Superhero Salesforce Records: Predictions with Einstein

If you ever done one or ever wanted to do one, it's quite painful.

All right, we're live on YouTube. So my goodness, we're live everywhere.

Make sure I'm not lying about the schedule here.

Yep eleven forty five that's Zach Berman is coming up next but right now we have Steve Marcel and we're gonna.

Steve Marcel is the lead and owner of Cloud Jungle and he's also an expert on Einstein Analytics. He presented at the last X Force webinar and I'm we're looking forward to this presentation on superhero top ten. So with that, take it away Steve.

All right, so thank you Leonard and a big thank you to the X Plenty team as well for inviting me back. So yeah, always pleased to be here.

And so yeah as we said this presentation is on Einstein Analytics and actually the presentation is fully in Einstein Analytics as well so we're going to be taking you through a dashboard going forward but just I thought we'd start off with this quote from Albert Einstein himself the important thing is not to stop questioning curiosity has its own reason for existing and this resonated with me when looking through some of the data sets that we're gonna present to you now.

So brace your eyes because we've some colourful slides coming up.

A quick introduction to me as I'm not a data scientist, I'm not a coder, I'm not really techy believe it or not and I'm definitely not interested in the superhero thing. I was passed some superhero data sets and somebody said see what you can do with this and I really didn't know anything about it and didn't understand that there were these magazines called I think it's one called DC Comics and another Marvel Superheroes or something like that and knew absolutely nothing about it.

Really I'm completely unqualified to do this but I think the idea is that with Einstein Analytics you don't need to be a data scientist, you don't need to be a coder and you actually don't necessarily need to understand your subject matter although it might help a little bit sometimes but the idea is you can get in and look at the data and try and make some sense of it. So this presentation really is all about what I call point and click Einstein ninety six so there's no special coding in here, it's all drag and drop sort of interfaces and the reason for all of the crazy colors is we're going to try and demonstrate some of the features that you can actually do on dashboard or to bring in things like hyperlinks and actually some charts and graphs etc as well and how you can format it the way you want so we just got a little bit crazy with the formatting.

So hopefully you've joined us and you're ready to enter our contest as well and the contest is about creating your own superhero and you can do that using the QR code at the top and it's going to pull up a Google Form but if you haven't got your browser enabled on your phone or you just want to use your standard computer browser go to bit.

Lyxforcehero and that will bring up a form where you can create your superhero.

By all means open that up now or there's plenty of time to fill that form out if you want to see a little bit more of the background info behind it all.

So what we're trying to do is demonstrate how we can make structured, unstructured data. We've got these couple of disparate data sets and we basically pull them together and then with that we're also looking at all this data to see if we can model it produce predictions.

Effectively what we're do is we're going to take things like superior eye colour, hair colour etc and see if that has an influence on the total power score of superhero. And we can perhaps talk about some examples of how you might not do superhero contests but apply this into the real life workplace as well.

Yeah at the bottom of the screen we've got our I've called a feature bar so those sort of familiar with Analytics these are on here these are the things that we've used on each page to demonstrate the capabilities of Python Analytics so things like page hyperlinks you saw us go from the the opening page to this page.

We've got text widgets which is plain text and we've obviously got our lovely styling, put things in containers so we can move them around the page, etc. So I'm now going to use a page hyperlink here, data sets, to put us onto the next page and in here we can see some lenses, some grays. And so what we've got here, going back to the story about the superheroes, is these are our two different data sets.

We were given several data sets and some of them were like total garbage and these two here there was at least some fairly decent correlation between superhero name and one was very much metrics around superhero durability, combat scores, strength scores etc almost like top trumps for superheroes and you ended up with a total superhero score.

Another one had some more physical features like height and weight, eye colour, the type of hair if they have hair, sex so male, female or other and what breed were they human, other, mutant I think we've got as options and the alignment is whether a goodie or a baddie and the publisher is which comic they first featured in apparently you have been told that some of these superiors would appear in more than one comic.

Anyway, so again looking at what we've got in terms of features on this page, here we've got our lens expiration. So this is basically IAS Analytics looking at some data sets and we've just formatted the charts here and you can see for example eye colour we've used some conditional formatting to relate the colors used to the wording.

Our challenge really is to see if we can merge these two data sets and make some sense out of it.

