What separates the best SaaS dashboard UX design from the rest
The patterns behind the best SaaS dashboard UX design: default views that answer a question, density that respects real data, states that are designed, and the performance budget nobody talks about.
Almost every B2B product eventually grows a dashboard, and almost every one of them is designed backwards: someone assembles the charts the data team can produce, arranges them on a grid, and calls it an overview. The best SaaS dashboard UX design starts from a different question — what is the first thing this person needs to know when they open the product, and what will they do about it?
That question, asked seriously, produces a recognisably different screen. Here's what editors see in dashboards that score well.
1. The default view answers one question
A dashboard that tries to serve everyone serves nobody, and then acquires filters to compensate. The strongest ones pick the single most common reason a person opens the product and answer that above the fold. Everything else is one click away.
The test is brutal and simple: cover everything except the top third of the screen. Can the user tell whether anything needs their attention today? If not, the dashboard is a data display, not a dashboard.
Data display
- Twelve cards, equal weight, no hierarchy
- Numbers with no baseline or comparison
- Charts chosen by what's available
- Filters as the fix for "it's not relevant"
Dashboard
- One answer, immediately, above the fold
- Every number against a target or a previous period
- Charts chosen by the decision they support
- A configurable default instead of a filter bar
2. Density matches the job, not the design file
Design files contain eight rows. Production contains eight hundred, with names that are too long, statuses that don't exist yet and a currency column that includes three currencies. Dashboards designed against tidy data fall apart on contact with real accounts — and editors always look at a real account.
- Test with the longest realistic string, not a representative one. Names, company names, email addresses, product titles.
- Decide what happens at 0, 1, 10, 1,000 and 100,000 rows. All five are different designs.
- Give the dense view its own density. Comfortable spacing on a table someone reads for six hours a day is not generosity, it's scrolling.
- Let power users bring their own density: a compact toggle costs little and earns loyalty.
3. The states are designed
Most of a dashboard's life is not the populated happy path. It's loading, partial, stale, empty, erroring, or permission-limited — and these are where craft is visible, because almost nobody designs them.
| Compared on | Designed | Defaulted |
|---|---|---|
| Empty | Explains what will appear and how to make it appear | "No data" |
| Loading | Layout-stable skeleton, no shift when it resolves | A spinner over a blank area |
| Partial | Says which panel failed, keeps the rest usable | Whole page falls over |
| Stale | Shows the timestamp and a way to refresh | Silently old |
| No permission | Says who can grant it | Panel simply missing |
| Error | What happened, what to do, no jargon | "Something went wrong" |
4. Filters are a last resort, not an information architecture
A row of filters at the top of a dashboard is usually an unresolved argument about who the screen is for. Every filter added is a decision passed to the user, and a decision passed to the user is a cost — one they pay every single visit, forever.
Better patterns, roughly in order of preference: a smart default based on the person's role; saved views they can name; a segmented control with two or three options; and only then a filter bar. If you must have filters, make the current state legible in one glance and always offer a way back to the default.
5. Speed is part of the design
Dashboards are the worst-performing screens in most products, because they aggregate. They're also the screens people open most often, which makes the cost compound. Editors score performance on every entry, and dashboards are where entries lose points quietly.
- Render shell first4Structure and navigation before data. The page should never be a blank rectangle.
- Stream panels4Each panel resolves independently; one slow query doesn't hold the page.
- Reserve space3Skeletons with the final dimensions — no layout shift when numbers arrive.
- Cache the last view3Show yesterday's figure with a timestamp while today's loads.
- Defer the heavy chart2The one nobody looks at first can load when it scrolls into view.
6. It works without a mouse
Dashboards are used all day by people who are fast at their job, and fast people use keyboards. Full keyboard operation isn't only an accessibility requirement — though it is that — it's a power-user feature that happens to be the same work.
- Every action reachable by keyboard, with visible focus that is never obscured by a sticky header.
- Tables navigable by arrow keys, with a clear relationship between the focused row and the action bar.
- A command palette if the product is complex — the single highest-leverage addition to a dense product.
- Shortcuts that are discoverable, not folklore. A
?overlay is the convention for a reason.
What we'd look for in your entry
If you're entering a dashboard for Product Experience or Interaction Design, the screens that will do the most for you are not the populated overview. Send the empty account, the dense table with real names, the partial-failure state, and a short recording of someone completing the main task from a cold start.
Then say, in one line, what question the default view answers. If you can't, that's the thing to fix before entering — and it will improve the product more than the entry. Submit a project when you have it.
Configurability without chaos
Every dashboard eventually faces the same pressure: different roles want different things above the fold. The two common answers are both wrong — one dashboard for everyone (serves nobody) or a drag-and-drop widget canvas (serves the 3% who configure it and abandons the rest).
What scores well is a middle path: a strong opinionated default per role, a small number of prepared views, and the ability to save a view and make it your default. The user's first decision is a choice between good options, not a blank grid.
Alerts, digests and the attention budget
A dashboard is often paired with notifications, and this is where products quietly lose trust. Every alert spends a little of the user's attention; once spent, it's gone for the important one.
- Threshold alerts4Only for things a person would act on today. Everything else is a digest.
- Digests4Batched, scheduled, scannable — with the one thing that changed at the top.
- In-product badges2Cheap to add, expensive in aggregate. Count them across the whole product.
- Email per event1The fastest way to train someone to filter you.
Exports are part of the design
People export from dashboards for one of two reasons: because they need the data elsewhere, or because your dashboard can't answer their question. The second is a design finding, and it is measurable — track what gets exported and you have a prioritised backlog.
Either way, make the export good: the columns as shown, the filters applied, sensible dates, a filename that says what it is. A CSV named export(3).csv with raw column keys is a small betrayal at the end of an otherwise careful product.
Questions we get asked
How many metrics should a dashboard show?
As few as answer the question it exists for — often three to five above the fold. If your dashboard has twelve equally weighted cards, you have a data display, and the hierarchy decision is still ahead of you.
Should dashboards be customisable?
Give a strong default per role, a few saved views, and the ability to set a default. Full drag-and-drop canvases serve the few who configure them and quietly fail everyone else.
What's the biggest dashboard mistake?
Designing against tidy demo data. Production has long strings, missing values, 50,000 rows and three currencies — and editors will ask for an account that looks like it.
How do we score well on performance for a data-heavy screen?
Render the shell first, stream each panel independently, reserve space so nothing shifts, and show cached values with a timestamp while fresh ones load. Perceived speed is a design decision before it's an engineering one.
Assessed work in business software
Chosen by score from the work the editors have assessed, not by hand.
- saas
- dashboards
- b2b
Written by
The UXE editors
The team that selects the work, uses it, scores the seven criteria and writes every verdict on this site.
Keep reading
Judging & criteria6 min read
Digital product design awards: a guide for SaaS and product teams
Why most digital product design awards are built for marketing sites, what a product team should look for instead, and how to enter a SaaS product so we see the work that actually took the year.
The UXE editors
Craft & practice5 min read
Information architecture that scores: labels, depth and the way back
Information architecture is one of the seven UXE criteria. What editors test — labels people would use themselves, depth that matches the task, search that works, and a way back from anywhere.
The UXE editors
Craft & practice5 min read
Performance is a design decision, and we score it like one
Speed is one of the seven UXE criteria. Why perceived performance is a design problem rather than an engineering one, what field data editors look at, and the five design choices that make products feel fast.
The UXE editors