Analytics Extensions Explained with Aidan Moore | Data Science in Tableau & more

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

0:02going to be taking another look at

0:03analytics extensions. After reaching out to

0:05the community about table extensions, I

0:07realized that analytics extensions were

0:09just a huge gap in my understanding. So I'm

0:11learning a lot over the last few days. And

0:13I really wanted to share the conversations

0:15I'm having with other people on this

0:17channel. In this video, Aiden Moore is

0:19going to be walking through some use cases

0:20for analytics extensions. And he's actually

0:22going to walk us through his thinking

0:24process when he discovered a solution that

0:26allows him to do something pretty cool with

0:29Salesforce.

0:29As ever, let's get stuck in.

0:31So Aiden Moore, I'm head of analytics at a

0:34company called Informer. I work

0:36specifically in the business operations

0:38team. So looking at a lot of Salesforce

0:41data and marketing data, that kind of thing

0:44.

0:44Amazing, amazing. And with with Salesforce

0:47data, it's probably worth asking the

0:49question, has that has that gotten better

0:51since Salesforce acquisition of Tableau?

0:53Are you finding there's more synergies

0:55between the two? I always ask that question

0:58when everyone works with Salesforce.

0:59It's a good question. I don't want to be

1:04rude to Tableau Salesforce. I think there's

1:06, is it getting better? I think there's ways

1:12, and actually some of the stuff that I want

1:14to demo to you today, I think overcomes one

1:17of the big problems that I find with

1:20integrating Salesforce data into Tableau.

1:22Amazing. Amazing. Okay, excellent. Yeah. So

1:26yeah, I sort of let you just show me what

1:29you want to demo. For the people who do end

1:31up watching the video, what I'll do is I'll

1:34explain the difference between extensions

1:37and analytics extensions, dashboard

1:39extensions and analytic extensions.

1:41And then I'll also explain ahead of time

1:44how table extensions use analytics

1:46extensions, if that makes sense. So you don

1:49't have to sort of worry about that clarity.

1:51I'll do that in the video itself. So that's

1:54all good.

1:55Did you get some good use cases for the

1:57table?

1:57I did. Yeah, I did. I did. I did. I'll send

2:01I'll send them through to you as a link. I

2:04got some from Tableau actually reached out.

2:07Francois reached out directly. He had an

2:10internal sales rep, not sales rep, sorry,

2:13sales engineer, who had sort of non AI/ML

2:16examples of them using table extensions.

2:19What was really interesting about that is

2:21quite varied. There's about seven examples

2:23in there, a lot of them to do with API

2:25calls, actually, which is turned out to be

2:27quite a sort of hidden gem within the table

2:29extensions, because you can essentially use

2:32it to call in call in datasets.

2:35But broadly speaking, I think I'm still in

2:37the early sort of phase of my understanding

2:40of the capability. So and my impression is

2:43whatever you code is whatever the

2:45capability you put into the extension. So I

2:48'm sure there's going to be hundreds of

2:50different things that get built with this.

2:53And it's just understanding the realm, the

2:55high level realm that this opens up rather

2:57than trying to understand every single

2:59individual use case. Yeah.

3:01Okay.

3:04I'm actually demoing some of this stuff to

3:06the Pittsburgh Tableau user group next week

3:10.

3:10Nice. Amazing. Amazing.

3:12My very, very first Tableau.

3:14This is your dry run.

3:15It'd be super interesting to see if there

3:17are bits that you're kind of interested in.

3:21Yeah, absolutely. Definitely. Yeah.

3:24Used to my advantage. So I've got here a

3:26very, very basic example. I take it you don

3:28't need me to go through how to connect to

3:30the analytics extension, do you?

3:33Why not? Why not? Just explain for everyone

3:35's benefit. I know you can repeat these

3:36things 100 times. People don't always get

3:39it. So why not?

3:40So I assume that people already have a Tab

3:46Pi set up to connect to it.

3:47It is very straightforward. We're going to

3:51help settings and performance. We're going

3:54to manage our analytics extension.

3:57You see here that I'm using a local host on

4:00port 9004. I can test my connection.

4:03I love that it sits under the help menu.

4:10I never understood that. Help, I need an

4:16analytics extension.