So what we do then in analytics next of all is we do some data prep and there are some various ways we can prepare our data and analytics. We have things called data flows and for this I used a recipe where we pulled some data in together.

And so what I'm going do now is open up the recipe so I've got a hyperlink up here so this is opening up live the recipe in Einstein Analytics just to give you a little bit of an idea of what went on in the background.

So I used like an augment or lookup facility to basically pull together two different data sets and the lookup key was 'superior name' effectively and then I found a quite good correlation between the superheroes in both data sets.

Think on one of the data sets I had about thirty seven thousand superheroes which was incredible.

Then I'll just x that, go back to our data prep and then once we pull them together, obviously you match things like obviously match them superhero but I can then have one data set where I've got those metrics of speed, combat score, power etc along with things like hair colour and eye colour etc etc and then I can look at each individual parameter like eye colour for example and I can use a bucket feature so you can see I've got all these buckets down here and so for eye colour for example, I would have grouped the popular ones together so blue to blue so this is like a formula field or formula column and so I've bucketed the blues together and then on one dataset we have blue with a capital B and one with a lowercase b etc.

So obviously that is just blue and all of the strange colours we just grouped as 'other'.

Okay so I've managed to bring my data down to sort of group together those small values.

We did that with the publisher as well so it was DC Comics, Marvel Comics and there was a few other ones but they weren't pretty insignificant so we grouped them as 'other'.

Now you'd probably be aware that when you know especially the data analysts out there you get given data and you get like a day column and it says next Thursday in there which is not a lot of use to anybody so it's an important aspect of any work in data science is effectively cleaning your data to start with make sure you've got something of quality that you can then go and analyse. So that's the data recipe piece and then from there what we do is we put it in ISO Analytics into a model.

So don't forget if you haven't made up your superhero yet and you can do but what we want to do with this model, as I said before, is to see if there's a correlation between our selected variables, so your height and your weight, your eye colour etc and these are things that you can adjust on the form and total power score. Okay so do you think hair colour is going to influence total power score?

Well this is what we can do with Einstein Analytics is look for some patterns in the data that probably as humans we wouldn't actually pick out necessarily, but they could actually have a good influence on our desired outcome.

So that might be for example if you're a retailer, you know where you place your product, for example top shelf, middle shelf, lower shelf or end of aisle, so in different products group next to other products etc. You've all these little variables and you might have a lot of data on that but you can put it into a model to see where you're going to have the most influence and that might be total sales for example choosing what product lines to do in what season, all sorts of different variables that you may choose to analyse. Or it could be something like engineering for example where you're looking at variables in terms of how you set up an engine, in terms of the overall efficiency of that engine or speed or your outcome may be fuel efficiency or something like that. There's lots of ways to use this tool.

So really you've basically got to decide what are your variables and then what is your outcome variable and for us our outcome variable is the total power score there.

So what we then do is we take all this data and we plug it into Einstein Discovery.

I'm not going to go through the whole creating the Discovery story on this session, I did it on previous one, on a previous seminar I did on the X4 Summit but what we can do is we can actually flip to our story and see what's happened. So basically what Einstein Analytics has done, it's crunched up all that data and looked for patterns, okay, as we started off to see what is behind the data, so what has happened and for those of you that have done the form already and picked a high intelligence score, well good on you because that was the number one variable that had the biggest influence on the total score, so if you picked an intelligent superhero then great.

So here, just to explain this, we've got the bar, so this is the ranges of scores that you might have chosen, so analytics sort of groups these together by default and blue is where there's a significant change from the norm and the grey where there is no significant change. So basically we're looking at our data sets and working at the average score of all the superheroes and we're saying that if you've chosen an intelligence score of one to fourteen, that is significantly different from the average.

Unfortunately if you did that, you're going to get a worse total outcome score and everything else was generally the outcomes variables your total score will increase with your intelligence typically. In that there are some other variations so if you've gone for the seventy six to eighty eight range and your breed is other then that was actually one of main factors to influence a really good score. So what analytics does, here's what intelligence orders it in order of the most significant combinations of variables that you have selected. Here we can see BREE is in human, intelligence is in that seventy six-eighty eight range and also the sixty four-seventy five range it does worse so if you actually pick that range for your intelligence and you've got really as human that's actually a negative factor so even though overall intelligence is a good thing not if you're human and you're in that sort of medium high sector and it'd be a bad thing.

