Tableau VIZql Data Service, Headless BI, Decoupling & Layers | Announced at Tableau Conference 2023

0:00Hey, it's Tim here. In today's video we're

0:01going to be covering a topic that has taken

0:03me a while to understand

0:04But it's gonna be VDS, visQL data service

0:07The concept of decoupling and this term

0:09that was mentioned called headless BI and

0:12then we're also going to touch on

0:13Some sort of small concepts around layers.

0:15Anyway, that's a lot to cram in. Let's get

0:18stuck in. Okay

0:19So at the Tableau conference keynote Pedro

0:21announced something called a visQL data

0:23service and he threw a few terms in for

0:26example

0:27Headless BI decoupling and so what I wanted

0:29to do is sort of really dig into that

0:32But I really can make a video immediately

0:33because fundamentally I didn't actually

0:35know what Tableau meant by the term

0:37Of course, you can go out on the internet

0:39and look at these things and sort of come

0:40to some sort of conclusion

0:42But I was actually able to get a bit of

0:43time with a few people

0:44internally at Tableau who knew a bit more

0:46about how the service has come about and

0:48Some of the thinking that's gonna come out

0:50and so I have to sort of lead with this and

0:52say that look this service has

0:54Not been announced. It's not been released.

0:56It's not in public. It won't be probably

0:58until next year

0:59But because it was in the keynote we can

1:01speculate a little bit about it

1:03So let's find out then let me show you what

1:05I think

1:05This service is going to be like let's hop

1:08into Xcalidraw where we can go through a

1:10diagram that I've prepared for you

1:12Okay, so I want to start off by setting the

1:15scene here

1:16What I want to do is essentially take you

1:18through a diagram

1:19So I've hidden the diagram and on the left.

