The BI Workbench

Your Tableau dashboard is slow. Check these five things first.

The request arrives as "the dashboard is slow", which is not a diagnosis. It could be the warehouse, the query, a calculation, the number of marks being drawn, or the fact that eleven people are refreshing it at 9am on a Monday.

People tend to start where they can see something — deleting a chart, thinning a colour palette, adding an index — and that is usually the wrong end. Work through this order instead. It goes from the most expensive and most common cause to the cheapest and rarest.

First, measure. Don't guess.

In Tableau Desktop: Help → Settings and Performance → Start Performance Recording, then load the slow view. You get a timeline split into Executing Query, Computing Calculations, and Rendering.

That split is the whole diagnosis:

On Tableau Server or Cloud, the view's own Performance tab gives you a version of the same thing, which is what you need when the complaint comes from someone whose laptop isn't the problem.

1. A live connection where an extract would do

Live means every filter change, every parameter toggle, every page load goes back to the source system and waits. Extracts move that cost once, at refresh.

The question isn't "which is better" — it's whether anyone would be harmed by data that is one refresh old. In most operational reporting the answer is no, and the extract is several times faster for the person clicking.

Where to be careful

Extracts are not free: they need a refresh schedule, they duplicate storage, and a stale extract is a silent correctness bug. If you choose an extract, put the refresh time on the dashboard so the reader knows how old the number is.

2. The custom SQL view is doing too much

This is the one I'd check first on any inherited workbook. A published data source built on a forty-line query with five joins, several CASE statements and a DISTINCT — executed fresh for every sheet that uses it.

Things that help, roughly in order of payoff:

3. Calculations running at the wrong level

Row-level calculations evaluate for every row before aggregation. On a few hundred thousand rows that's fine; on tens of millions, a handful of them can dominate the timeline.

Two habits that pay off:

Table calcs are a separate case: they're cheap to compute but expensive to compute often, because they depend on what's in the view. A table calc across a wide table responds slowly to every filter change, and no amount of tuning makes it feel instant.

4. Too many marks

Tableau draws marks. A scatter plot with 300,000 points is 300,000 things to draw, and it will feel slow on a good laptop and unusable on a shared one.

Common offenders: a text table with tens of thousands of rows, a map with every individual record plotted, a line chart with a mark per transaction rather than per day.

The fixes are boring and effective — aggregate to the grain the reader cares about, bin or hexbin a dense scatter, and put the detail behind a filter or a drill-down rather than on the opening view.

5. Filters that re-query everything

Every quick filter that lists values runs its own query to populate. Ten of them on one dashboard is eleven queries before the reader has clicked anything.

The order matters more than the tricks

Most slow workbooks I've been asked about were slow in the data layer, not in the visual layer — a live connection to a view that joined eight tables and returned columns nobody needed. Deleting a chart from that workbook changes nothing measurable.

Record it, read the split, fix the biggest bar, record it again. The second recording is the part people skip, and it's the only thing that tells you whether the change you made was the change that helped.

Reviewed 19 November 2026 against current Tableau behaviour. If something here doesn't match what you see in your version, tell me — corrections get fixed and the date above changes with them.