Generative BI in Amazon QuickSight

Right. Hello everybody. Hello, reInvents. Are you guys having a good conference? Yes. Yes, I love this conference. What I like most about it is the opportunity that we have. And I think it's a special thing with AWS to develop these connections, these individual and personal connections with people so that we can really understand each other's situation. And I think this is the magic that helps us to build better products in the future.

Many of us here are, are building today. So I'm Zach, I lead product for AI/ML at uh QuickSight. And uh how many are familiar are you familiar with QuickSight? Yes. Ok. About half. All right. That's great. Uh so, uh I've been doing this uh sort of building software thing for 25 years and uh a lot of companies uh that you may know Microsoft Salesforce Tableau uh and some start up work as well that you probably would not know.

Uh and uh the reason that I do this work is because I believe in the power of data as a tool because with data, we can, we can see something we can accept a common truth and that maybe if we can see the same thing, we can agree with each other about a question which can help make the world a little bit better place.

So the question is we have all this data that we have been collecting all of us together for years and years and years and years. What can we do with the data to help us improve our business? And so uh for you a question, uh what do you think it is that gets in the way of people being able to use their data effectively? Anyone sorting through it, sorting through it, sorting through it? Absolutely. Oh, it's echoing. Ok. It's too loud. I can it be quieter? Is that, will that be better? Oh, multiple systems. Yes. Yeah. Your data is can can be spread out and it's like inaccessible because it's not organized. It's not easy to get to. Yeah, this is true.

So what Gardner says is that the thing that prevents people most is a data skills and staff shortage. Now, the evidence that's provided is, hey, 68% of the time business analysts are building dashboards reports. Hey, they're doing undifferentiated, heavy lifting, work, rebuilding things that have been built before the business users who need to use their data in order to drive a decision. These business users, 75% of the time, they're just not confident enough in the data. They don't believe the data don't quite understand what the data is about.

So I look at this and um I see a different problem which is we need better software. So that's why we created Amazon QuickSight. So QuickSight is a general purpose BI service. Uh you can use it for all BI needs uh you know, it's dashboards reports, uh analytics and uh it's of course, because it comes from Amazon automatically scalable. So, you know, have to think about, hey, what are all my different sort of uh sort of resources that I have out there? Am I gonna be able to, you know, provide the load that I need to be able to provide for my service? So it kind of happens automatically and it's inbuilt with all these generative AI capabilities. And this is one of the things that makes it really different from some of the other things that I've worked on and we're gonna talk about that a little bit today and then it also fits itself uh on cost to your workload.

So a lot of customers are using QuickSight today. It's, it's more all the time. I'm just astonished. I was here last year, uh relatively new to AWS. And, and, and it's amazing to me how rapidly we're growing, like how many more customers we have and then even just a year ago. And so with it, of course, you can create dashboards and, you know, this is a very common thing that people do with their uh analysis tools, uh you know, interactivity, different kinds of visualization.

Um you can also build applications. So it turns out that this is actually the bigger part of uh our business at QuickSight is that people build these custom apps, they embed analytics in other things and then maybe make them available to their customers or externally or maybe they make them available inside that that also happens.

So um another capability that actually launched last year at reInvent is Paginated Reports. And so these are inbuilt and there are a bunch of people that are like doing, you know, a sort of legacy to cloud migration kind of questions and they really haven't been able to move their, this like last thing is Paginated a report. And so that's why we did this so that it's within your BI service. And so it can leverage the existing sort of uh KPKP I data that you have uh that you're using for QuickSight.

So um the part that I work on is our AI and ML and uh key investment that we have here is QuickSight Q and so this is our natural language service and uh Quick IQ has been around for a little bit. So uh it's like 2020 is when we first launched and Quick Q helps business users to ask and answer questions of data.

Last year reInvent we launched Forecast and Why those are capabilities that are a little bit more advanced than you might expect, but they are specifically targeted to help these business users answer questions easily that are kind of common needs. So we've been building QuickSight. Q we've got all of this additional capability and modeling that we've been doing around. Hey, can you detect outliers, you know, can you, can you figure out tops and bottoms like additional kind of uh ML that we've built over the years and at the same time, there's this AI revolution, you know, there's all of this new capability that the fun the foundational models, the large language models bring. And so the question is how can we combine these things together in order to do more for people to help them with their data?

