Video |

Tableau | Data visualisation | Data prep | Analytics

Tableau 2020.2: The new data model in Ask Data

Ask Data now speaks the new data model, so you can ask questions of related tables without ever building a join.

Tim Ngwena •6 May 2020 •1 min read •4 min watch •2.7K

Part ofWhat's new in Tableau 2020.2

Watch on YouTube

Tableau 2020.2 lets data sources built on the new data model be used inside Ask Data with full functionality
The new data model uses relationships (the 'noodle') rather than physical joins, keeping Orders and Returns as separate logical tables on a single connection
Tableau auto-suggests matching fields like Order ID when you relate two tables, with optional performance settings you can usually leave alone
Taking an extract before publishing lets you optimise the file ahead of pushing it to Tableau Server
Ask Data preserves dimension and measure distinctions per table, but natural-language questions about relationships may need rewording to get the view you expect

0:00In Tableau 2020.2, Tableau have added the

0:03ability to use data sources built with a

0:06new data model inside of askdata.

0:09So you get all the functionality that you'd

0:11expect.

0:12I'm going to show you how to do this from

0:13scratch.

0:14So first I'm going to connect to an Excel

0:15file.

0:16I'm going to use the superstore sales files

0:18that comes inside of your Tableau

0:20repository for data sources when you

0:22install Tableau.

0:24That's normally in your documents folder,

0:26my Tableau repository, data sources.

0:28Then you can see here 2020.2 and then the

0:31version.

0:32So that's pretty much how you get to that

0:33file.

0:34Now when I open this in the older versions

0:36of Tableau, of course, I drag in orders

0:39like I just did.

0:40But then what I normally do is do a left

0:42join on returns based on order ID.

0:45Now if I drag that now, you'll see I'm

0:46using the data model.

0:48You can see the noodle that it's affection

0:50ately termed as.

0:51If I drop that into the view, Tableau

0:53actually makes a suggestion here that order

0:56ID on orders equates to the returns on, equ

0:59ates to the order ID in the returns table.

1:02Okay.

1:03There's a couple of performance options,

1:05but I don't really need to change this

1:06because this is fine.

1:08I don't need to give it more information.

1:10And now I have that relationship.

1:12If I click on orders, I can see that table.

1:14Click on returns, I can see that table.

1:17There's no join being done here.

1:18It's a single connection to two logical

1:21sort of tables, but it's creating some new

1:24sort of connection as it were.

1:26And so I just have this one connection.

1:28Okay.

1:29Now the next thing I'm going to do is take

1:31an extract of this data just to sort of

1:33optimize it.

1:35And I'm just going to leave it here.

1:36Actually, I'll call this beta just so that

1:38I don't confuse my other data sources.

1:41I'm just going to go ahead and publish this

1:43data up to Tableau server.

1:45The reason I took an extract is because I

1:47actually wanted to optimize the file a

1:49little bit before I publish it up.

1:51And then the thing I want to publish it to

1:52is actually this is the 2020.2 release

1:55folder.

1:56Okay.

1:57And I'm going to call this ask data version

1:59.

2:00Okay.

2:01So what we should be able to see is the

2:03same sort of lines that you get in the data

2:05model, signaling the difference between

2:08dimensions and measures.

2:10So let me go ahead and publish that up now

2:12so that we get that up to Tableau server.

2:16And when that's finished, it automatically

2:18opens up Tableau server.

2:20So here we are with our data source.

2:22Okay.

2:23And not only that, we're already in ask

2:25data.

2:26And just for, you know, good measure, you

2:28can now see that we've got our two tables

2:30here, returns with dimensions and measures

2:33there.

2:34And you've got orders here at the top with

2:36dimensions and measures.

2:38And so you can ask, you know, the same old

2:41questions as like how many orders with

2:44returns.

2:46Let's just see what it does with that.

2:48Okay.

2:49Let's just see.

2:51You see count distinctive orders count and

2:53count distinctive returns count.

2:55It's struggling a little bit to sort of

2:57make that relationship of how they work.

3:00But this is where you can kind of play with

3:02, you know, some sort of way of wording the

3:05question.

3:07I think it's actually got this correct here

3:08.