So etc etc there's lots of ways you can analyse this data in terms of what has happened in the data but then we can move over to why things have happened okay and we've looked at intelligence, let's just scroll down and look at something else here what we've got going at random. Height one hundred ninety nine to nine seventy five, quite tall superheroes has a very good outcome on the score. We can see our average score here, two twenty five our average superhero score and just by selecting the height of one hundred ninety nine to nine seventy five, that adds thirty one to the score.

In fact here, if you did that one hundred ninety nine to nine seventy five and you selected Marvel Comics then that would deduct four points from your score.

In fact what we'll just do now yeah we'll go to what could happen.

Now we'll keep on what happens and we're not going to do a compare what's the difference?

So for example, you might want to think about what the difference is between blondes and brunettes so we've got hair blonde and hair brown, okay, and to see which one is better. So here we've got the average when the hair is blonde, okay, however if you're blonde and blue eyes that's a bad thing etc etc and then we can see where your blonde and publisher is Marvel Comics that is a good thing for example. So the combinations of the two then add up to the difference between when your hair is blonde and hair is brown and generally the brown haired superheroes do better than the blondes but then we can perhaps add something in like sex so let's stick that in there as well and what that has done is it's the same chart as we saw before but now we can just split out to say actually when the sex is male we can see that the brown haired factor is even stronger and we will be pleased to know the blondes do better than the brunettes if they're female.

Then we have the whole sex as other category as well where there doesn't seem to be a lot of variation going on there but there are some minor combinations of variations.

So you can actually really get into the detail of all these different variables.

So does this model that we've actually deployed, is it any good? So we can actually go and look at the success of our data model here and I'm gonna go to the evaluation here and so what this shows is we've taken our data and ISO Analytics pretends it doesn't know the total power score but just looks at the variables that we've selected and runs a prediction on that data set itself and then compares it with the actual power scores and you can see that there's a really really good strong correlation of predicted outcome against actual you know in fact there's quite a few dots like bang on the line and then we've got a few that are slightly off and for those of you that are into your statistics you'll know things like r squared values and all these sorts of things and different folds of data and you can really examine the actual details behind it all but really we can just see from the previous chart there is a really good correlation or a prediction correlation so we can see those factors that do influence total power scores other than those just power metrics themselves.

Okay right let's fire on to our dashboard.

Brilliant, so we can see plenty of you on here are actually doing some scores which is great.

So what happened is you filled out a Google Form and then we used the Xpence integration technique to integrate Xpence with Salesforce and what that did was it created a Salesforce record in the background.

In fact I'm just gonna edit this chart very quickly because I just realized we're missing some vital information which is Creator so we can see actually who the actual people are.

There we go, a bit of development in real time.

So Leonard! This has got to be a fix right? So Leonard has chosen Captain Zombie America and so far he is ahead in our leaderboard at five point six as a score so he's and honestly he's not actually seen this presentation before so done Leonard followed by Anonymous and Nick Hills. So we'll refresh this in a minute to see if there's any change.

What we've got here on this dashboard and just again demonstrating some features of Einstein Analytics is we've got our all time top ten so these are from the big data set of six hundred odd superheroes these are our top ten and you can do things like put a heat map on there and then a bar to show you the average so obviously they're all above average since they're the top ten. And then on the right, these are the ones that have been created today and so what we used here to do this, obviously we mentioned the integration that we've got to create these records and normally, so what we then have to do with this goes into Salesforce but then we need get that data into analytics and you can do that with various replication processes in ISO Analytics but we use a new feature called Salesforce Direct so you've actually got a live link between ISO Analytics and Salesforce which means as this data comes in as you fill out your Google forms we can see the scores come live in real time.

So I'm just going to refresh the in fact what we should do first of all is quickly save that because we made a change to the dashboard and let's give it a refresh to see if there's any change in the overall board.

Bear with me let's go on back to the start looking like a live demo and that didn't say properly because we had the error but Lend is still at the top of the leaderboard there so well done Leonard.

So that's it in summary, to give you a quick whiz of some of the key features of Einstein Analytics in terms of how you can basically look at your data, prepare data for the analytics using either data flows or recipes, also the new Dataprep tool which is just coming out I believe which makes life a little bit easier to pay data.

Then you can model it so you can then ask Einstein Discovery to go and model your data where you have to pick an outcome and then provide it with a list of variables it can then give you some information behind your data and then you can get some predictions so these predictions are written straight into Salesforce from Einstein Analytics so we can see then this predicted score for Captain Zombie of America and display everything in a live dashboard to explore further.