1:22I've got two arrows one is green and one is

1:25purple now

1:25These are conceptual arrows

1:28I just want you to be absolutely clear that

1:29what you're about to see is not technically

1:32correct that sort of it's conceptually

1:34correct

1:34But it's not technically correct and so on

1:36the left here

1:37We have the journey and we have the

1:39analytical flow which starts from the

1:40bottom and goes up and then on the right

1:42you have

1:43The consumption workflow which goes from

1:45the top and goes down

1:46essentially what I'm trying to show here is

1:48that if you're an analyst building some of

1:50these assets whether it's a Tableau pulse

1:52view or

1:52Dashboard or metric whatever you're

1:54building you start from the bottom of this

1:57flow going upwards

1:58And then when you're actually using these

2:00tools when you're actually consuming this

2:02data consuming this analysis

2:03You actually go the other way around you go

2:06from the top down and that's an important

2:07sort of journey to understand

2:09And but it also means that the journeys are

2:11actually two different things consumption

2:13and creation are completely separate

2:16They don't sort of always marry up, but in

2:18the Tableau platform of course

2:19They can actually overlap in lots of

2:21different ways, and that's why this sort of

2:23blue purple thing is a little bit confusing

2:25But let's take a look at this diagram. Let

2:28's select this and move this across now in

2:30today's world

2:31And this diagram on the left is sort of

2:33accurate it really what you should be

2:35looking at is a Tableau architecture

2:37diagram

2:37This gives you an accurate view of where

2:40all the services are sat, but it doesn't

2:42help me with my explanation

2:44So this is what I've made instead so if we

2:46go from the bottom up

2:47and let's look at the analytical flow when

2:50you're building a

2:51Dashboard or metric or something for an end

2:53user to consume you'd start here with your

2:55data sources at the bottom

2:56And it could be in Google Drive. It could

3:00be in the cloud

3:00It could be wherever you want it the data

3:02sources can come from a multitude of

3:04sources including flat files on your

3:06machine

3:07Once you have that once you've connected to

3:10your data source you see the next step

3:12I go to here is server and cloud now

3:14I've done this because in a world where we

3:16're bordering is a predominant experience

3:18and in a world where data sources mostly

3:21live on

3:21Tableau server and cloud this is

3:23technically true

3:25But if I was if I was being strict where

3:27you'd actually start from you know an

3:30analytical workflow is you'd you'd start

3:32off in

3:32Test stop then go down to your data sources

3:35connect to your data sources

3:36And then once you've connected to your data

3:38sources you then start building dashboards

3:41Once you've built your dashboards you then

3:43publish them back up to server

3:44And then now you've published them to

3:47server they can start to do all the other

3:48things for example

3:49You can use the embedded data source within

3:51the workbook to build a new view you could

3:53even publish the data source within that

3:55dashboard

3:55To tableau server and cloud and then use

3:58those in other dashboards and other metrics

4:01so this diagram starts to get a bit

4:02confusing

4:03Then once you've got your metrics or your

4:05dashboard sets up from those you can then

4:08go and do a bunch of things one of

4:09those is do an embedded experience where

4:11you put it inside of an application or you

4:12put it in a web page or

4:14You go on the left hand side, and you send

4:17it out to a client the client could be

4:19Tableau desktop or Tableau prep

4:20This is in essence me connecting back to a

4:23data source

4:23That's hosted in server or cloud if you're

4:25using desktop or prep if you're on mobile

4:28Then this is essentially just the Tableau

4:30server Tableau cloud app on

4:31Mobile consuming dashboards and metrics if

4:34you're on the web

4:35Then you're browsing the Tableau server

4:37Tableau cloud portal browsing the explore

4:40tab looking for

4:41Visualizations and using them and of course

4:43at the very end you've got your end users

4:44here at the top

4:45They're the ones who are fundamentally just

4:48you know users. They could be explorers

4:50viewers whatever they are

4:51They are the analysts coming in to look at

4:53some data and so in this flow

4:55This is sort of broadly the worldview today

4:58, and this is how it works again

4:59This is not a hundred percent accurate, but

5:01just bear with me as I try and explain this

5:03so that's the world today

5:05And Tableau through this term called head

5:08less BI and what I was trying to sort of

5:10represent

5:10What do they really mean well in that

5:12strict flow where you're starting from

5:15desktop?

