Metabase Review

Founded 2014 in San Francisco. Profitable. Self-hostable Docker container, or Cloud Starter at $85/month. More than 50,000 companies on the install base. Those four numbers say survivor.

The strategic call is who owns analytics in your org. If engineering owns the data warehouse, Metabase is the obvious pick — it sits next to Postgres without explaining itself to security. If marketing or finance owns the dashboards, Tableau or Power BI ship a polished sales motion Metabase doesn't try to match.

Metabot, the natural-language query layer, is the new competitive front against Hex and Sigma. The catch is the BI category is reshaping fast around AI-native entrants, and Metabase's lead is in trust, not novelty. Pilot it on three internal dashboards for 60 days. If your analysts stop exporting to spreadsheets, you've found your standard.

Look at Metabase Models. They sit between raw warehouse tables and end-user Questions, with documented columns, types, and joins. That's a semantic layer in everything but the marketing word. Compare LookML: more powerful, more verbose, harder to onboard. Metabase Models hit the 80% case at 20% of the LookML learning curve.

The X-Ray feature is the entry-point that makes self-serve analytics actually self-serve. Click any column, get a generated dashboard of distributions, time series, and outliers. That's strategic onboarding for non-technical stakeholders, and the design choice says someone on the team has watched real users freeze in front of a SQL editor.

The ceiling shows up at scale. Metabase doesn't replace dbt for transformation, doesn't replace Looker's explore-as-API for embedded enterprise analytics. The lock-in lives in saved Questions and Collections, both exportable as JSON. Sensible architecture for an org that wants BI without BI vendor capture.

Cloud Starter: $85/month for up to 5 users. Cloud Pro: $500/month entry. Enterprise on contact-sales. Self-hosted Community: $0 in license.

50 users on Cloud Pro × 12 = $6K/year baseline. Add seat overage at category-norm 30% over 24 months. Year 3 lands closer to $9-10K. Compare Looker at $5K-12K per user per year — different category, different audience. Compare Tableau Creator at $75/user/month: same 50-user math is $45K/year. Metabase is 5-7x cheaper at the same headcount.

The self-host path is the real wedge. Community Edition runs on a $50/month VM for a small team. The catch is operational FTE — call it 0.25 FTE of data engineering at $150K loaded cost, or $37K/year hidden. Honest math still puts self-hosted under $50K/year all-in. Cloud Pro is cheaper if you don't already have a data team.

First sprint reality: the Question builder is the surprise. You stack steps — filter, summarize, group by, custom expression — and each step shows the resulting data inline. That's the workflow Tableau buries behind drag-and-drop guesswork and Looker hides behind LookML. Metabase makes it visible.

The dual-mode design holds up. When the builder hits its ceiling — window functions, complex CTEs, pivoted aggregates — you flip to Native Query and write SQL with autocomplete from your schema. Compare Mode: SQL-first, harder for non-SQL stakeholders. Compare Looker: explore-first, harder for analysts who think in SQL. Metabase doesn't make you pick.

The friction is dashboard polish. Dashboards on Metabase look like dashboards from 2019 — functional, not delightful. Compare Sigma's spreadsheet-style live blends or Hex's notebook polish. Both ship a more modern feel. Metabase is the analyst's tool, not the executive's pretty deck.

You can tell Metabase was built by people who use it. The Question builder shows you the rows after each step instead of just promising they're filtered. Empty states are real, not screenshots. The X-Ray feature is genuinely useful instead of demo theater.

What hurts is the dashboards. They work, they render, they share by URL or embed. But compared to Sigma or Hex, the visuals feel like 2019. Nothing makes you want to screenshot it for Slack. For $85/month on Cloud Starter, fine is honest. For an exec who expects Tableau-pretty dashboards, the gap shows.

Mobile is read-only and basic. For analytics that's normal — nobody builds dashboards on a phone — but if your CEO only checks metrics on mobile, plan for it. The product is solid where it counts and slightly tired where it doesn't.

Founded 2014. Profitable. AGPL open-source. More than 50,000 install base. Most BI startups from that vintage are gone — Periscope absorbed into Sisense, Mode into ThoughtSpot in 2023, Looker into Google in 2019. Metabase stayed independent and shipping.

What worries me isn't what they say. It's the AI-native cohort — Hex, Sigma, the LLM-first BI wave. Metabot is the answer, but it's AI added to a 2014 product, not built around AI from day one. Incumbents who add AI on top often lose to ones who reorganize around it.

The exit story is honest. AGPL self-hosting means you keep your data, your Models, your Questions — all exportable. Compare Looker where LookML is the lock-in. That portability is why I'd still bet on it for the next 24 months.