Designing data-dense dashboards: role-based views that surface the next action
Most dashboards are built from the data outward. Someone lists everything the system can measure, a designer arranges it into a grid of charts, and the result is technically complete and practically useless. The person looking at it still has to figure out what matters and what to do.
I worked on dashboard products for Communications Service Providers (CSPs and ISPs) at Plume from 2018 to 2024, during eight years at the company (2016–2024) leading brand and then design, and as Head of Design from 2021 to 2024. The lesson I’d put first: a dashboard should answer "what should this person do next?" Density is fine. Data-dense UI is often exactly right for experts. What isn’t fine is density without hierarchy.
The problem we were solving
Before platforms like Plume’s, ISPs could see almost nothing inside their subscribers’ homes. They knew how much speed they were serving and how much bandwidth was used, and not much else. They found out about problems when support lines lit up, often for WiFi issues that weren’t their service at all: interference from neighbors, a badly placed device, a weak radio in someone’s laptop.
Plume’s platform could see home network health all the way to the individual device. Our job in design was to turn that visibility into tools: customer support portals, web dashboards and advanced ISP utilities that gave operators awareness and control they’d never had. Those tools had to sit alongside a calm consumer app and still feel like one company. The Plume web dashboards case study has more on the scope.
Start from the call, not the screen
The best dashboard brief I know is a single sentence: who is looking, and what must they be able to do in thirty seconds?
For a support agent on a live call, the answer might be: understand whether this is a home WiFi problem or a service problem, and tell the customer what happens next. For a network operations team, it might be: notice that something is going wrong across many homes before the call volume tells them. For leadership, it might be: see whether the service is getting better or worse this month.
Same data, completely different jobs. Starting from the job keeps the design from becoming an inventory.
Role-based views
Different roles need different views of the same truth. This is where today I’d go further than we could then, and design each role’s view around one recommended next action:
- Frontline support: a calm view, a few signals, plain-language likely causes, the evidence one click away, and an escalation path that carries everything already learned, so the next person never starts from zero.
- Network operations: a dense view built for scanning: patterns across many homes, emerging incidents, and the ability to drill from fleet level to a single device.
- Executives: trends, not incidents. Is the experience improving, and where?
The data model underneath stays the same. Permissions and density change. One component set can run at different densities, with role permissions deciding which modules appear.
Hierarchy under density
Experts don’t need less data. They need data arranged so the important thing is obvious and everything else is quiet. The approach I’ve used for years, which I wrote about long ago in Why Great UIs Run on Grey, still holds: let neutral tones carry structure, let content carry contrast, and save one strong color for the action that matters.
In practice for dashboards:
- One loud thing per view. If everything is highlighted, nothing is.
- Group by question, not by data source. "Is this home healthy?" is a better module than "RF metrics."
- Make comparisons direct. Put the number next to its normal range, not on a separate chart.
- Let detail live behind a click. Summary on the surface, evidence on demand.
Designing the drill-down
Dense dashboards live or die on how people move from overview to detail. The overview tells you something is wrong; the drill-down tells you why and what to do. Get the path wrong and experts open five tabs to answer one question.
The principles I follow:
- Keep context when you go deeper. When someone clicks from a fleet-level alert into a single home, the reason they clicked should stay visible at the top of the detail view.
- Make the way back obvious. Breadcrumbs or a persistent summary beat a browser back button that loses filters.
- Show the same entity the same way everywhere. A device, a home or an incident should look identical on every screen it appears, so people recognize it instantly.
- Put evidence next to conclusions. If the interface says "likely interference," the supporting signals should sit right beside the claim, not on another tab.
Plume IQ, our monthly published data insight, was good practice for this. Each month we had to find the one insight worth showing and make it readable in seconds. That same discipline, one clear point with evidence close by, carried into the larger dashboard work that followed.
States that matter
Dashboards are judged on their worst days. A dashboard that only works on clean data fails in the first week of a real outage. Today I’d test prototypes against replayed real incidents, and design these states as carefully as the default:
- Loading that shows structure immediately, so people know where to look.
- Stale data, clearly labeled with how old it is. Silent staleness is the most dangerous state in operations software.
- Partial outage, where some data is missing and the UI says so instead of showing zeros.
- Empty, which should explain what will appear and why it isn’t there yet.
One system across consumer and enterprise
Consumer apps and operator consoles often look like they come from different companies. At Plume we refused that split: the operator was looking at the same product the homeowner used, just from the other side. We ran the consumer app, the dashboards, email and brand as atomic systems with shared foundations. The Plume app and Plume brand case studies show the consumer side.
For a service provider, the dashboard is the brand, the screen their teams look at all day.
When a dashboard becomes agentic
The natural next step for operations dashboards is to suggest the next action, not just display the state. "This home’s issue looks like interference; here’s the evidence; here’s what to try." That’s useful only if the suggestion shows its reasoning and the person can override it easily. The patterns for that, plan, show its work, hand off, are in my agentic UX patterns catalog.
A dashboard’s job isn’t to show everything the system knows. It’s to help one specific person do the right thing next. Design from that, and density stops being the enemy.
Related Product and brand as one system: what eight years at Plume taught meConnected hardware: designing the box, the app and the brand as one launchDeep-tech brand and product: making an invisible product visible