4:17So we can test it and it should work. I don

4:21't know if it'll look quicker than that. I

4:22don't know if that's because I'm sharing my

4:23screen or something.

4:27There we go. I've done a fast thing.

4:29I'll just cut the gap. We'll blame Tablai.

4:33Yeah, good stuff. Okay, good. So super

4:36straightforward example of one of the

4:38classic use cases, I think, that

4:39came out as soon as Tab Pi was announced in

4:43vision 10.2, for those of us that can

4:45remember

4:46that far back, was sentiment analysis. So

4:49essentially what we have here is a very

4:53simple

4:54dashboard. I can type in a text string

4:57and it's now going off via Python to use

5:03the natural language toolkit to do a

5:08sentiment analysis

5:10of the statement that I've put in there. So

5:12we can see that this has a very positive

5:1493 percent score. If I make an adjustment

5:23that has a very negative score.

5:26Interestingly enough, it's only 91 percent

5:30negative or its

5:30love is 93 percent positive. So I think we

5:33can prove statistically that love is

5:35stronger than

5:35hate. I wonder what gets 100 percent minus

5:38100 percent. It's probably one word.

5:40Does make you wonder. I won't ask. Actually

5:44increasing the number of exclamation marks

5:46sometimes. So maybe. Yeah, right. But yeah,

5:51so that's the kind of thing. And then

5:52people came

5:53up with great ideas of looking at things

5:55like song title, song lyrics. So we could

6:00do that.

6:01So a little bit of Gloria Gaynor, 81

6:05percent. This is Nine Inch Nails. Although

6:09I think Johnny Cash

6:09did it better. Sad song, 77 percent, minus

6:1377 percent. So what's happening in the

6:16background is

6:19I just opened this. So here we have a

6:24script real table calculation. So what is

6:27happening is we're

6:28importing the natural language toolkit in

6:32Python. And we are passing to it a text

6:35string. The text

6:36string is underscore arg one, which is a

6:39positional argument. So that's coming from

6:41this first

6:42attributes here. Right. And then in Python

6:45terms, we're creating a list here for the

6:48scores.

6:49And then for each of the words in our text

6:52string, we're getting the polarity of it.

6:56Then we're comparing that and adding it to

6:58our list of scores and then returning the

7:00score

7:01back to the dashboard. So that is what's

7:03happening. Very nice. Very interesting.

7:06Really cool.

7:08It's also quite simple as well. I mean,

7:11whenever you hear Python analytics

7:13extensions,

7:15I think people immediately just see tons of

7:17code in their eyes. But actually for a very

7:20powerful set of features, it's quite an

7:23easy sort of follow along example.

7:26Yeah, yeah, absolutely. And anybody who,

7:29well, I don't claim to be a professional

7:30Python code at

7:31all. I think anybody who's middled with

7:34Python before will completely get this.

7:38Right, right. And just to sort of give some

7:42users some background context, I understand

7:45that that

7:46import ni, is that n-i-t-k? N-l-t-k,

7:49natural language toolkit. N-l-t-k, natural

7:51language

7:51toolkit. That's a library that sits inside

7:54of the Python. And have you added that to

7:57the

7:57TabPy setup or is that just coming natively

8:01built within TabPy? So it doesn't come

8:03natively with

8:04TabPy. It depends which way you install Tab

8:07Py. So there are, to my knowledge, three

8:10different

8:10ways of doing it. The most straightforward

8:13way, in my view, is to use the Anaconda

8:15distribution,

8:16because that comes with a whole bunch of

8:18libraries pre-installed. So in that

8:21scenario,

8:21I didn't have to go and install the natural

8:24language toolkit. If people were just using

8:28native Python and had downloaded TabPy from

8:31GitHub, then they may well have to go and

8:34pull

8:36some of these in. Cool, and then the rest

8:39of it is just from the documentation, I

8:42guess. You've just

8:43followed the guide for nltk and you fed it

8:45an argument and then it comes back with a

8:47score.

