Dashboard design best practices: 14 rules from real product dashboards
By Ingot UI13 min read
A good dashboard answers one question per screen, puts the three to five numbers that answer it in a KPI row at the top, explains each number with one well-chosen chart, and lets people act on the rows behind it without leaving the page. The rest of the rules below (states, density, colour, tokens) exist to keep that promise as the product grows.
Summarize withChatGPTClaudePerplexity
On this page
- Start with the job, not the layout
- The first screen
- Charts that explain the numbers
- Tables, drawers and actions
- States, density and trust
- Navigation and page structure
- Accessibility on data-heavy screens
- Dashboards on small screens
- Build it faster with an AI agent
- A checklist before you ship a dashboard page
- Common dashboard design mistakes
- Where to start if you are building one now
These rules come from rebuilding real product dashboards as 9 design systems, among them audience analytics (Loopo), email marketing (MailPulse), AI agent governance (Auton), workflow automation (Axon), security (Veriq) and business admin (Vantage). Together they hold 578 source screens and 2,652 components, so every rule below has a real screen behind it, and most link to one you can open.
This is a long guide. If you only have five minutes, read rules 1, 2, 5, 9 and 12. They fix most dashboards we see.
Start with the job, not the layout
1. One question per screen
Write the question the screen answers before you draw anything. “How is revenue doing this month?” is a screen. “Revenue, audience, campaigns and settings” is four screens squeezed into one. When a team can’t agree on the question, the result is a wall of equal-sized cards that nobody reads.
Each page template in our systems was designed this way. The Loopo revenue page answers “where does the money come from?”, the Loopo audience page answers “is the list growing, and from where?”, and nothing on one page tries to answer the other’s question.
2. Decide who opens it, and how often
A founder who checks the overview once a morning needs a different screen from an operator who lives in a queue all day. The first wants a summary and a few trends; the second wants a dense table, keyboard shortcuts and bulk actions. Most dashboards need both kinds of page, and the mistake is designing every page for one of them.
| Visitor | Opens it | Needs on the first screen | Example page |
|---|---|---|---|
| Founder or manager | Daily or weekly | Three to five KPIs, one trend, what changed | Vantage home |
| Analyst | Several times a week | Filters, comparisons, export | Loopo analytics |
| Operator | All day | A dense queue, status, bulk actions | Axon review queue |
| Admin | Rarely | Clear settings, safe destructive actions | Veriq settings |
The first screen
3. Put a KPI row at the top, and keep it short
The KPI row is the most-read part of almost every dashboard. Three to five cards is the useful range: fewer feels empty, more turns into a table of numbers without hierarchy. Each card needs the value, a label people understand without a glossary, a comparison (vs last period or vs target), and the period it covers.
Our systems have 161 KPI and stat components. Compare the Loopo KPI strip, which carries a metric selector, with the Axon run stats, which show throughput, p95 duration, failure rate and queue depth for an operations team. Same pattern, different questions. For the details of a single card, see our KPI card design guide.
4. Show the change, not just the number
“12,480 subscribers” means little on its own. “12,480, up 4.1% vs last month” tells people whether to relax or dig in. Use a delta chip with a direction icon as well as a colour, because red and green alone fail for colour-blind users and on poor screens. Say what the comparison is against; “vs last 30 days” and “vs same month last year” lead to different decisions.
5. Let the first screen answer the question without scrolling
On a 1440 × 900 laptop, the page header, KPI row and the main chart should all be visible before the first scroll. Everything below the fold is supporting detail: breakdowns, tables, recent activity. If the main chart sits below a row of decorative cards, people will never see it.
Charts that explain the numbers
6. Pick the chart from the question
Most dashboard charts answer one of five questions: how did it change over time (line or area), how do these compare (bar), what is it made of (stacked bar, or a donut for three to five parts), how is it distributed (histogram), and where are people dropping out (funnel). When a chart is chosen for looks, people have to work out what it says.
| Question | Chart | Real example |
|---|---|---|
| How did it change over time? | Line or area | Loopo transaction value trend |
| How do these compare? | Bar | Loopo weekly bars |
| What is it made of? | Donut or stacked bar | Vantage plan mix |
| Where do people drop out? | Funnel | Vantage funnel |
| Do cohorts stay? | Retention grid | Vantage retention grid |
There are 396 chart components across the 9 systems, and the charts page groups them by type. For a full decision guide, read dashboard chart types.
7. One idea per chart, with a title that says it
Give each chart a title that states what it shows and a subtitle with the period: “Revenue by channel, last 30 days”. Two y-axes, more than five colours, and a legend people have to match by colour are the usual signs that one chart is trying to be three. Split it.
