Grace Teoh Profile picture
11 Oct, 32 tweets, 15 min read
Listen to @JustynaLucznik’s conversation with @SchreiberLars on #AI in #PowerBI!

👇🏼Here’s a detailed thread of my key takeaways from their conversation with (with approx. timestamps) and my own reflections:

1/
09:45

To start off, @SchreiberLars quoted @matvelloso’s tweet 😜. Lol.

2/
@JustynaLucznik: The term “AI” has a lot of hype around it. Can be little overblown.

In context of #PowerBI, it is to create intuitive and trustworthy experience for users that meet their needs regardless if they’re data scientists, analysts, business users.

3/
While #AI in #BI is often thought for predictive uses, this only scratches the surface of democratising #PowerBI’s capabilities for users.

Hence, focus on wider use cases to infuse #AI through user’s entire journey to reduce barrier of entry in interacting with data

4/
and answering top-of-mind questions. To liberate analysts from having to build bulky reports so as to anticipate every possible thing on user’s mind.

E.g. Key Influencers visual (uses regression and decision trees), and Q&A Natural Language.

5/
> I agree that there’s a lot of hype and hot air surrounding #AI but not enough understanding.

I remember when I first started @lightdotlab and trying to define my niche (and flailing badly), someone once said to me that if I was only “peddling” #Excel, and not get

6/
on the #AI bandwagon, I’ll be “finished”. Being so “green” back then, his grave remark tipped me into an anxious spiral and I felt like I was doomed. Couldn’t sleep for days.

It’s laughable now that I even took his words seriously. 😑 FFS he was a HR consultant upstart and

7/
really had no business messing with my mind.

> Unless you’ve been that poor sod trying to anticipate ALL the possible questions that get thrown at you in the boardroom, you have no clue of the insane amount of work that goes into preparing all the various possible views

8/
different stakeholders can have in wanting to look at the data.

Before #PowerBI, it was a blend of:

• Experience reading how people thought and what they cared about,

• Understanding of the business and the data,

that helped me prepare the 10 key presentation slides

9/
and 831857302 hidden slides, just in case they asked. And they always did.

With #PowerBI, I find the earlier bullet points still relevant, but it’s so much easier with slicers, drilldown, and drillthrough to show different perspectives that facilitate real-time discussions.

10/
So far I enjoy the Decomposition Tree the most. It can be easily explained visually. If we had to verify the breakdown done by #AI, it can be done.

The same can’t be said for Key Influencers. While it’s a really powerful tool, it’s not something I would put in front

11/
of executives at the moment.

First, it is a black box to most business users (which can be scary to take a leap of faith to just trust it).

Second, it does take a bit more effort to digest, understand, and explain what it’s getting at.

Hence, I recommend using it

12/
in tandem with other visuals, and to cross-check with our business understanding.

14:44

@JustynaLucznik: We’re not trying to replace tools that data scientists use #AzureML #R #Python, but about complementing that story and expanding the audience for those technologies.

13/
Given data scientists are in great demand, and don’t have the bandwidth to tackle every problem, the idea is to democratise such capabilities.

Hence, they took the same technology in #Azure and built it into #PowerBI so users could build #ML models with point-and-click.

14/
16:37

@JustynaLucznik: If you have something highly business critical, where every small fraction of percentage makes a difference, then you probably want to have your models in #Azure so you get full control on tweaking everything possible.

But not everyone has the skills

15/
to optimise such #Azure models, and not everything will materially benefit from the extra percentage.

So for the latter users, #PowerBI’s #AI can democratise such capabilities to help answer questions like:

“Who should we target with our marketing campaigns?”

16/
“Who’s most likely to convert?”

18:00

VERY IMPORTANT QUESTION by @SchreiberLars:

How can I explain the <number> that Key Influnecer visual gives me? My boss is sure to ask me to explain how it came about.

17/
@JustynaLucznik: Team intentionally decided not to enable prediction at individual row level in Key Influencers (vs. #AutomatedML can score likelihood row by row).