So I think that pretty much sums it up and open up to any questions.

Thanks Steve, we do have a first off, I did not see this presentation before it was given. All right. So I apparently I'm just good at picking superheroes though. I think you're probably the only person on earth who knows less about superheroes than I do honestly. But we have a question anonymous attendee asked, so they were they think their predictive analytics are super cool. How would Einstein analytics compare with Tableau?

Very good question. So I'm not a Tableau expert and I think from it is interesting how Salesforce are positioning Tableau and Einstein Analytics because most of you probably know that Salesforce have bought Tableau and it's in their portfolio.

So Einstein Analytics they say use Einstein if you are a Salesforce user currently say let's say you're a sales guy and you're using Salesforce natively then use analytics. I think they've still got some integration work to do with Tableau and Salesforce and so it's kind of embedded in there and a lot of the great applications and a lot of the clouds that Salesforce are now deploying have analytics component built in so they have their pre built dashboards for example and reports which are great.

They're saying that if you are not typically a Salesforce user but you want to look at some analytics and data prep then go down the Tableau route.

It's yeah I the features are the fact that you know especially with the Salesforce Direct side of things it's going to weigh towards ice analytics and I think traditionally some people believe that some of the data in the graphs, the way things are presented in Tableau potentially may be superior in the past but I think analytics have caught up so it's a tough one really but I guess we just go with the sort of Salesforce line so if you're a standard consumer of Salesforce then you should definitely stick with analytics and I also think probably look at your capabilities in your organization as well so if you've got people that are skilled on one or the other that's gonna certainly help.

The other key difference I suppose is that Ice Analytics is one hundred percent cloud whereas you can have an on premise version of Tableau as well.

Sorry if it's not a great answer but I thought it was pretty good.

I've used Tableau. I've used Tableau quite a bit. And what I would say one of the differences you don't have any of predictive analytics at all. Know, just demonstrated none of that in Tableau. I mean, you can do some of that.

This is something if you're not comfortable with statistics, this will go and give you a whole bunch of insights into the data.

Tableau doesn't do that. You to find them yourself basically. This is what I would characterize as a difference.

I was gonna add a couple of questions to ask and by the way, anybody in the audience has a question just throw it in the Q and A.

When you started the demo you talked about the data that you're matching two data sets and you did a lookup on a superhero name. I think we've all seen this use case before where someone has some external data that they wanna match into data that's already in some database or in Salesforce or whatever.

What kind of fuzzy capabilities or special matching capabilities does Einstein Analytics have if any help with the inevitable issue of some of the keys aren't matching up, if your data isn't perfect?

Yeah, it depends what you're trying to do.

That data that we just saw was, there was a pretty straightforward hit with a key in terms of bringing in on superhero names. So that was quite straightforward, but things like that bucketing functionality was great. And then in the data flow editor, you've got some more advanced features, shall we say. So you can create your own fields. I think we saw before I created that the eye color field based if it was brown with a lowercase b or brown with an uppercase b we're going to call it brown.

Dataflow you've got some more advanced features and one of the tough ones that we see is around sort date formatting as well and there's sort of ways of splitting out single lines into multiple rows. So especially if you've got a row with lots of date values in there but you want to use those date values independently of each other. So let's say you've got something like an actual day, forecast date and a baseline day or lots of total target dates with slightly different descriptions but you wanna be able to see those sort of side by side but they're on the same line, then you can do things like you can split those out.

So it's all different ways of splitting rows of data or joining rows of data together, various lookup filters, outer joins, inner joins in terms of how you augment your data together. So you do have a lot of options and when you get stuck you just go to the community and you sort of pose the question and generally somebody's already asked it and come up with a really good solution. I think that's one of the powers of Violence Analytics is the analytics community is really strong and really helpful. There's a huge passion around the platform out there.

So hopefully that answered that question.

Can you tell us a little more about Salesforce Direct and what you know about it?

Yeah, basically it's a live link between analytics and Salesforce. So literally as you were filling out those forms there, then it was bringing the data in straight into analytics and then you may say well why don't you do that for every record? Well I think you just need to select certain objects where it's important to get that information in and that does have some limitations, is kind of a straight in shot for getting object data into Salesforce and it's not going to work if you do have to do lots of transformations of that data or if you want to bring in multiple levels of a hierarchy say for example account and contacts etc all linked together and it's really a one object straight in shot piece of data.

