Original subtitles

Instructor: Last but not least,

we have arguably the most powerful AI visual of all,

Key Influencers.

And the purpose of the key influencer visual

is to help you understand

these statistically significant factors

that drive specific metrics or outcomes.

And this is quite a bit more complex

because now we're not just visualizing

or summarizing information, under the hood,

Power BI is running linear and logistic regression models

that are actually quantifying the relationships

between the variables in our data set.

And this can be used for a number of different purposes.

We can analyze categorical outcomes, which are discreet,

like an email being marked as spam

or a customer review being marked as positive or negative,

we can analyze continuous outcomes,

like the factors that impact the price of a home,

and we can also identify top segments

based on different combinations of factors.

So in this particular example,

we're identifying factors that are highly correlated

with owning a home,

and here we see that parents

are 1.59 times more likely to be homeowners, all else equal.

And in our top segment view,

we can identify customer segments

where this particular outcome, home ownership, is likely.

So for example, 93% of married customers

with children and a bachelor's degree own a home

compared to only 67.6% overall.

And that could be an incredibly powerful tool

in the real world

for things like customer segmentation

or defining ideal customer profiles or ICPs,

for instance, we might run a similar analysis

to analyze things like customer churn or retention,

purchase probability, and so on,

and use the results to help us do things

like refine our marketing strategies or brand messaging

in ways that we may have never even considered

without this type of data-driven analysis.

So let's go ahead and open up our AdventureWorks report

and dig a little bit deeper

into how this key influencer visual actually works.

All right, so if you'd like to follow along,

let's go ahead and add one more report page here.

We're gonna call it Key Influencers,

and we'll insert that visual

right here in our AI visuals group,

kind of looks like a lollipop chart.

Let's drag it out.

And our three build options

are 'analyzed' and 'explained by'

just like our decomposition tree.

And a third for 'expand by,'

which we'll talk about in just a bit.

For now, we're gonna focus on these first two.

And let's stick with our simple

kind of categorical analysis,

trying to understand what drives

whether or not a customer owns a home.

So that means for our analyze field,

we can pull in homeowner,

which remember, is a binary yes or no outcome,

that means it's categorical, not continuous.

And for 'explained by,'

this is where we just drop in any fields

that we think might impact the probability of a customer

being a homeowner.

So for instance,

we might think that income has something to do with it,

so we could pull in annual income,

perhaps education is a factor as well, so education level,

whether or not a customer is a parent,

maybe related to whether or not they own a home.

Similar story with marital status,

we could pull that field in as well.

And last but not least,

maybe occupation has something to do with it too.

That seems like a pretty good representative set of factors,

but feel free to add other fields

that you think might be correlated with home ownership.

So now that we've built out our fields

in our key influencers visual here,

you see that we can toggle

between the categorical outcomes 'yes' or 'no.'

So in this case, we're interested in understanding

what influences homeowner to be yes,

so we can switch that toggle.

And here we're seeing

the statistically significant drivers of ownership,

marital status, is married, is parent, is yes,

annual income is between 30,000 and 120,000,

education level is graduate degree,

occupation is management, and so on and so forth.

And we also see the strength of this impact as well.

So in other words,

when a customer is married all else equal,

the likelihood of them owning a home

increases by 1.62 times.

And if we click any of these factors and expand our visual,

now we see some additional detail

here in the column chart on the right.

And this is showing the percent of homeowners

for each category within a given field,

in this case, marital status.

And what this red line shows

is the average excluding the selected category,

which in this case, is just marital status equals single.

And this tells us that 51.12% of single customers own a home

compared to 81.5% of married customers.

And if we look at another example,

this is the 'is parent' field,

we see that 47.93% of customers who are not parents

own a home, versus 75.18% who are parents.

Let's look at one more example here

with a couple additional options.

In this case,

customers who fall in this 30 to 120,000 income range

own homes at a rate of 72%

compared to the average of all the other categories

at 58.31%.

So these strength values, the 1.62, the 1.59, the 1.23

are essentially derived

by comparing the difference in home ownership

for a specific category

compared to the average of all the others.

Now, what's also helpful

is that if you hover over these buttons,

it gives you some additional information,

including how much of the data set is represented

by this particular field or factor.

So in this case, married customers

represent approximately 54.49% of the data.

And we can actually show that too

using some formatting settings.

If we drill into analysis here, we can toggle on counts,

and that's gonna show this kind of subtle ring

around the outside of the bubble

that represents that same percentage.

Now, by default, we're sorting by impact,

which I think makes the most sense,

but you can also sort by count as well.

So that's a quick summary

of the key influencers part of this visual.

What's really cool

is that you can also toggle over to Top segments

and explore these customer segments

that Power BI has identified.

So this is Segment 1

which is defined by these characteristics,

education level is bachelors, is parent, is yes,

and marital status, is married.

And within this entire segment,

91% of those customers are homeowners.

And it's telling us that's 23 percentage points higher

than the average of 67.6.

It also tells us that there are 2,552 data points or records

that this segment represents,

which is 14.1% of the data as a whole.

So under the hood, Power BI is running a cluster analysis.

It's combining all of these factors

that we've determined here,

and it's combining them

into specific customer segments or profiles

that together are highly predictive of home ownership,

just incredibly powerful stuff that we're able to do

with just a few clicks.

Now I wanna show you one more example here,

I'm actually gonna keep this version,

and I'm gonna add another instance

of the key influencer visual.

And this time, I wanna show you a continuous outcome

instead of categorical.

So in this case,

my analyze metric will be product price, right here.

And I want to explain product price

by product cost, like so.

And I recognize, this example is a little bit contrived,

but it does help show some of the differences

when you're comparing categorical outcomes

versus continuous outcomes like this.

So now, instead of selecting

different categorical outcomes here,

we're saying what influences product price

to increase or decrease?

And you can see in the Analysis type here,

it's defaulted to Continuous.

If we change that to Categorical,

it wouldn't really make much sense, right?

We'd see all of the different potential product prices,

which in this case, is just meaningless.

So continuous is the right move here.

And you'll notice now

that the visuals look a little bit different.

Now we see a linear regression.

We notice a very strong correlation

and positive relationship between these fields,

which isn't surprising at all.

Generally the more product costs,

the higher the retail price will be.

But in this case, the slope of this line

is how we derive the strength

of these influencers or these factors.

So Power BI is telling us

when the sum of product cost goes up $516.73,

the average product retail price increases by $865.

And last thing I wanna show you,

this is where we can play

with this 'expand by' option as well.

So note that we're breaking down the data

at the product name here in this scatter plot,

but if we expand by a higher level product field

like subcategory, we're gonna see an error at first,

because it says the field in our Analyze pane,

product price is not summarized.

So it tells us to either summarize that value

or remove the 'expand by' fields.

So instead of product price, what we can do

is pull in a summarized field that we've calculated,

like average retail price,

and now we're gonna see that same linear regression,

except now each point in our scatter plot

represents a product subcategory,

road bikes, mountain bikes, touring bikes, and so on.

So obviously, a lot to digest here,

you can go much, much deeper

with these key influencer visuals,

but hopefully that quick overview

helps you understand how these are working

and exactly how powerful they are.