So these guys uh help a little AI robot, you guys uh what can they do? Like what are they good at? So it turns out I've been looking into this, I've been thinking about it a lot. Uh there are certain things that they are very good at and they're best for. So for example, they can assist us in our work by providing additional information by, by doing work to summarize and just help us interpret and understand information. They can even do some really magical things to generalize ideas. And this is maybe one of the more surprising things that they can do, but they can make sort of cognitive leaps where they can say, hey, you know, this information may be related to that information over there that you weren't even thinking of. And so that's maybe one of the more interesting things to me, they are not the perfect solution for precise knowledge in many cases.

And so what I mean by that is, for example, today, we don't really ask these large language models to do math because we know that they're not very good at it. So we look at other solutions like generating SQL or generating code or you know, other things outside and this is an evolving space. And so hopefully they will, they will get better at it. But today, we don't do that. And in data math is very important, uh we also don't really use them for well-defined tasks.

So if you have a model that you have been building and you know, it does forecasts and it knows your data and you trained it to do a specific thing in a specific place, probably going to be better than asking the large language model to do a forecast. Uh and then finally, we need to keep the human in the loop. So this idea of verifiable truth in the end, we don't want the AI just making all the decisions like we want this to the element of human judgment, which is very important to remain.

Ok. So with that, we are announcing uh announced actually yesterday by Adam that we're bringing generative BI capabilities with Amazon Q in QuickSight. Uh there are three sets of experiences that go with this. So we have uh an AI accelerated dashboard offering that helps business analysts quickly build dashboards. We have AI answers to questions of data and this is for business users who need to inform a decision. And then we also have this uh AI assisted data storytelling, which is a new kind of thing that helps business users to examine their data and then share it with other people in order to drive decisions as a team.

So the dashboard authoring experience includes three capabilities. You can build visuals, you can build calculations and then you can refine those visuals to get a precise experience, the AI answers to questions of data. So there are two parts to it. There's an executive summary capability. So Q can summarize key insights out of dashboards can do that kind of a automatically. Um and then we have an enhanced Q&A experience. And so uh we've learned through watching customers try to use this experience, business users kind of outside the context of the, the BI team. We've learned that it's sometimes hard for them to know even what questions they can ask. And so we're doing some things, we suggest questions they can ask, we show them the scope of data in a very simple format so that they can take a look and understand what's in there. And then when they answer a question, they are sometimes confused about like what this answer means. So we're presenting a more complete contextual answer, that's a visual in multiple parts, explaining the data. And we even describe it in uh in narrative so that they can see key insights. And then finally, we have support for vague questions. And these are, it turns out are the kinds of questions that business users typically ask, like, how can i make this better or how much does that cost or things that are kind of open ended or, or sort of difficult to match to specific fields in a database?

Ok. So the last part is our AI assisted storytelling. So this the first part of this is uh being able to actually create a story. So a story is a new kind of thing. It's not a dashboard, it's not a report, it's more like a document or slides. And so then we can actually generate these things directly from your data and we can make them available to people uh automatically, we can also then kind of refine that content and improve it using the AIs. And then uh we, because we're like within the BI service, we get to participate in the existing governance model. And so that way the data isn't sort of leaving to go to some external document store.

All right, let's take a look at what this is. Switch over to my demo here. Ok. Ok. Uh so, uh were any of you in my session a little while ago? No? Ok. Great. Uh so, we, we have two demos here but I'll go with the sales demo. Uh so, uh here we have QuickSight and so along the left, this is just a field list. Uh these are, these are fields that people can use and then uh at the top you'll see there's a new entry point to ask you to build a visual. So I'll go up there and then I'll just tell you what I want the outcome i want to achieve. So sales by product as and city is a map. So you can see now our sales are pretty broadly distributed. I want to look at the profit though and here you'll see the profit is kind of flat. So that's not very good. So I wanna analyze that a bit.

