Original subtitles

Hey everyone so for this demo I want to dive into another one of power buys a.i. Driven visuals called

the key influencer visual.

Now instead of applying this to our current adventure works project what I want to do is show you how

we can use this visual to explore a new demo dataset that's a little bit more appropriate for this kind

of analysis.

So what I've done here is create a new power b I file it's called Power by A.I. visuals and you'll find

a completed version of it as well.

Both are available for download.

You can follow along or just watch.

Totally up to you.

But the dataset that we're working with here there's one query in this file and it's called Kickstarter

projects.

Now for anyone not familiar with Kickstarter it's basically a platform where entrepreneurs can post

projects or business ideas and try to get them funded from other users.

So let's take a quick look at what we're dealing with here.

Going to pop into the Data tab and we have one record to one row for each project got a unique project

I.D. project name.

It's categorized into subcategories and categories.

The goal amount is basically the target level of funding.

In order for the project to be successful we know when that project launched and ultimately we know

what the outcome was.

Was it successful.

Did it fail.

Is it currently live or in progress.

That's what we have in this column here.

And then we also know the number of backers the number of people who pledged or supported the project.

We know what country it came from.

We know the total amount pledged in U.S. dollars.

And we've got a year field here as well.

So I filtered this down to just 20 17 projects.

Got about forty thousand records here in this table.

But if you're interested jump in to power query you'll see the full unfiltered dataset which is about

three hundred and fifty thousand records so you can play with the full one you can cut it up different

ways totally your call.

So let's head back to our report view and we're gonna go ahead and insert the key influencers visual

looks like this little kind of like a lollipop chart looking thing with the A.I. light bulb in the corner.

So let's drop it in.

If you don't see that visual make sure you've updated your power by desktop version.

I'm on December 20 19.

So make sure your current.

And let's just stretch this out to take up almost our entire canvas.

Now here's the thing before we just start dragging and dropping all willy nilly here.

Let's take a minute and understand what the purpose or goal of this visual really is.

Because this is not your average visual.

Most chart types and templates like bar charts and pie and areas and histogram whatever they're designed

to visualize concrete data they just change the format they bring it to life in a visual form.

What the key influencers visual does is much more sophisticated.

It actually helps us understand and expose the individual factors that drive some sort of an outcome

and that outcome could be categorical like a project being successful or unsuccessful like we have in

this case or maybe it's a customer review being positive or negative or that outcome could be numerical

or continuous like determining what factors impact the price of a house or make the price of a house

increase or decrease.

That's what this visual is all about.

And for anyone who wants to kind of dig deeper under the hood into the actual statistics and data science

behind this it's outside the scope of this course.

But basically what we're working with here are regression models either a logistic regression for categorical

variables or linear regression for numerical variables.

Basically the idea is to understand the dependence and the interrelationships correlation between variables

here in our table and specifically how changes to some independent variables impact the predicted value

of our outcome or our dependent variable.

So all right that was a lot of words.

I apologize.

Getting a little heavy there.

Now let's start playing with this and see what this is all about.

So we've selected our visual here we've got three different fields or wells that we can play with and

the visualizations tab analyze explained by and expand by so analyze is the outcome that we care about.

So let's go ahead and start with that project outcome field which is categorical.

And as you can see it says All right what influences project outcome to be failed live or successful.

I want to see what impacts successful projects and select successful there.

And now my next step is to drag fields into this explained by well based on basically any field that

I think might influence this outcome.

OK.

So if I think that country might be a factor here I could just drag country drop it in and the visual

updates.

So we see a lot of information here on the left side.

We'll have a list of any key influencers any factors that power b has deemed to be statistically significant

and influential in impacting that outcome.

In this case we just have one and then on the right side we have visual representation of our entire

data set for a given field.

So we've got five countries in our dataset.

UK US Canada France and Australia and power RBI has told us that there's one significant influencer

here which is that when the country is the United Kingdom the likelihood of this project outcome being

successful increases by 1 point 1 1 times.

All else equal.

And the way that that value or that factors derived is by comparing the success rate or the project

success outcome in the UK compared to the average country.

So all other countries averaged out.

So in the UK for the fifty eight hundred and fifty six projects Forty two percent of those were successful.

When we compare that against all other countries averaged out we see that only thirty six point two

four percent are successful otherwise.