8:48That's it. For full disclosure, I think I

8:51found this online a couple of years ago,

8:53maybe. I don't

8:54know who to attribute it to, but yeah, this

8:58is something I got. But in 2019, I was very

9:02lucky

9:02that I went to the Tableau conference in

9:04Europe in Berlin and also to the Tableau

9:06conference in

9:06Vegas, which was really cool. But it was

9:09actually the Berlin one that for me turned

9:12out to be much

9:12more cool because I had a bit of downtime

9:15and I didn't know what to do. And there was

9:18this great

9:18big hoard of people that walked past me and

9:20went up some stairs. And so being a bit of

9:22a sheep,

9:23I followed this hoard of people and ended

9:25up in an absolutely fantastic presentation

9:28by Rachel Bose.

9:30It was called "From Hello World to Saving

9:32the World with Python."

9:34Nice. And at that point, I hadn't even

9:36thought about any of this.

9:38Okay. Well, Rachel's presentation was

9:40really, really cool and it had some stuff

9:42like this.

9:42And so I investigated it a little bit more.

9:44I can send you a YouTube video that she did

9:46.

9:46Yeah, yeah. I'll be sure to put a link to

9:48it in the video as well.

9:50What really excited me when I saw this... I

9:54can tell you the bits that excited me. It

9:57was...

10:16Another bit that really excited me was this

10:19bit. Because what I'm doing here in Python

10:24terms is

10:25I'm creating a list. I'm adding zero to

10:27that list and I'm returning zero.

10:29Okay. So I'm effectively doing nothing. But

10:33what that means is...

10:36And the bit that I want to demo to you

10:39today is that knowing that I can do nothing

10:42in Python

10:44almost means that I can do anything. So

10:46instead of just pulling in the natural

10:50language toolkit

10:51and doing sentiment analysis, instead of

10:55pulling in some SKlearn to do some

10:58statistical data

11:00science-y kind of stuff, I can do anything

11:02I want. So long as I can do it in Python.

11:05Right. How did you arrive at that? Sorry.

11:08How did you arrive at that sort of...

11:12I would never come to that conclusion.

11:13Because I can do nothing. I can do

11:15everything.

11:15How do you sort of arrive at that

11:17conclusion? It's a good question. Maybe it

11:21's some dyslexic

11:22thinking. Maybe it's that I had a lot of

11:24beer on the day-to-night out. I don't know

11:26what it was.

11:27Right. But I left with the distinct

11:29impression that if I could re-raise those

11:32bits and if I

11:33could then stick anything that I want to in

11:35the middle. Right. Right. So let me give

11:37you some

11:37examples of what you can stick in the

11:42middle. So that's my safe one. Let's take a

11:47look at this.

11:47So here I'm doing what we've just had a

11:51look at. I'm creating my list of scores. I

11:55'm then adding

11:56zero to that. But now instead of importing

11:58natural language toolkit, I'm also creating

12:01basically a true false thing. Because this

12:03is a table calculation, it will fire

12:05every time something is filtered or not. I

12:08only want this to fire when I choose to do

12:11it.

12:11Instead of pulling in one of the

12:14statistical libraries or something like

12:16that, I'm actually

12:17pulling in PyODBC. And PyODBC is what you

12:20can use to interact with databases. Oh,

12:24nice.

12:24So here I've got... I'm setting up a

12:27connection. I'm writing an insert statement

12:30. I'm adding to

12:31that insert statement the value of arg1,

12:34which you can see down here is my username

12:37in Tableau.

12:38Yeah. I'm also adding in arg2, which is

12:42just a piece of text, after the name of the

12:45dashboard.

12:45So if I go here, let me just maybe blur

12:50that out. Shouldn't be better preferred.

12:53Yeah.

12:54So if we have a look at that, we can see

13:03that I have an empty table.

13:04Excuse the column headings. But what I can

13:08do in my non-safe version, if I hit my SQL

13:13button,

13:18I now go to my SQL table. You'll see I now

13:23have a row of data.

13:24Oh, yeah.

13:26Corresponds to my username and the string

13:29of text that I have been there.

13:30So I'm able, by using the analytics

13:34extension, to do things that people would

13:37more commonly

13:38associate with things that are in the

13:41extension gallery.

13:42Right. I see.

13:43But I'm able to code that in-house using

13:46the functionality of Python.

