All language subtitles for 004 Q_A Visuals_en

Chris: Next up, let's talk about Q&A visuals.

These allow users to explore and visualize data

from your model using intuitive natural language

prompts and queries.

So for example, you might type a prompt

like show me total revenue by country,

Power BI is gonna take its best stab

at interpreting what you mean

and produce the most relevant value

or visual like the map that we see here.

And the important thing to know about Q&A visuals

is that they're only as useful and only as accurate

as the data model behind them.

And they typically require a significant amount

of training before they're truly effective.

So what exactly do I mean by training?

Well, inside of Power BI desktop

we have a few different options to refine our model

and teach Power BI how to handle different types of queries.

First, we have field synonyms.

This is where you can add human readable synonyms

for the tables, columns, or measures in your model.

Next, you can actually review questions

or real prompts from users to help refine your model.

Note that this requires that you publish your report

to Power BI service.

You can also teach the Q&A

by adding example questions to try to find missing

or misunderstood terms so that you can add new synonyms.

And last not least, this one's less about training

and more about the user experience.

You can pre-populate a list of suggested questions

that users can choose from by default.

So with that, let's jump into Power BI,

practice building a Q&A visual ourselves.

All right, so for our Q&A demo

we're gonna insert a brand new report page

and let's call this Q&A.

And on this page we're gonna insert a new Q&A visual

from our AI group right here.

And let's go ahead and resize things.

And what I'm also gonna do is add a matrix

to this page as well so that I can kind of spot check

the values from the Q&A visual.

So let's start with category name on rows

and total orders on values.

So when it comes to interacting with the Q&A

it's as simple as just typing what you're looking for.

So for example, total orders for bikes,

it spits out a card with 14,000.

And if we check that against our matrix, 13,929 looks good.

So what if we get a little bit more specific here?

Total orders for blue bikes.

Now we get 1,263, we go to our matrix

and we pull in color on rows,

we can drill into bikes and blue bikes 1263 looking good.

So we're two for two off to a good start.

Now let's get a little bit more detailed.

Total orders, four blue bikes by start of month.

And I'll see a couple interesting things happen.

First of all, Power BI produced this

monthly line chart, which looks good.

This is accurate.

And you'll also notice that we see this red underline.

So the blue underline means that Power BI identified

what these terms actually mean.

The redline means it's not entirely sure

but it made a guess.

So in this case, start of month can actually be found

in a few different places.

The calendar lookup table, the rolling calendar

or inside of our date hierarchy.

So you'll notice beneath the prompt it says showing results

for total orders for blue bikes

by calendar lookup start of month.

So that's the guess or the assumption

that Power BI made to produce this visual.

And in this case that's accurate.

But if I were to type something that's just

complete gibberish like this, I get a double red underline

which basically just says we give up, right?

We don't know what you mean.

Try a different term or add this one

by creating a definition.

And that's one of the ways that you can train your model

which we're gonna talk about in just a bit.

So let's get rid of that gibberish

and expand this query one more time.

So total orders, four blue bikes by start of month

but now we don't want a line chart.

I want a column chart.

I could say as column chart.

And there we go, visual updates.

And what's even better is that just like any other visual

we've got cross filtering effects

as we select certain values in the chart.

So notice our matrix updating with each

of these month selections.

Now some of you out there are probably thinking,

well Chris you're typing very specific terms

because you are so familiar with the data model.

But an average user probably wouldn't be using terms like

total orders or start of month.

They'd probably type something that's a little less clear

or a little more general like orders by month, right?

That's a much more realistic query

from a user who's not as familiar

with the underlying data structure and data model.

And this is where things get interesting.

And here's a good example, right?

With this query orders by month initially it looks accurate

we see a line chart, but the problem is that

there's only 12 data points here

and we've got two and a half years of data in our model.

And if you look closely, you'll realize

that Power BI is aggregating the order volumes

for each distinct month number.

And by doing so it's creating this kind of meaningless trend

that's quite a bit misleading to be honest.

So I'm gonna show you how to solve this in just a second.

But another example of where a general query

can cause problems is something like revenue by category.

So what we see here we're hoping to see kind of

product categories, right?

Accessories, bikes and clothing.

Instead, we see this bar chart broken down all

the way down to the skew category level

which in this case is not what I was intending.

So that's a good segue into the training options

that are available to make these Q&A visuals

more accurate.