8. Tooltips should finish the sentence
A good tooltip shows the exact value, the date and the comparison, so people never have to read the axis. The Loopo comparison tooltip puts the month, your value and the benchmark for similar businesses in three short lines. Keep tooltips readable at a glance: two or three lines, numbers aligned, units shown.
Tables, drawers and actions
9. The table is where the work happens
Charts get attention, tables get used. Once someone sees a number move, they want the rows behind it: which customers, which campaigns, which findings. Design the table as carefully as the chart: a toolbar with search and filters, columns in order of importance, status chips, right-aligned numbers, a row menu, and an empty state for when filters match nothing.
The 9 systems include 320 table and list components. The Vantage customers table and the Veriq secrets table show two very different densities. Our data table design guide covers columns, states and bulk actions in detail.
10. Open details in a drawer, confirm in a modal
Clicking a row should open a drawer that keeps the list visible, so people can move through records without losing their place. Use a modal for short, blocking decisions: confirm a delete, set a target, name a segment. The rule of thumb: if people need to compare the detail with the list, it’s a drawer; if they must decide before continuing, it’s a modal. More in modal vs drawer.
11. Keep actions next to the data they change
Put “Pause flow” on the flow, “Request a fix” on the finding and “Change plan” on the subscription. A global “Actions” menu in the page header forces people to remember what they selected. Bulk actions belong in a bar that appears when rows are selected, like the Loopo bulk action bar.
States, density and trust
12. Design the empty, loading and error states first
A new account has no data, a slow query shows a spinner, an integration fails. These are the states new users see most, and they decide whether the product feels finished. An empty state should say what will appear here, why it’s empty and the one action that fills it. Loading states should keep the layout steady with skeletons in the shape of the real content. Errors should say what failed and what to do. See empty state design.
13. Choose density on purpose
Overview pages can breathe: larger numbers, more space, fewer elements. Operational pages should be dense: 36–44 px table rows, compact filters, small type that is still legible (13–14 px for body text in tables is common). The worst outcome is the middle: big padding on a page people scan all day, or cramped cards on a page people open once a week.
Compare the airy Vantage business page with the dense Axon runs and logs page. Both are right for their visitors.
14. Use colour for meaning, and put it in tokens
Pick one accent colour for primary actions and selection, a neutral scale for everything structural, and a small set of status colours (success, warning, danger, info) used only for status. Charts get their own palette. Then store all of it as tokens, not hex values in components, so changing the brand or adding dark mode is a token change instead of a rewrite.
Check text contrast against WCAG 2.2 contrast guidance: 4.5:1 for normal text and 3:1 for large text. Pale grey labels on white are the most common failure on dashboards. Our design tokens guide shows how the systems share one token structure with different values.
Navigation and page structure
Rules 1 to 14 are about a single page. A dashboard is a set of pages, and the way people move between them matters just as much.
- Group the sidebar by job, not by data type. “Audience, Campaigns, Automations, Analytics” reads as a workflow. “Tables, Charts, Lists” reads as a database. The Loopo sidebar groups pages under workspace and audience headings.
- Keep the sidebar to one screen. If it needs to scroll, merge pages or move rarely used ones (billing, integrations, settings) to the bottom.
- Use tabs inside a page for views of the same data. All, live, draft and paused are tabs with counts, as in the Axon tabs; a separate page for each is not needed.
- Give power users a command palette. One shortcut (usually ⌘K) to jump to any page or record. The Loopo command palette and Axon command palette are two takes on it.
- Repeat the page header pattern everywhere. Title on the left, date range and primary action on the right, the same on every page, so people never look for them.
Browse every navigation pattern in the navigation category and every page header in page headers.
Accessibility on data-heavy screens
Dashboards are some of the hardest interfaces to make accessible, because so much meaning lives in colour, position and tiny text. The basics go a long way:
- 1Never use colour alone. Every status chip has a label, every delta has an arrow, every chart series can be told apart by label or pattern.
- 2Charts need a text equivalent. A short summary sentence above the chart (“Revenue rose 8% in March, led by email”) helps screen-reader users and everyone else.
- 3Tables must be real tables. Use proper table markup with header cells, so assistive technology can read a cell with its column name.
- 4Everything works with the keyboard. Menus, drawers, modals and date pickers open, move and close with the keyboard, and focus returns to where it was.
- 5Respect reduced motion. Count-up numbers and animated charts should be still when the user asks for less motion.
Interactive parts in Ingot UI use Base UI primitives for focus and keyboard behaviour. Our accessibility page explains what is covered and what you still need to check in your own screens.