With the replication services they got with iSanalytics, we typically run for our clients data syncs on an hourly basis but then again it depends on you've got your limit of how many data flows you can run-in a twenty four hour period.

So just it really depends on use case there in terms of the urgency for getting that data and some of the data may not change that much from day to day so a sort of daily sync may be adequate but other stuff whereby you're trying to do a presentation with audience participation and you want that data in straight away you don't want enough to go into your data flow monitor to run that data set to run the replication to bring it in so it's nice the fact you've got the option to bring some stuff in life.

And Salesforce Direct I'm assuming is extra cost option, it's not?

No, it's all there, it's a standard feature.

Oh, it's part of the base Einstein.

Things change, I'm shocked.

Another question I had was, noticed that, you know, part of the whole predictive analytics that whole, you know, stuff you showed where basically, you know, what does eye color have to do with hair color? And how does that relate to the power score?

Now there's a whole bunch of powerful statistics behind that.

Have you found that people sometimes who aren't looking at the underlying statistics very closely that some of them are get themselves in trouble maybe with correlations that aren't that great or sort of synthetic correlations that aren't really there or is Einstein pretty good at only giving you stuff that is statistically significant?

Yeah, no, is very good. So actually when you build the story it flags up if you've got too tight a correlation between your variables.

So for example, I don't know, let's say height and chest size, for example, if that was one of them, then there might be a really strong correlation and it probably it might skew the model too much and it suggested that you might want to delete one of those. And especially if you're putting in lots of columns of data into the model, it would actually say that you could be better off just ignoring some of these variables and relying on the strongest ones.

Then also there might be some overfitting of data whereby something is really going to skew the outcome too much and so and there might be an obvious one. So I guess in our model we looked at intelligence, it's a huge factor going forwards. So perhaps we could have run that again but taken out intelligence and that would have put a bit more focus on the other variables.

So yeah, is a bit of an art as well to look at it and one thing I would say with Einstein Analytics as well, it should all make sense. If you know your business as well, it's giving you insights and there shouldn't be too many big surprises in there but it's kind of a reconfirmation and it's helping you making the right decisions and being able to do that fast decision making and putting it right in front of where it's needed in that Salesforce record. So for example, you're looking at if anybody's done any of the trailheads, there's a super badges with Iris Analytics are based around a telecoms company and one of them is and it looks at attrition rates.

Customer likely to attrite this month and it gives you a percentage chance and so then you can actually have a dashboard which lists the attrition possibilities so you can have like you know your top fifty customers that are likely to attrite and you can then go and look at that and make sure that you're looking after them somehow or even if somebody phones up about the plan and you can see that they're an at risk person but they're perhaps not suggesting that they are on the phone call, you can, you know if you've got information in there like how old their router is, you could go hey how about we send you a new router and actually you can use I think you've got session later on next best action for example whereby you can make some suggestions to help improve the or in this case actually you want to reduce your attrition possibility. So an outcome variable can be a positive or it could be a negative, you might want to reduce the number or you might want to increase the number.

That's where it comes very useful.

Great. Well, Steve, thank you very much. This was a fascinating presentation. And I really liked the whole superhero thing. And I liked that my superhero was the best.

That was uncanny. Nobody could believe that Leonard, that was generally was not a not a fix so it must have come as a real surprise to you.

I guess that's one of my talents I didn't know about superhero creation.

But thanks for your time. It's been great talking to you at both of the X Force conferences. I just want to remind everyone as Steve did that if you're interested in Einstein Analytics, at the top of the hour, wherever you are, we have Einstein Next Best Action.

And that's with Sika Bide and Laura Gayadar and they're both going to talk about the pros and cons of next best action and including some issues with deployment about you know where you want to use it and what the limitations are. I think that that would be a great panel if you have some insight interest and coming up next we have Zach Berman who is our partner engagement manager at X Plenty, who's going to talk a bit about an org merge that he and an org merge ETL use case that he did with one of our consulting partners. So I think both of those are good. Hope that you all tune in for all the rest of the presentations. And once again, thanks Steve and thanks to the audience for joining.

And thanks Xpence and thanks Zed. Okay, bye bye.

Take care, bye bye.

‍