So quite an increase there and that's why when country equals UK that's identified as a key influencer.

Now note that we only have one influencer here in the list on the left but we're seeing five countries

here on the right.

And that's because whether or not a value or factor is noted as an influencer has to do with a number

of things.

For one it has to do with the difference of success rate compared to the average.

So for instance U.S. and Canada are very very similar to the average and also has to do with volume.

So you can see the count of Kickstarter projects from France only nine hundred seventy nine from Australia

fifteen hundred compared to the US which is almost thirty three thousand UK which is almost six thousand.

So those factors all go into this regression model behind the scenes in order to determine this list

of key influencers.

So at this point we're only looking at one factor country but we obviously know that there are other

things that determine if a project ends up being successful or not.

So all we need to do is think about okay what other factors might possibly be influential here and maybe

category is a factor as well.

So we can drag category in right next to our country and now all the sudden we see a totally new list

of influencers here.

In fact that country equals the US or UK influencer is now pushed way down to number seven in the list

and we have these category influencers which are outweighing it significantly.

So now we're actually seeing that when you factor in or consider country and category the number one

influencer is when the category equals comics followed by dance projects theatre projects music projects

and so on and so forth.

And you can continue with this process.

Right now we're looking at two categorical fields but we could pull continuous or numerical fields in

here as well like the number of backers or the total amount pledged in US dollars.

And now what you're seeing kind of as you'd expect are that these new fields we just pulled in are the

new biggest or most influential factors.

And it makes sense.

The number one factor is that if you have a project that raised more than four thousand nine hundred

sixty eight dollars you're almost four times more likely to have a successful project.

All else equal and that makes sense.

Same story here with backers more than 100.

Here's the thing you raise more money you get more backers you're more likely for your project to hit

its target so that all is intuitive.

That makes sense.

And when we click on one of those factors that's based on a continuous field what poverty does here

is it kind of bends those values just like you would with a histogram into different chunks.

So more than forty nine sixty eight between twenty six thirty five and forty nine sixty eight you can

kind of see how each of those bins of values compares to the average.

If we had a clear linear relationship here power behind might plot this as a scatter plot with a line

of best fit but the one thing that's kind of missing to this point is that we're sorting by default

based on the impact these influence factors here.

But what we don't know at first glance is OK.

This is an influential factor but how much of the data set does it actually represent.

Is it a large portion.

Is it a very small piece and what you can do is hover over these bubbles here and it will tell you so.

This influencer contains twenty five point thirty eight percent of the data this one contains 19 percent.

This one down here contains under 1 percent.

And what you can do to make this a little bit more clear is actually go into the format pain here drill

into the analysis options and you can enable counts and what that does it's kind of subtle but it adds

that little ring around each of these bubbles which represents the percentage.

Right now it's based on an absolute percentage so 25 percent actually looks like a quarter of the circle.

You could change that to relative which makes the largest factor 100 percent and then index is all the

other ones accordingly.

Sometimes that can make it a little bit easier to see and interpret.

And the last thing that that actually does is it gives us an option here to sort either by Impact which

we're doing by default or we can sort by the count and now we see that actually this factor here which

isn't a very big influencer is a big piece of our data it represents almost 26 percent of our dataset

now.

Last thing to cover really quickly here before we move on to the next demo which is going to cover continuous

variables is this top segments tab and what the top segments Tab does is it actually runs a cluster

analysis behind the scenes and what power behind doing here is it's combining factors into segments

or populations that seem to have very high level of influence.

So you can click on any of these to see more information about them.

So segment 1 is defined as a segment where the number of backers is greater than one hundred and the

category is not games.

OK.

So within that segment almost 90 percent of those projects were successful.

And that's compared to an average of only 37 percent.

So a very very influential very successful segment that we've defined here can also see that it contains

about sixty four hundred records which is about fourteen point six percent of our data.

You can learn more and explore down here but that's basically the idea you can click through see which

types of segments power eyes defining.

And this can help you understand things like you know who your target audience really is you know maybe

you realize that middle aged women in the Northwest are your best customers and you're really not resonating

with men in the south.

Those are the types of insights that this sort of segment analysis can really enable.

So there you go.

That's the key influencer visual looking at a categorical outcome like we are here in the next demo

we'll keep this example going and instead of looking at a categorical outcome we'll pull in a continuous

variable there and check out some of the differences.

Stay tuned.