keep people informed about what is going on. So say that the data didn't load, and say how old it is. Orders couldn't load. The last update was at two minutes past six yesterday evening. Retry.

All four tells come from one gap. The AI doesn't know what the screen is for, or who is reading it, so it guesses. Part two is about telling it.

Your exercise. Take a dashboard you or an AI built, and write the one sentence it should answer. Mark which of the four tells you can see. Then fix the first one: put the answer at the top, with what it's compared to. You're done when you have a before and an after, and the after leads with your answer.

This lesson was adapted from Why AI dashboards look like slop, by Savva Sicevs. The bakery examples were added by Aesthetary. Read the original on his site, then go on to part two.