Original subtitles

Instructor: All right, next up I'm excited to introduce

my favorite AI visual, decomposition trees.

So decomposition trees allow you to visualize

how your data is distributed across multiple dimensions.

So, in this example here,

we're looking at how total orders break down by category,

subcategory, and product name.

Now, what I love about this visual

is that you can use it in so many ways.

You can configure it manually

for ad hoc data exploration, or EDA,

or you can leverage its AI functionality

to support things like root cause analysis.

Now, this is a relatively simple visual.

There are really only two components to it.

You've got your analyze field,

which is the measure or aggregate

that you want to explore or analyze,

and you have your explain by fields.

These are the dimensions that you want to use

to actually break down the data.

So in my opinion, definitely one of the most flexible

and intuitive AI visuals.

Let's jump into Power BI and see how this works.

So for this demo, let's go ahead and add a new tab.

We're gonna call it decomposition tree.

And we're gonna go ahead to the insert menu

and you'll find the decomposition tree right here.

It kind of looks like a tree map icon.

And to get started,

why don't we start with kind of

a hypothetical root cause analysis?

And suppose that Adventure Works leadership came back to you

and said, "You know what?

Our return rate is higher than we're comfortable with.

Can you help us start to analyze

what might be the root cause

or diagnose what's driving that return rate?"

Well, that's a great use case for decomposition trees.

This will be the perfect visual

to start conducting that type of analysis.

So what we can do is we can pull in return rate

into our analyze field like so, and then explain by,

this is where we're gonna drop in any dimensions

that we want to use to break down these return rates.

In this case, I like to think about it kind of like

a top down approach.

So I'd wanna see return rates

at a very high level like product category,

also at the product subcategory level,

also for individual product names.

And we could keep going.

We could include other fields like country,

like customer fields, and so on.

But for the sake of example,

I think this is gonna be a good start.

And that's really it from a build standpoint.

Now what we can do is look at the overall return rate,

2.17%, and we can click this plus icon

to start breaking it out or segmenting it down.

You'll see two AI options noted by these light bulbs

at the top, high value and low value,

or we can manually start breaking down these return rates

by any one of our explained by dimensions.

So let's start with a little bit

of manual exploratory analysis, starting with category name.

So by default, we can see that we're sorting

by return rate in descending order, which makes sense.

Bikes are generating the highest return rate at 3.08%.

Accessory is the lowest at 1.95.

From there, we can break down to the next level subcategory.

And within bikes, touring bikes are at the top of the list

with a 3.3% return.

And then same story for product name.

So already starting to see an interesting trend.

It looks like this Touring-2000 blue bike size 46

is being returned quite a bit,

8.33% compared to 5.56 for the next touring bike.

And this is all dynamic, right?

So we can drill into road bikes here too,

in which case we see an even more distinct trend.

This Road-650 bike is returned almost 12% of the time,

which is really notable compared to the others.

And what I love about this decomposition tree

is that it can cross filter other visuals,

it can react to filter context.

So it can be used on your report pages

just like any other chart type.

And to show you an example,

let's pull in a simple card here to show total orders,

and we'll just see how this reacts to filter changes

in our decomposition tree.

So we can drill into bikes, 14,000,

touring bikes, 2,124,

that Touring-2000 was ordered 96 times.

We can kind of filter and cross filter in a very,

very simple and intuitive way.

And because we have data

at the product level here inside of our decomposition tree,

it also means that we have drill through options available

that connect right to our product detail drill through page.

So again, really powerful, really flexible type of visual.

And that alone,

that manual analysis by itself is pretty valuable.

But the AI functionality is pretty great too.

So now instead of breaking it down manually,

let's let AI do the work and select high value.

Now, check this out.

Instead of starting at the highest level category,

Power BI has essentially scrubbed all

of our dimension fields and isolated the single item

with the highest overall return rate,

which in fact is that Road-650 bike

with the 11.76% return rate.

And we could do the same thing with the low value as well.

And that will identify the AWC logo cap

with a 1.11% return rate.

So let's show one more example,

this time with a volume metric instead of return rate.

So we'll keep the same explain by fields here,

but let's swap out return rate

and pull in something like total orders.

And now if we use that same kind of AI functionality

to identify the high value,

it's just gonna work through this in hierarchical order.

So it's gonna find the highest ordered category name

because that's our highest volume dimension.

Then it's gonna find the highest subcategory,

and then it's gonna find the highest product.

So it's nice to see the distribution this way,

but there's really not much

in terms of artificial intelligence going on

behind the scenes.

If you want to get a bit more sophisticated,

what you can do is actually head to the format pane,

drill into analysis,

and change the analysis type from absolute to relative.

Now, instead of finding the highest item

by pure volume alone,

what Power BI is gonna do is it's gonna look for items

that are relatively high compared to the average

among that field.

So in this case,

it's actually highlighted the 30 ounce water bottle,

even though this product of course wasn't ordered more

than some of our subcategories or categories as a whole,

but it was the highest one compared to

the average order size across all of our products.

So in other words,

Power BI is now looking

for the biggest relative differences

within each of our explain by fields.

So let's go ahead and just build our tree out one more time.

I wanna see kind of how the orders break down

at each different level,

and I'm gonna show you some basic formatting options

that we can use to style this.

So we head to our format pane.

Of course, we can add a title here,

and we can format what the tree looks like as well.

Let's change the connector color to dark gray,

and let's go to the bars and make those a dark gray as well.

You can change positive or negative values independently.

And there you have it, decomposition trees.

Such a great visual for ad hoc data exploration

or root cause analysis.