Business intelligence tools compared on more than the sales demo

Business intelligence tools compared on more than a sales demo means Microsoft Power BI, Tableau, Looker, Qlik Sense and Metabase, judged on price, query behavior and lock-in. Every vendor demo works. The differences that decide your next three years show up under load, under change, and on the day you try to leave.

What to take away

Power BI, Tableau, Looker, Qlik and Metabase, and what each is

Microsoft Power BI. A semantic model plus a report layer, sold per user or as Fabric capacity. Pro lists at $10 per user per month and Premium Per User at $20. Fabric capacity is priced by SKU: F64, the tier comparable to the retired Premium P1, is listed near $8,000 a month pay-as-you-go, less on reservation.

It is the cheapest seat of the five at list, and the tightest fit if your data already lives in Azure or SQL Server. Query behavior depends on the mode. Import copies data into Microsoft's store, DirectQuery pushes it to your source.

Tableau. Salesforce's visual analytics platform. Its engine, Hyper, holds an extract, so most dashboards read from Tableau's copy rather than your warehouse. Best suited to analyst-led exploration and mixed audiences. List pricing runs $15 per user per month for Viewer, $42 for Explorer and $75 for Creator, billed annually, with a separate server or cloud tier.

Looker. Google Cloud's platform, where the data stays in the warehouse and the logic lives in LookML. Every query is generated from that repository, so the warehouse pays for the compute. Suits teams with version control habits and a warehouse they trust. Google sells platform contracts by edition rather than a list rate; deals typically start in the mid four figures per month.

Qlik Sense. An associative engine that indexes relationships across the whole model rather than answering a fixed query path. Good for finding what you did not think to ask. Qlik Cloud sells Analyzer and Professional user types plus capacity, with list rates typically from about $20 to $90 per user per month by type and term. Its lineage runs back to QlikView, the earlier in-memory product.

Metabase. Open source, self-hosted or cloud, aimed at teams who want a question answered without a build queue. It reads your warehouse directly, so compute lands on your warehouse bill. Self-hosting is free. Metabase Cloud lists a Starter plan under $100 a month, with Pro and Enterprise tiers above it. Row-level permissions are coarser than the other four.

Where the differences actually live

Three rows decide most outcomes. Query behavior sets whose compute bill grows with adoption. Version control decides whether a definition change is reviewable or just happens. The price row decides whether wider use raises the bill slowly or in a step.

Cost shape, in your own numbers

No vendor's list price tells you your bill. Build it from four variables: seats, peak concurrent readers, warehouse compute per query, and refresh frequency. Multiply seats by the per-user rate, then add compute for the queries that leave the warehouse.

A 200-seat comparison at list shows the shape. Power BI Pro at $10 per user works out to $2,000 a month. The same 200 seats at Tableau Creator rates list at $15,000. Mixing in viewer-only seats, capacity tiers and negotiated discounts moves both figures.

The trap is the shape. Per-seat pricing punishes the adoption you want. Capacity pricing punishes idle time. Warehouse-push tools move cost to a bill your data team already owns, which is easier to forecast and easier to blame.

Design the trial so it can fail

Connect to your real warehouse, not an extract. Have your own people build, not the vendor's engineers, and time how long the awkward problem takes without help. Implement the hard permission rule first, not last.

Then break something on purpose: fail a pipeline and watch what the reader sees. Silently stale is worse than visibly broken. Finish with two named users doing real work for a week. Measure time to an answer and how often they asked for help.

Lock in, and how to price it

Every tool creates lock-in somewhere. Definitions held inside the tool are the worst case, because they are the asset. Looker's LookML is portable. A Power BI semantic model is not, in practice.

Content built by hundreds of people becomes unmovable through volume alone. Embedded views create dependencies outside your team's control. Ask every vendor how a full configuration export works, then run it during the trial.

When a new tool is the wrong purchase

If your numbers disagree, or nobody owns the metrics, changing the presentation layer will not help. Write down the three complaints that started the search. If none is about tool capability, the money belongs in the modeled layer.

Decide what a tool must support before shopping: publication tiers in the three tier model and function economics in business intelligence.

The wider category is covered in analytics tools. For what the delivery layer must get right regardless of vendor, see dashboard design. For ownership rules any tool must respect, see data governance.

Common questions

How long should an evaluation take?

About three weeks of real work, spread over six. Longer and the organization loses interest. Shorter and you only see the demo surface.

Should we standardize on one tool?

For governed output, yes, because consistency is most of the value. Analysts will still explore in their own environment, and stopping that costs more than it saves.

Is open source a different decision?

The evaluation is the same. The cost shape moves from a license to your engineers' time, which is real and easy to underestimate at the start. Metabase is the free option here, and it still carries staff time.

What about the natural language features?

Test them against your own model with your own ambiguous terms. They perform well on clean demo schemas and expose how well defined your data actually is.