Dashboards on small screens
Most dashboard work happens on laptops, but people do check numbers on their phones. Decide per page instead of squeezing everything:
| Page type | On a phone |
|---|---|
| Overview | Show it: KPIs stack in a two-column grid, the main chart goes full width, breakdowns follow. |
| Queue or inbox | Show a simplified list with the one most common action; send complex work to desktop. |
| Builder or editor | Read-only view, or a message that it works best on a larger screen. |
| Settings | Show it as a single column; it is mostly forms. |
Tables are the hardest part. On narrow screens, turn each row into a card with the two or three most important fields and open the rest in a full-screen drawer, rather than forcing a horizontal scroll.
Build it faster with an AI agent
If you build with Claude Code, Cursor or Codex, most of these rules can be handed to the agent. The problem is that agents default to generic components and invent colours. Give them a design system to work from. With the Ingot UI MCP server connected, a prompt like this produces a page that follows the rules above:
Using the Ingot UI Vantage system, build app/(app)/overview/page.tsx. One question: how is the business doing this month? KPI strip with MRR, active subscriptions, ARPU and churn, each vs last month. Below it a 12-month MRR chart, then a customers table with status chips, filters, a row menu and a drawer for details. Include empty, loading and error states. Use only --fu-* tokens. Run audit_ui at the end and fix what it reports.More on this workflow in Claude Code frontend UI and Cursor rules for a design system.
A checklist before you ship a dashboard page
- 1The page answers one question, and its title says so.
- 2Three to five KPIs at the top, each with a comparison and a period.
- 3The main chart is visible without scrolling on a 1440 × 900 screen.
- 4Every chart has a plain title, a period and readable tooltips.
- 5The table has search, filters, status chips, a row menu and an empty state.
- 6Rows open a drawer; blocking decisions use a modal.
- 7Actions sit next to the data they change; bulk actions appear on selection.
- 8Empty, loading and error states are designed, not left to defaults.
- 9Density matches how often people open the page.
- 10Colours come from tokens, status colours mean status, and text passes contrast.
Common dashboard design mistakes
- Equal-sized cards for everything. When every card is the same size, nothing is important. Let the main chart take more space.
- Vanity numbers in the KPI row. “Total signups since launch” only goes up. Show numbers that can go down, so the row is worth checking.
- Charts without a decision. If nobody would act differently after seeing a chart, remove it.
- Filters that reset. Keep filters and the date range in the URL so people can share a view and come back to it.
- Settings mixed with data. Configuration belongs on its own page, not in a card on the overview.
- Default component styling. Unchanged library defaults make a product look like every other one. Change tokens first; see shadcn/ui alternatives for how.
Where to start if you are building one now
If you are designing from scratch, start from the closest real product. Each Ingot UI system comes with page templates, the components inside them, and React + Tailwind code for the ones that passed visual QA against Figma (2,636 components so far). The Loopo system is free, so you can test the approach before paying for anything.
| You are building | Start from | Pages to open first |
|---|---|---|
| Analytics or reporting | Loopo | Overview, Analytics |
| Email or marketing tool | MailPulse | See the email marketing dashboard guide |
| AI agent product | Auton | See the AI agent dashboard guide |
| Automation or ops tool | Axon | Workflows, Runs and logs |
| Security or compliance | Veriq | Vulnerabilities, Overview |
| Subscriptions or business admin | Vantage | Subscriptions, Customers |
For more inspiration across the systems, see SaaS dashboard examples. If your team builds with Claude Code or Cursor, the MCP server lets the agent use these components instead of generating generic ones.
Frequently asked questions
What are the most important dashboard design principles?
Answer one question per screen, put three to five KPIs with comparisons at the top, explain them with one well-chosen chart, give people a usable table to act on, and design empty, loading and error states. Density and colour should follow how often people use the page.
How many KPIs should a dashboard show?
Three to five in the top row. If you need more, the page is probably answering more than one question and should be split, or the extra numbers belong in a table below.
What size screen should I design a dashboard for?
Design for a 1440 px wide laptop first and make sure the header, KPI row and main chart fit above the fold at about 900 px tall. Then check 1280 px and a large monitor, and decide which pages truly need a mobile layout.
Should dashboards have dark mode?
Many users expect it, especially in developer and operations tools. Build colours as tokens from the start so dark mode is a second set of values rather than a redesign.
Is it better to buy a dashboard UI kit or design from scratch?
A kit saves the weeks spent on tables, drawers, states and charts, which every dashboard needs. Design from scratch only the parts that make your product different. Check that the kit has code that matches its Figma file, or you pay for the gap later.
About Ingot UI
Ingot UI is a library of nine dashboard design systems built from real product screens. Every generated React component is checked against its Figma original before it ships. How we write guides.