3:09So count of orders as a count by count of

3:12order ID returns.

3:14So let's just see what this does.

3:16It might not be correct, but we can of

3:17course adjust this if it's not correct.

3:20It's sort of difficult to sort of take that

3:23sort of way of thinking and sort of adapt

3:26it very quickly.

3:28And I think this is working, but it's not

3:30built the view that I would have expected

3:32it to build.

3:33I sort of wanted something a little bit

3:34more intelligent than this.

3:36So you can see the count of orders and you

3:38can see these are return orders.

3:40And so these items here don't have a return

3:44.

3:45Right.

3:46So if we could just get rid of that, then

3:48these are your orders with return.

3:50So it kind of built the right thing.

3:51But I had to guide it down that one more

3:53step just to get the view that I wanted to

3:55see.

3:56Okay.

3:57That's pretty much it in a nutshell.

3:59Our state now supports the new data model.

4:01So by all means, start using it, start

4:03publishing update sources and get involved.

4:05I'm Tim.

4:06Thank you for watching.

4:07If you've got any feedback on this video,

4:09please get in touch in the comments below.

4:11Otherwise, I'll catch you in the next video

4:12.

4:13Take it easy.

4:14Take it easy.

4:14[BLANK_AUDIO]

Rollup

Tableau 2020.2 extends Ask Data support to data sources built on the new relationship-based data model, so you can query related tables with natural language without ever building a physical join.

Tim builds a data source from scratch by connecting to a multi-table Excel file, relating two tables (orders and returns), publishing it to Server, and then testing it in Ask Data.

The Breakdown

Connect and drag in your first table 0:12

Connect to your source and drag in the primary table as normal – this part hasn't changed from older Tableau versions.
Relate a second table instead of joining it 0:40

Rather than doing a left join, drag the second table onto the canvas to use the new data model's relationship (the 'noodle'). Tableau auto-suggests the matching key field between the two tables, and you can usually accept its performance options as-is.
Understand what a relationship actually gives you 1:02

With a relationship, each table stays a separate logical table on one single connection rather than being physically joined – you can click into either table and see it independently.
Extract before publishing 1:29

Taking an extract lets you optimise the file before pushing it to Server – a good habit whenever you're preparing a data source for wider use, not just in this data-model workflow.
Publish to Server 1:41

Publish the extracted data source up to Tableau Server; once it's done, Server opens the source directly in Ask Data.
Query the related tables in Ask Data 2:20

In Ask Data, both tables appear with their own dimensions and measures preserved, and you can ask ordinary natural-language questions across them just as you would with a single flat table.
Reword questions that touch the relationship 2:42

Straightforward cross-table questions can trip Ask Data up – it may return separate counts instead of the combined view you want. If that happens, rephrase the question (e.g. asking for a count of one table's key filtered by the other) to guide it toward the right result, then adjust the resulting view manually if it's still not quite right.

Worth Knowing

Ask Data can struggle to infer the intended relationship logic from a simple question and may need the question reworded before it builds a sensible view.
Even after rewording, you might still have to manually tweak the resulting view (e.g. removing extra fields) to get exactly what you wanted.

Use It When

Reach for this when you want business users to ask natural-language questions across multiple related tables without you having to pre-build joins or flatten the data first.

How this Rollup was made provenance & method

A Rollup is drafted by AI from the video's transcript, then reviewed and edited by Tim. Everything used to produce this one is listed below — the model, the exact prompt, and the source video — so the process is transparent and reproducible.

Transcription: On-device — NVIDIA Parakeet v3 for recent videos, OpenAI Whisper large-v3 for earlier ones. The transcript never leaves the machine or gets published.
Drafting: Claude Sonnet 5 in the cloud, from that transcript.
Prompt: The exact Rollup prompt (v2) — the full system prompt, unedited.
Source video: Watch on YouTube
Drafted: 5 July 2026 at 09:38
Reviewed & edited: 5 July 2026 at 09:41 · by Tim Ngwena

Model + prompt + video is everything you'd need to recreate a Rollup like this yourself. The one thing we don't share is the transcript.

Rights. The video and its transcript are the property of TN Media Ltd. Unauthorised use or download is prohibited. © TN Media Ltd.