13:48I see. I see. So what you've done here is

13:51you've just grabbed your username, I guess,

13:55and you've just inserted it with the name

13:57of the data source, right?

13:58Yes.

13:59Username functions there. Yeah, perfect.

14:00Yeah.

14:01Cool.

14:02That was the first thing that I tried to do

14:04.

14:04Okay.

14:06I then took it a step forward. I thought,

14:07well, could I...

14:08Okay. That's the safe one without any

14:14credentials in it.

14:17Let's do that.

14:18I then thought, could I pull in one of the

14:21other libraries?

14:23Could I pull in...

14:25Doesn't like me now because I haven't got

14:28any credentials.

14:30Could I write a slightly longer script?

14:36And this time I'm pulling in SMTP lib,

14:39which helps me to connect to libraries,

14:42to get stuck in an email.

14:45I'm able to create a method. I'm able to

14:48pass, have some variables that are passed

14:51to that method.

14:52And ultimately what this will do is it's

14:58going to send an email to the address that

15:00I give it,

15:00saying this stuff. I clicked the button a

15:04few seconds ago. And if I go to my email...

15:08Do I have an email?

15:14Okay. I have an email. So I was thinking

15:16about things, maybe some offline training

15:18material,

15:19perhaps having it as a repository for forms

15:21for the business operations teams use.

15:23Then I extended a little bit further. I'm

15:29hoping this journey is useful.

15:32It does lead to just lead somewhere.

15:37I can feel like you're building up little

15:40components.

15:43Yeah. So I have a fantastic automation tool

15:47that we use in the team called UiPath.

15:49There are lots of other tools there, Autom

15:51ation Anywhere, Blue Prism,

15:53Power Automate, MuleSoft, even if we want

15:56to stay within the Salesforce ecosystem.

16:00When this loads, we will see...

16:08Okay. So this is a queue in UiPath, which

16:24is basically just think of a

16:26queue as a list of things that need to be

16:29done.

16:30Need to be done. Yeah.

16:33So I'll turn the safe one. So at the moment

16:35, just to clarify, at the moment you can see

16:38that my

16:39queue is empty, has no new items in it.

16:41Yeah.

16:42Yeah. But perhaps predictably now I'm going

16:45to click my UiPath button.

16:47Then if I refresh this,

17:00you'll see I now have a new item. Yeah.

17:03Nice.

17:03So I've been able to take some information

17:06from Tableau.

17:09And this time in between my scores, I'm now

17:15using the request library, JSON.

17:17I'm running methods to authenticate against

17:22the API.

17:23And I'm then taking some bits of

17:26information. In this case, I've hard coded

17:30it.

17:30I'm able to take these bits of information.

17:33And if I have a look at my new item here in

17:35this queue,

17:36you will see that it matches. So I've had

17:40some information in Tableau.

17:43I've been able to click a button in Tableau

17:45, which is an intuitive part of the

17:47dashboard itself.

17:48Yeah. If you look out of place on the

17:50dashboard, I've been able to ping that

17:51information into

17:52awaiting API. As part of doing that, I've

17:55also been interacting with APIs and that I

17:58've had to

17:59go off and get an authentication credential

18:02in order to do that. So I'm within the

18:04Python

18:05environment and I'm managing the process

18:07for exactly how that authentication goes on

18:11.

18:11Yeah. So hopefully that was good background

18:16for what I'm about to show you now.

18:27So if I go... I thought I had it open.

18:31Okay. That's Salesforce.

18:49I'm now in Salesforce. Yeah.

18:53And this is an opportunity record and this

18:55opportunity record has some products on it.

18:59There used to be a process to do pricing

19:05approvals for our Salesforce records,

19:10but it was very manual, very clunky, took

19:12ages, wasn't particularly reliable and

19:15generally caused

19:16a whole lot of headaches. They looked at

19:19getting CPQ, but due to some restructuring

19:22in the business

19:23wasn't the right time to really explore. So

19:26the question came to me.

19:29Sorry. For everyone's benefit, what's CPQ?

19:32So customer price quote or customer price

19:35quote. It's a tool to do pricing approvals,

19:39but it also does it from additional bits.

