Tableau Conference 2025
Your dashboard is not slow. Your data is
I gave this talk on the main stage at Tableau Conference 2025 in San Diego. It is about the six things an analyst can ask a data engineering team for, and why nothing changes until you know how to ask. Here it is written down, in about three minutes of reading.
The recording
Six data engineering tips for Tableau developers
Thirty nine minutes, recorded live and published on Tableau's own channel. Nothing loads from YouTube until you press play.
Nothing is requested from Google until you press play. If you would rather not, the talk is also on YouTube.
Tableau Conference 2025, San Diego. Recorded by Tableau.
The six tips
None of these moves a dashboard from sixty seconds to five on its own. They add up, which is the point. Small changes, in the right place.
Ask for the right thing
"I need user data in Tableau" gives an engineer nothing to build. Say which data, which platform, how often it refreshes and which fields you filter on, and you get an answer instead of a question. You are the owner of the dashboard, so you are the one who knows this. If you do not ask for it, you will not get it.
Understand the data model
A very wide table with orders, invoices and salespeople mixed into it carries duplicates and breaks in ways that are hard to see. Splitting it fixes that, but it moves the joins into Tableau, where they cost you every time the workbook opens. Split it, pick only the tables you need, then have the result built as a table in the database.
Index what people actually filter on
Without an index the database reads every row to find one name. With one it halves the search at each step, so ten times the data costs almost nothing extra. Pick the filters your users really use, and prefer an ID over a long URL or a country field. You cannot index everything, so this is a choice, not a switch.
Load less data in the first place
If you only ever report on one product line, there is no reason to pull fifty million rows covering all of them. Split the table. The same works for time: the last thirty days in one place, the last year in another, five years in a third. A board dashboard about this month should not be reading a decade.
Do the maths in the database
Fifteen million rows and seven rows answer the same question if the question is a monthly total. Aggregate before it reaches Tableau. Business logic belongs there too, because logic rebuilt in three workbooks becomes three slightly different numbers, and then you are in a meeting about which one is right.
Use the right platform
A transactional database is built for writing quickly, one row at a time. A warehouse is built for reading a lot at once. Running analytics on the first one is slow by design, and no amount of work in the workbook fixes it. You may not get to choose today, but a reason on record is how these things eventually change.
The part that makes the other six work
Your data engineering team is not opening your dashboards. Nobody is coming to tell you a filter is expensive.
Where this goes next
Doing this once for your own dashboard is a fix. Doing it so the next analyst does not repeat it is an operating model.
- Ownership, written downThe talk keeps arriving at the same place: someone has to own the data, and it has to say so somewhere. That is the difference between one fast dashboard and a team that stops rebuilding the same table.
- The cheat sheetA short read on data products, ownership and contracts. Free, with nothing to fill in.
Next step
Want me to look at yours?
Bring one slow dashboard. Thirty minutes is usually enough to say whether the problem is the workbook or the data underneath it.
Thirty minutes, straight to the problem. No deck, no pitch, and a written summary afterwards either way.