Picking a BI tool: Looker Studio, Lightdash, Power BI or Metabase

A business intelligence tool pulls data from a source, does some arithmetic with it and puts the result on a screen. Charts, tables, filters, a dashboard you share with your team.

The tool usually stores nothing itself. It sends a query to your warehouse or your ad account and draws whatever comes back. That is why two dashboards side by side can disagree: they ask the same source a different question.

For a marketing team the question is rarely whether you need a BI tool. The question is which one, and what you put underneath it.

Free, connects to GA4, Google Ads and Search Console in about two minutes, and everyone on your team can use it. For a ten-chart report on a single source it does the job.

It starts to hurt as your dataset grows. Every time someone opens the report, a query goes to the source. With BigQuery you pay per byte scanned, so a dashboard that stays open in a browser tab turns into a line on your bill. The extract connector caches a copy that loads faster, but it caps the size of each extract, so large tables will not fit.

The second limit is governance. Calculated fields live per report, sometimes per chart. If someone edits the formula behind "revenue", there is no review, no diff, and no easy way to see who changed what and when.

Open source, and built on top of dbt. You define your metrics in the YAML that sits next to your dbt models, in the same repository as your transformations.

That is the difference that matters. A metric like revenue is defined once, in code, with a pull request around it. Every chart using it calculates identically. Change the definition and it changes everywhere at once, with the git history explaining why.

The cost is that you need dbt. Without it, Lightdash is a detour. With it, you stop writing your metrics down twice.

We run our own dashboards on Lightdash on top of BigQuery, for exactly that reason: the definition of a number belongs with the transformation, not with the chart.

Power BI is strong when the company already runs Microsoft. Permissions through Entra ID, reports inside Teams, Excel users who feel at home on day one. Its model, with DAX and a semantic model inside the file, is powerful and takes time to learn. The desktop application you build in runs on Windows, which is a practical objection in a Mac team.

Metabase is the low-threshold option: open source, set up in an afternoon, and colleagues without SQL can ask their own questions through the query builder. For a team that wants to start looking at its own data, it is the fastest route to something usable. For heavier modelling you hit its limits sooner.

When three dashboards show three different revenue figures, it is rarely because the tool miscalculates. It is because nobody wrote down what revenue means.

Does an order count on order date or invoice date? With or without VAT? Shipping included? Do you subtract returns, and if so, in the month of the order or the month of the return? Four reasonable answers, four different charts.

Without one place where that choice is recorded, every analyst builds their own version in their own report. That place is your warehouse with a modelling layer on top, or the semantic layer of your BI tool. What it cannot be is a calculated field in a dashboard that one person created once.

• Start with definitions, not with the tool. Write down how your five core metrics are calculated, including VAT, returns and the date you attribute them to.
• If everything runs on Google and your report is small, Looker Studio is enough. Keep building in it until it crawls and you have only moved the problem.
• If you use dbt, look at Lightdash. Your metrics then live in the same repository as your models, with review and history attached.
• If the company runs on Microsoft, Power BI wins on permissions and adoption, even when another model is technically nicer.
• If you want to start without a data team, put Metabase on your warehouse and see in a quarter what your team actually opens.
• Measure what your dashboards cost. BigQuery reports bytes scanned per query, and a busy dashboard without caching is usually the biggest item there.