19:41But the pricing approvals was the really

19:44key part

19:45for my business. So the question came, was

19:48there anything that we could do to improve

19:51this

19:52entirely manual process that they had?

19:54Right.

19:55And I thought, well, knowing what I now

19:58know about the fact that I can click a

20:00button somewhere and

20:01it can interact with an API, or I can click

20:03a button somewhere and interact with SQL.

20:07I thought, yeah, I imagine there's some

20:09stuff that we can do. I already had an if

20:12rame. And so

20:14we do a lot of reporting as an opportunity

20:18level via my Tableau iframe in Salesforce.

20:23So we're looking at this now. And here we

20:27have something for pricing approvals.

20:30What I'm going to do is I'm going to submit

20:33for pricing approval.

20:35Now what this is doing, this is using

20:46replicated data of Salesforce. It's a Table

20:50au dashboard.

20:51It's pulling in some bits of information

20:53and what the currency is, what the

20:55opportunity number is,

20:56the products, how many seats, some fake

20:59list prices, what the sales prices are that

21:03is on

21:03the opportunity record. And then Tableau is

21:05doing a calculation of what sort of

21:08discount

21:08percentage that is. Because of the size of

21:11the discount percentage, who needs to

21:13authorize

21:13this opportunity? And then down at the

21:16bottom, there's a submit button.

21:21And this submit button is doing exactly

21:22what I've just demonstrated. It's firing

21:24something

21:25out into UiPath. And when it does that, you

21:27don't have to use UiPath. You could just

21:29have

21:30a back script, SQL table, or any other

21:33automation tool. But that will then result

21:36in a bunch of

21:37emails coming out to say this opportunity

21:39needs approval. People then click on the

21:42button,

21:42and then that button will turn into another

21:47Tableau dashboard where they can go and

21:50click

21:50a button that says either approve or deny.

21:53So again, it's using all this functionality

21:55of

21:55being able to interact with APIs via click

21:58buttons in a Tableau dashboard.

22:00But there is-- having shown you how great

22:05this is, and it is a great process,

22:07I'm now going to tell you why it's a bit of

22:09a pain. Because if I go back to--

22:11If I go back to here, we can see that I've

22:14got some sales prices here. And I've got

22:18some sales

22:18prices. And those sales prices, as I

22:20mentioned, are being pulled by replicated

22:23data.

22:24Right. Yeah. But the replication, it

22:28happens, let's say, every 15 minutes or so.

22:32So if I'm

22:32a veteran, I'm talking to a customer. And I

22:36want to get the deal through as quickly as

22:38possible.

22:39So let's say I'm going to come in here, and

22:41you and I are talking about this particular

22:44product.

22:51And we're going to say-- I like you, Tim.

22:54So I'm going to give you this for 20,000.

22:56It's a special deal just for you. 20% off.

23:00Wow.

23:00So I'm going to give you that. So we've

23:05agreed that. And now I want to push it

23:07straight through

23:08for approval. Yeah.

23:10I go into my analytics. And I select my

23:15pricing approval.

23:20I'm going to submit it for pricing, please.

23:22Nice.

23:23But you'll notice that the 20,000 that we

23:36just agreed--

23:38It's not on there. Yeah. Yeah. Now, that's

23:41a bit of a pain,

23:41right? Because I'm an ambitious rep. And I

23:43want this to go through--

23:44You want it now. --pricing approval.

23:46Yeah. But so I guess the question is, how

23:50do we get around that process?

23:54Yeah. So let me show you how we get around

23:59that process. And this is where I think

24:00there's a lot of opportunity for the table

24:02extensions.

24:03What we're going to do now is-- we've

24:11talked a bit more. I'm going to edit the

24:14products.

24:15We've talked a bit more. And actually,

24:18maybe I don't like you so much anymore. So

24:20now I'm

24:20going to charge you 40,000 pounds. Oh, no.

24:23Inflation. You've just cost yourself 20,000

24:27.

24:27I'm going to save that. So I'm showing us

24:3840,000 here.

24:41I'm going to the Analytics tab now. But

24:48instead of hitting on my pricing approval,

24:49I'm going to hit my super secret button

24:51down here.

24:51Is that just the white box that any unit

24:55exists? Yes. That's it.