Focus instead on top influencers. E.g. customers from region X is Y times more likely to convert.

18/
Use the visual as a HINT as to where to focus your attention on. Not trying to prescribe. But to highlight interesting patterns, point user where to look into, and then leave it to users to determine if that’s an influencer they want to action on.

19/
It’s like a shortcut to slicing and dicing the data yourself.

It’s a hard balance: Too prescriptive vs. too hands-off with messaging.

You need to do the actual slicing and dicing yourself, we just help to automate a bit of that exploration based on historical patterns.

20/
There’s still a lot of work to be done there to make it more explainable.

> Kudos to @JustynaLucznik and team for being attuned to the nuances of the Key Influencers’ use cases.

Often, if we can’t explain the <number>, it’s a no-go. Worse, meetings can get out of hand

21/
if overly enthusiastic attendees start trying to figure out the visual in detail. The more senior ranking people there are at the meeting, the lower fidelity the presentation materials should be to keep them to focused on the higher level strategy

22/
and first principles thinking.

24:00

@JustynaLucznik: On Smart Narratives, still getting feedback and prioritising what to work on. Coming up: Full coverage of templates over all visuals. Have more intelligent templates that focus on insights, not just telling the trend.

23/
Factors to determine what gets GA:

• Feature-readiness
• Number of features
• Feedback if GA-quality or not

Combo: Anecdotal feedback from Twitter, LinkedIn, YouTube, email + structured feedback from enterprise, community votes make for well-rounded features.

24/
34:10

@JustynaLucznik: There’s #AI in data prep and visuals, but not so much yet in data modelling. No specific plans yet. Usually 6-month road map. Please tell the team where you’d want to see #AI being applied in Data Modelling. They don’t want to only build things that

25/
are cool but actually solve pain points.

38:00

@JustynaLucznik: Studied Economics and Politics as she aspired to be a diplomat. Started in consulting services for #Microsoft as project management. Preferred to be in product side. Attracted to #ML ‘cos of statistics during

26/
Masters research. Hence, the move to current role.

43:30

@JustynaLucznik: Advice she would give her younger self — Don’t be afraid to ask stupid questions. ‘Cos often other people also want to ask the same questions to get clarity.

27/
Imposter syndrome is real. It doesn’t go away but it’s perfectly fine. It’s healthy ‘cos you’re challenging yourself and not letting yourself get too comfortable. A lot of people feel that way. And it’s okay.

> I TOTALLY GET THIS. In fact, the more I learn, the more

28/
I realise I don’t understand even more things. Like who am I to be training and coaching folks when I haven’t got everything figured out?

What I found helpful in managing this never-good-enough feeling was when I first enrolled in @powerbipro’s Pro+ programme.

29/
Avi shared how we can help the people just a few rungs below us in their learning journey.

In fact, being closer proximity to their position, and having been in their shoes not too long ago, actually helps us better connect with their pain points, see things from their

30/
perspectives, and thus better able to help them.

Sometimes I also remind myself that we don’t need PhD holders to teach children in schools.

That said, I’m also making use of the uncomfortable feelings that arise from having Imposter Syndrome to invest in my personal

31/
development and growth. Looking forward to the next couple of months as I start my mini-MBA designed by myself for myself to hone my #PowerBI craft. 🤓

Thanks @JustynaLucznik for your reassurance and speaking up about it!

32/32

• • •

Missing some Tweet in this thread? You can try to force a refresh
 

Keep Current with Grace Teoh

Grace Teoh Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!

PDF

Twitter may remove this content at anytime! Save it as PDF for later use!

Try unrolling a thread yourself!

how to unroll video
  1. Follow @ThreadReaderApp to mention us!

  2. From a Twitter thread mention us with a keyword "unroll"
@threadreaderapp unroll

Practice here first or read more on our help page!

Did Thread Reader help you today?

Support us! We are indie developers!


This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3/month or $30/year) and get exclusive features!

Become Premium

Too expensive? Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal Become our Patreon

Thank you for your support!

Follow Us on Twitter!