5:15You're going to your data source, then you

5:18're going up to your dashboard you publish

5:20it to server and cloud

5:21Then you go up to one of these experiences

5:24and these experiences all require

5:27Metrics or dashboards to work so you can't

5:30for example

5:31You can't and if I sort of bring my

5:33annotation tools back you can't

5:36really

5:38Give a user any sort of experience unless

5:40it has a metric or a dashboard

5:43They have to start that even if you just

5:45wanted the data source

5:46You'd still have to create a dashboard that

5:48the JavaScript API could query or some

5:51other API could query and then from there

5:54You could go on and build your experience

5:56They need a dashboard and that's

5:57essentially not headless headless BI is

6:00essentially this concept where you don't

6:02need a dashboard

6:03you're essentially decoupling which is the

6:05term tableau used and

6:06The infrastructure so you can choose at

6:09what point you're taking part in the

6:11journey

6:11I'll come to a diagram a little later that

6:13explains this in more detail, but this is

6:15sort of the world today

6:17Let's take a look at the world in the

6:18future and for this I've added Tableau GPT

6:21and

6:22Tableau pulse into this matrix so you can

6:24get a better view of sort of where

6:26everything might sit again might sit

6:28because I don't

6:29Know we haven't seen the features. So let's

6:30go ahead and remove this. So

6:32This is a new diagram and what we can see

6:35here is that actually is very similar

6:37The the the way end users and the way

6:39people work is exactly the same

6:41However, if you go down here to the server

6:43and cloud section, it's a little bit busier

6:45. And what you can see here is that

6:47fundamentally and you now have where before

6:51you had

6:52Metrics and dashboards and you now have

6:54essentially three options you have Tableau

6:56pulse, which is just over here

6:58You have visQL data service

7:00Which is this there and you have your dash

7:02boards and the way to think of this is these

7:05three are equal citizens

7:06To the data source. They all have the same

7:09access to the same data sources. So here's

7:11the first big change

7:13This means you could build three separate

7:15experiences that are all pulling from the

7:17same data source

7:19That is either published to Tableau server

7:20or cloud or is going through a virtual

7:22connection or anything

7:23Any of those things sit on top of the

7:25connection layer. And so the way to think

7:28of this is it's actually layers

7:30It's actually a really good way of thinking

7:31about it

7:32so if I go back here and strictly speaking

7:35we think of these as like a

7:36Tower you start with your data source you

7:38go and connect to your data source again

7:40You'd be doing this in desktop, but just

7:42bear with me. Let's start from the bottom.