24:57Love it. And I'm going to submit my pricing

24:59approval this way.

25:00Right.

25:09Now, let's see. It's instantly--

25:17Yeah. Yeah.

25:21Yeah. So how did I do that? Because if you

25:24're connecting Tableau to Salesforce,

25:28you can only replicate every 15 minutes on

25:30an extract. You know that my replicated

25:33data

25:33in SQL takes between 15 minutes and

25:34sometimes a couple of hours. So how have we

25:37done that?

25:38Yeah. You have to pull it in somehow, right

25:41? Yeah. Exactly. The way that I have done it

25:45,

25:45and up until the recent announcement about

25:48Snowflake and Tableau having

25:50basically live replication, I believe I was

25:53the only person in the world who had pretty

25:57much real

25:57time access to Salesforce data. Amazing.

26:04And if I can find my safe one, I'll show

26:08you what I'm doing.

26:10It opens.

26:16Yeah. You'd hope that in the future, Table

26:23au just figure this out. So you're always

26:26seeing

26:26the latest version of the truth, right?

26:28Yeah. You'd expect so. But in the

26:32meantime, what I'm doing-- so similar to

26:35what I was doing with the UI path and

26:38pulling the request.

26:38I'm also pulling in PyODBC. So I'm creating

26:42a method to enable me to interact with SQL

26:45Server.

26:46I'm deleting anything from a particular

26:48table, which has the ID of the opportunity

26:52that I'm in.

26:53I'm then connecting to the API from Sales

26:56force. I'm asking Salesforce to give me

26:59everything

27:02for the opportunity, the opportunity to

27:04line items, the price, et cetera,

27:06and then using Python to insert that into a

27:09SQL table. And then when I do the hop from

27:12that

27:12button to make a pricing approval, that's

27:18then moving onto a dashboard, which is

27:20making a live

27:21query to this particular table. And in

27:24doing so, I'm able to have a live-- I say

27:29live in inverted

27:31commas connection to the Salesforce data. I

27:35see. So you're creating almost like a--

27:41you're staging the real-time data. Exactly.

27:44And then over time, it will eventually

27:46replicate. It would just turn up for this

27:49use case, right?

27:49Exactly that. Amazing.

27:51So by using the analytics extension, I'm

27:53able to, first of all, create the

27:57functionality within

27:58Tableau to enable me to do things like

28:00pricing approval. And the benefit of that

28:03is it's keeping

28:03people within the Tableau and Salesforce

28:05ecosystem. We're not sending them off to do

28:07some manual

28:08process. It's really fairly straightforward

28:10to set up. It's self-coded in Python, which

28:13is

28:13widely available within the data analysis

28:16community. And then I'm able to do funky

28:20stuff like interact with APIs, pulling in

28:21data, putting that data somewhere, staging

28:23it, as you've

28:24said, and using it in tools like this. That

28:28's really impressive. I had no idea that--

28:33in my head, I think I had this very narrow

28:35perspective that analytics extensions were

28:37only used for, I guess, data science-like

28:41analytic capabilities.

28:44Yeah. You were using the middle of the

28:46scores. Yeah. As you say, because you can

28:48do nothing,

28:49you can do everything, right? Yeah. You're

28:52using that opportunity in there to bring

28:56other things in

28:57and also send things out, I assume, to make

28:59things work really nicely. That's really

29:02clean.

29:02Do you see that-- I guess in your

29:05particular instance, this is a fairly

29:08tactical deployment,

29:09right? So do you find it easy to scale?

29:13When you put this on Salesforce, and I

29:16guess you're

29:16setting this up, do people find it easy to

29:19use that sort of button and just know it

29:22works? And

29:22it works really well, right? Yeah,

29:24absolutely. Yeah. Absolutely. It's simply a

29:26one-click button.

29:28Everything else is being pulled in for them

29:31. Previously, they had to literally go

29:33somewhere

29:33else and type all these things out. Heck

29:36yeah. God, yeah. And it's an interesting

29:40one because

29:40in this use case, if I heard you sort of

29:42explain the use case at the start, I would

29:44have said,

29:45surely that needs a write to database

29:47extension or something like that, right?

