Dashboards
Dashboards are single-screen displays that show a small set of live or regularly refreshed metrics, so a specific audience can check status, spot exceptions and decide what to do next without building a report.
What are dashboards?
A dashboard is the last layer of an analytics stack: it reads from modeled data and turns a handful of numbers into a status check. Underneath sit a source system, a pipeline, usually a warehouse, and a semantic layer that defines each metric once.
Most fall into three types:
Operational: minute-to-minute status, such as orders in the queue or tickets breaching SLA.
Strategic: weekly or monthly trends against targets for leadership.
Analytical: filterable views that let an analyst slice by region, channel or cohort.
Which type you need decides the refresh rate, the level of detail and who the default viewer is. A useful page also states its own limits: a visible last-refreshed timestamp, the definition behind each metric on hover, and a named owner to ask when a number looks wrong.
How do you design a dashboard people actually use?
Start from the decision, not the data: every tile should answer a question someone acts on. A practical build order:
Name the audience and the two or three decisions they make with it.
Pick the KPIs behind those decisions and write each formula down, e.g. conversion rate = orders ÷ sessions.
Put the headline numbers top-left, with the comparison (vs target, vs last period) beside each one.
Add drill-downs and filters below, not more tiles.
Set the refresh cadence to match the decision speed: hourly for operations, daily for most marketing.
Keep a first screen to roughly 5 to 9 metrics; beyond that, split it by audience.
What do dashboards look like in practice?
A first dashboard usually replaces a weekly spreadsheet ritual, and its value shows up as a shorter, calmer meeting. Picture a 60-person direct-to-consumer skincare brand that reviews performance in a Monday meeting built on six exported spreadsheets. Say the ops lead spends most of Friday stitching Shopify orders, Meta and Google Ads spend, and Klaviyo revenue together.
The team loads those sources into BigQuery with Fivetran, models revenue, CAC and repeat rate in dbt, and builds one Looker Studio page per audience: a founder view with five headline KPIs and an ads view with spend and ROAS by campaign. In this example, the Friday stitching disappears and the Monday meeting starts from the same numbers for everyone, so the debate moves from whose spreadsheet is right to which campaign to cut.
Two months in, the team adds an alert rather than another tile: if blended ROAS drops below 2.0 for three days, the ads owner gets a Slack message. The page stays small because new questions go to ad hoc analysis first and only earn a tile once they recur.
When do dashboards fail, and when should you not build one?
A dashboard fails when nobody owns the decision it was built for. Common failure modes:
Metric sprawl: 40 tiles and no hierarchy, so nothing stands out.
Silent definition drift: two pages show different revenue because filters differ.
Stale data with no freshness label, which erodes trust fast.
Slow queries on raw tables; pre-aggregate anything a page loads often.
Skip it for one-off questions; an ad hoc report or a single SQL query is cheaper than a page nobody revisits after the first month.
Which tools are used to build dashboards?
The tool matters less than the data model behind it, but the main options split cleanly by team and budget.
Power BI: strong in Microsoft-heavy companies, priced per user.
Tableau: deep visual analysis and large extracts through its Hyper engine.
Looker and Looker Studio: Looker adds a governed modeling layer (LookML); Looker Studio is free and quick for marketing data.
Metabase and Apache Superset: open-source options for self-hosting.
Grafana: built for time-series and infrastructure monitoring.
Custom React, Dash or Streamlit apps: when the layout, screen size or interaction model is unusual.
A scorecard is a narrower cousin that compares results to targets; see scorecards for the difference.
Dashboards vs reports
Dashboard | Report | |
|---|---|---|
Purpose | Monitor status and spot exceptions | Explain results in depth |
Refresh | Live or scheduled, always current | Snapshot at a point in time |
Length | One screen, 5 to 9 headline metrics | Many pages, tables and commentary |
Interaction | Filters and drill-downs | Mostly read-only |
Best for | Recurring decisions | Audits, board packs, one-off questions |
How does VIZIO AI build dashboards?
VIZIO AI treats the dashboard as the visible end of a pipeline and engineers both for the speed the audience needs. For a UK cybersecurity research organization, the Growth Studio built a real-time ransomware monitoring dashboard on a Python and Apache Kafka pipeline feeding Tableau Hyper, with partitioning and pre-aggregations so 500,000+ records render in under one second and new incidents appear with under 60 seconds of latency.
For a Minnesota furniture and lighting maker, the DX Studio built a React 4K signage dashboard fed by Shopify and order data through AWS Lambda and Firebase, which increased operational data visibility by 85%. Where to start: list the three decisions the page must support before choosing a tool.
Dashboards in VIZIO AI case studies
- Real-Time Ransomware Monitoring ›A Kafka-to-Tableau monitoring dashboard renders 500,000+ records in under one second and shows new incidents with under 60 seconds of latency.
- 4K Screen-Compatible Real-Time Sales ›A React dashboard for 4K signage combined Shopify, order and calendar data and increased operational data visibility by 85%.