7:44You start from your data source

7:45That's just your you know, raw data source,

7:47whether it's Tableau server Tableau cloud,

7:49whatever it is you start there

7:51once you've done that you publish up your

7:54server and you publish up your connection

7:56to server or cloud and

7:57Once the data is there, you now have three

8:00options you can either

8:01Build a metric go and define a metric and

8:04then have it live inside of Tableau pulse

8:06You can ask an analyst to build something

8:09very bespoke. Maybe a dashboard maybe

8:11something nice

8:11maybe something that even uses Tableau

8:14pulse on the metric built in Tableau pulse

8:16or

8:16You could ask your developers to use vis

8:19cure data service to query the same data

8:22source

8:22To then go off and build an application

8:24interface that uses d3 in a web browser

8:27somewhere else

8:28okay, so those are your three options and

8:31so and you can kind of look at vis cure

8:33data service in this instance at

8:36Primarily being for embedded services and

8:38applications essentially this cure data

8:39service the only real reason you need to

8:41use this is if you want

8:43To build something external somewhere else

8:44right that that's the only real reason it

8:46makes sense

8:46Otherwise you just use the front end and

8:49set up the tablet give you through its own

8:51product

8:52Which just comes out of the box ready to go

8:54and is seamless

8:55You don't have to sort of rebuild or put

8:57any dev time behind it

8:58But for companies who do want to build a

9:00more tailored experience like Tableau

9:01showed during the keynote

9:02This this would be one of the ways of doing

9:05that. Okay

9:06Now the other thing to bear in mind here is

9:08that Tableau GPT is sort of sitting around

9:11These data sources so Tableau GPT was

9:13described by Tableau as being and while

9:15Tableau's own efforts are generative

9:18generative generative AI can't say this I'm

9:21having another like

9:22What is it? I just can't say these words

9:26are so confusing anyway

9:27these

9:31Technologies this technology is going to

9:33sit around this and the way Tableau showed

9:35it was side by side

9:36But really strictly speaking Tableau GPT

9:38needs access to all of these things to be

9:40able to enhance your experience

9:42And then server and cloud remains what it

9:44is today the secure sort of you know

9:47governed space that IT

9:48admins can use to control and manage

9:50everyone and

9:51Everything runs on already the

9:54infrastructure that's there for server and

9:55cloud

9:56So it's all set up nice and it's sort of

9:57nice and easy to work with

9:59And so that's sort of how this starts to

10:02work and so you can hopefully start to

10:03realize that look this actually changes

10:05In a very meaningful way a couple of things

10:08not just how users consume

10:10information and data as Tableau showed but

10:13actually also workflows and so what I did

10:14is I tried to make it a

10:15Much simpler diagram just here on the

10:17bottom. So let's go ahead and move this out

10:20of the way

10:21so if I just move this out of the way and

10:24Just look at today in the future you see on

10:26the left you have today's world. So you

10:28have your end user, okay, and

10:30They can choose their window into a data

10:32source. They can either use Tableau metrics

10:35. They can either go to a dashboard

10:36And they can essentially just you know

10:39browse their their setup how they want but

10:42it's actually quite narrow

10:43Everything has to originate from the

10:45dashboard pretty much

10:46There is no real way around it

10:49The other way the other thing you could do

10:51is you could you know have have your data

10:54sources going into Tableau

10:55Available through other platforms and other

10:57systems, but you'd have to build those

10:59outside of Tableau. That's obviously always

11:01been possible

11:01but yeah, it's a very sort of different

11:04setup to the one we have here on the right

11:06which is

11:07Really the ability to choose the window

11:09into your data source

11:10You can choose three ways of looking at the

11:12same thing

11:12And there is an interface to define the

11:14metrics that is decoupled from the dash

11:16boarding experience

11:18Previously it was very much part of the

11:20dashboarding experience

11:21And then the other thing is you can request

11:24data without having to put anything in

11:26front of it

11:27You don't have to have any sort of show

11:29showroom for your data

11:30You can go ahead and pulp pump it into

11:32whatever other service or API you have and

11:35for the record, you know

11:36There are many services and API's that have

11:38ready-to-go systems out of the box that you

11:40can do this with

11:41So this is a this is a diagram. I'm still

11:44working on I'm still not sort of super

11:46happy with it

11:47There's a couple of things that are wrong

11:49And you know in a way there's never going

11:51to be a right version because depending on

11:53what you're talking about

11:54This diagram will need to change

11:56It's kind of like a 4d chess where you've

11:58got to look at it different dimensions for

12:00it to make sense

12:01and but hopefully this hopefully starts to

12:04explain sort of all these concepts and so

12:06I think the big change for analysts if you

12:08're a data analyst today and you kind of

12:10wonder well

12:10How is this going to change my workflow? I

12:12think this this this is fundamentally it

12:14you're gonna need to actually start

12:15thinking about

12:16I don't think visco update service will be

12:18one of the big things unless you're a

12:20developer an API

12:21That will be pretty much a standalone thing

12:23and

12:24but

12:25you'll need to think whether you want to

12:26define metrics so that they're available in

12:28tableau pass with all the goodness that

12:30comes there or

12:31You're gonna build a dashboard that does

12:33the same thing, but we'll probably not get

12:35the same love and attention

12:37long-term as

12:39Something like tablet pass get you know

12:41that tablet pass is gonna get a whole bunch

12:43of suite of features

12:44They're gonna hook into newer technologies,

12:46of course, obviously running in tablet

12:48cloud

12:48My hunches tablet pulse is a hundred

12:50percent going to be a cloud only feature. I

12:53can't imagine

12:55you know

12:57tablet putting the kind of work required or

13:00even admins necessarily and

13:02Being willing to beef up their infrastruct

13:04ures to the level that can let something

13:06like tablet pulse run

13:08Um, if you really want to know and for

13:11tableau to sort of train an LLM to do this

13:14kind of work

13:15it's a serious investment if you look at

13:17the likes of chat GPT and

13:18You know open AI there and they're

13:22investing. We're literally talking

13:23thousands of

13:25hardware thousands of GPUs thousands of

13:29computing hardware a

13:30Ton of time it takes a really long time and

13:34whilst all of that is happening

13:35It costs money and electricity and that is

13:38just an investment and that's just stage

13:40one

13:40There are multiple stages to training an LL

13:43M

13:43It can take months on end using all this

13:45hardware with the right expertise and at

13:47the end of it

13:48You have something close to what chat GPT

13:51has and there's something called an ELO

13:53score

13:53Elo and essentially that rates how good

13:56LLM model is and so it's going to be

14:01interesting to see how tableau GPT

14:05sort of embraces this where that sort of in

14:07that that sort of

14:08Investment takes them and sort of what kind

14:11of capability are we seeing?

14:13With a technology like this versus the

14:15things we are seeing more generally through

14:17companies like Microsoft open AI and Google

14:20Anyway, we're getting off off track here

14:23I just wanted to describe what's going on

14:26with viscos data service using this diagram

14:28So hopefully at some point next year when

14:30it comes out we can dig a little bit deeper

14:32and find out more about it

14:34Anyway, thanks for watching and I'll catch

14:35you in the next one

14:36You

14:46[ Silence ]