29:49Because at that point,

29:52I mean, that's sort of just in my

29:53simplistic head, that's the most obvious

29:56way to do it.

29:56But you're kind of almost abstracting the

29:59need to have an element in the interface,

30:01and you're just building it into an action

30:03rather than having to create an interface

30:06for it,

30:06which is quite nice. Yeah, exactly that.

30:09Yeah. Good. It works brilliantly.

30:13Yeah. There were other things that I'd like

30:15to try with perhaps with the table

30:17extensions.

30:17Because you can interact with APIs, it's

30:21out there. You can do pretty much anything

30:24you like,

30:24right? But yeah, Twitter data live and

30:26doing sentiment analysis. So combining the

30:29non-traditional and the traditional use

30:31cases for the analytics extension, bringing

30:34those together

30:35would be awesome. I don't know. I think it

30:38's one of these features where I was talking

30:39to

30:40Charles Laporte from Bistry yesterday. And

30:44we were talking about sort of where does

30:47analytics

30:48extensions and table extensions really sit

30:51within the day-to-day analytical flow? It

30:54doesn't feel

30:54something that's mainstream across, let's

30:56say, all the creators who have access to

30:58that license.

30:59But it maybe is more of what I call explor

31:02atory analysis sort of tool initially for

31:06someone who's

31:07just investigating data. And then when you

31:09come to productionizing it, you kind of

31:11have to do a

31:11little bit of the work you've done to

31:13simplify it into a button to kind of really

31:15package it up

31:16nicely and then move it on. Or you could

31:18move that work further back. But in your

31:20case, that's not

31:21possible because of the setup with Sales

31:23force. But you could potentially move some

31:24of that further

31:25back into the stack and have it come

31:27through in the day source natively. Yeah.

31:29Yeah. Yeah. I agree.

31:31And you mentioned Snowflake as well. And I

31:34was going to ask you, have you done any

31:36work with

31:37Snowpop, which is their sort of take on

31:39running Python on Snowflake, if that makes

31:42sense?

31:42Not yet. We are currently an AWS house. But

31:47we are.

31:47Oh, right. Right.

31:48I've just started testing Snowflake.

31:50Snowflake. Nice. Of course. It's funny

31:52because Snowflake run on AWS as well, right

31:55? So

31:57they're also an AWS house, if anything. But

32:01the thing I'm trying to sort of see if

32:05anyone has any

32:06understanding of is where is the best place

32:09to be doing this kind of data science work?

32:13Is it

32:13alongside the database as Snowflake is

32:16suggesting with the data? Or is it sort of

32:20bringing it closer

32:20to the people who actually know how to

32:22write these tools? Or is it actually a

32:24mixture of both? You do

32:25it locally with Tab High, then you move it

32:28further back into something like Snowpark

32:30or AWS Lambda.

32:31Yeah, potentially. And with the stuff that

32:35I've demoed, although I've been showing the

32:37table calculations with the actual Python

32:40scripts in, I could deploy those as models

32:42to my

32:43Tab High server. For example, not entirely

32:45the same as what you're talking about, but

32:48you don't

32:48have to have great big long Python scripts

32:50within your table calculations. You can

32:52have the script

32:53saved up to your Tab High server and then

32:55call them. Right. And so if you call them

32:57as modules,

32:57they're almost working like functions in a

32:59piece of code, right? Yeah. So you just

33:01pass the

33:01attributes to it that it needs. Yeah.

33:04Amazing. But it's not so fun to demonstrate

33:06if it's just

33:06like one line of attributes. Yeah, that's

33:10the thing. People always want to know what

33:12's under

33:12their head. What's in the black box? That's

33:14always the ephemeral question, right? Yeah.

33:17Good. Well,

33:18thanks so much for that. That's been really

33:21fantastic. What I'll do is I'll edit this

33:24up.

33:24Do you want me to send it to you ahead of

33:26time so you can just watch it, make sure

33:27that

33:28everything's good? I'll do a good job of

33:30going through and sort of cleansing

33:31anything that

33:32shouldn't be in public. If you can make me

33:34look handsome though, if you've got some

33:35filters or

33:36something. Yeah. Oh, God.

33:40[silence]

33:46[ Silence ]