And we can access those training options

through this gear icon here that's gonna open up

a brand new window where we can access all of these

different training options.

So let's start with synonyms.

This is where you'll be spending the most time

training your models.

And the first thing to note here is that you can

actually hide entire tables or individual fields

from the Q&A.

And this is great if you have things like parameter tables

or fields that you never want to be shown

in a Q&A visual.

You'll also notice that Power BI often takes a stab

at creating synonyms for individual terms.

Month short, synonym is short, and some of these

really don't make much sense at all.

In this case, start of month has a synonym of start.

Same with start of quarter, week, and year.

So while we're in here because these fields

are the ones we typically use for time series analysis

really want to clean up these synonyms.

So start of month, let's update

and add a synonym called month.

Start of quarter let's get rid of start and add quarter.

I'm gonna do the same thing for start of week.

And for start of year.

So that's gonna help clean things up quite a bit.

I can also go up to that month field,

which was causing us to kind of aggregate

those order numbers by month number.

And I can say, you know what?

We only want to match with that start of month column.

Let's just hide the month field

from the Q&A visual entirely.

And we'll jump back and see what impact that had

in just a second.

Last thing I want to do here while I'm managing my synonyms

is try to understand why that category query

was mapping to skew categories.

So that skew category field lives in product lookup,

can scroll down to skew category.

And you can see that Power BI had automatically

created a synonym called category.

And that's why we were mapping to this field in our query.

So let's delete or remove that synonym there.

Let's collapse this table and head to our category table

and make sure that category name which is the field

we wanted does have a proper synonym for a category.

So we'll add it here.

And then I actually know there's one other place

where that word category shows up

and that's in our customer table in the education category.

And you can see we've got a category synonym

for this field as well.

So we're creating all sorts

of confusion with this Q&A visual.

Let's remove it here.

This is really a good real world example

about how messy this process can be

and how much effort it typically takes to properly

train a model and get consistent accurate results.

So I could go through all of the tables and fields

here in my model, but I think for now

that should do the trick.

One thing to call out here is that you've got this

managed terms tab and that's where you can see

all of these synonyms that you just created.

So with that, let's jump back to our visual

and check this out.

Now, our revenue by category prompt returns

what we'd expect, product category breakouts,

bikes, accessories, and clothing.

And if we go back to that other prompt,

we tried orders by month,

which gave us the aggregated totals.

Now we're mapping to that start of month field

and we're showing the proper time series trend.

So those updates we made by adding those synonyms

had a really significant impact

on how this Q&A visual functions.

Now another feature I wanna show you in here

is the review questions option.

And again, this is only relevant

if we've published the report to service,

but this is where we can see actual questions

that users asked.

We can also see if they gave the results a thumbs up

or thumbs down rating.

And we can see through these underlines

how well Power BI was able to interpret the prompt.

So in this case, we see that a user typed

orders by product type we don't have any field

in the data set called type but they're most likely looking

for subcategory level data.

That might be a good hint for us

to go into our field synonyms and add some new synonyms

to make sure this type of question is generating

the proper result.

Next up, we've got this Teach Q&A option here.

This is where you can kind of test different queries

and help Power BI define terms that it doesn't recognize.

So here's one example, orders for bicycles

and we submit that prompt.

Power BI is gonna say, Hey we have no idea

what you mean by bicycles.

Can you help define it here?

So in this case we can say bicycle refers to

category is bikes.

You can go ahead and save that once it loads

based on my preview, that actually looks correct.

And now in managed terms, you'll see that update right here.

Bicycle is defined as category name is bicycle

if category name is bikes.

Now last thing to show you here is

the suggest questions option.

And this can be helpful if you want

to kind of train users how to ask effective questions

or if there are specific questions that you want people

to ask about the data.

So for example, orders by month we can add that.

Could do anything here like return rate by category.

These are just examples.

But let's go ahead and save those two.

And now when we close this box and head back to our visual

and if we clear this prompt you see

those two suggested questions that we just added.

Return rate by category, orders by month.

Now last but not least,

if you kind of have a visual that's been produced

that you really like and you're done interacting

with Q&A, you can click this button here

next to the gear.

And what that's gonna do is convert it

into a standard visual so we're no longer a Q&A visual.

We're now a standard line chart.

So there you have it.

I know that was a lot to cover

but that's your crash course in Power BI's Q&A Visual.