So I'll create a calculation and here I can just say what I want the calculation to be about profit ratio and it can automatically come back in QuickSight's calculation language. So now I've got this calculation. What can I do with it? Well, I can use it with the other features. So I can say, hey, I want to use that profit ratio and then I can combine it with other fields and then I can show those visually as a scanner and thereby see that I've got a few products that are performing really well on profitability and others that are performing really well on sales, which is not ideal.

So let's examine that a little bit. Tell me which products are worth the most. Now, this is kind of interesting because this is one of those vague questions that I was saying, like business users want to ask this, right? But what is worth mean? So in this case, we've interpreted this to say total profit by product. But at the bottom, you'll see there's a, did you mean there are other things in your data that can match that word worth? So we've got sales, we've got profit ratio and here we're helping these business users to see what their question uh matches to in the data.

Ok. All right. So if I move forward, which I can do by hitting the play button, we'll add this to our analysis and I'll show you the final capability which is refining these visuals. So uh here I can ask you to make some changes visually, change it to a table, add region, color profit uh using conditional formatting and it can just do that kind of stuff automatically.

So now we're going to switch over from the business analyst building the dashboard to the business user. So now I'm a sales manager and I have questions about like, how we can improve our sales performance. Uh so what I'm gonna do to start is I'm gonna just take an executive summary to see, hey, how are things going? And sure enough, I can see right away that the products that are driving sales are not the same as the products that are driving profit. So that's, that's questionable. I want to look into that and figure out what's going on.

So when I ask you, it not only suggests questions I can try but actually shows me specific data. So customer details are in here and order details. And so I can ask about for example, sales and profit for Contact Match our top selling product in a PJ, which is one of the regions that I oversee. And I can see here a really troubling trend on on the right. I've got a breakdown of all of the sales over time and it's comparing to the profits. And so you can see the sales are going up, the profits are coming down in the center. I've got a total of all sales and profit and at the bottom, I've got order details. So if I want to understand if i wanna believe or or sort of gain some confidence in this data, I can see the specific data that's being used in order to generate this answer.

So now if I want to understand how it's coming to specific conclusions that it's described on the side, I can actually just hover over the text to see where that insight comes from. So you can see here looking at year to date totals as of July 17th, the total profit has decreased substantially 3000%. Right. That's not, so that's not good. And it doesn't say that anywhere else on the screen, right? It's just looking at that uh chart over time and it's saying, hey, you're, you're, this is not a good situation to be in. So it's bringing those additional insights out.

All right. So let's uh examine a little bit more deeply. Um you know, these regions that I, you know, i'm looking after, so which are the worst performing. So here we're showing that vague question support again and, and it's chosen by default to go with sales as a performance metric, but that's not what I want. And it's ok because it's saying, hey, did you mean i've got this alternative? And so if I just click that, it'll just change the question. So now I'm asking about profit and it'll produce a different visualization. So sure enough, I can see APJ when it comes to profit is, is significantly underperform. So we should do something.

So I'm gonna take this finding and share it. So I'm gonna build a story. When I build a story, I get to choose between a document or slide format and then I just describe what I want the story to be about. So here I'm gonna say build a story, explaining profitability trends, give me ideas about how sales managers can improve their performance, I'll pick a couple of pieces of data. I want to go into it and then Q will build a story in multiple parts, examining the business.

So you can see here, it says profitability across regions, a data driven approach to enhancing sales performance. That's what it's going to tell me about. And as I go through here, it's gonna bring out specific visuals from my data and it's analyzing the data, interpreting the data and coming to a conclusion.

So it says, hey, 70% of sales come from these top three products. But a certain products that are like low sales, but high profitability are an opportunity because we've seen some success in other markets. I can include additional data. like for example, I can get some uh details from marketing channels and our effectiveness there to include here, I can ask the AI to refine the document. So, hey, make this into bullets instead that'll be easier for people to read. I can apply a theme which is very straightforward, just kind of uh gives it a little bit of a pop. Then I can quickly review the final product before I share it out to, I'll share it to Shannon who uh you may have seen uh this morning in Swami's keynote.

So uh so that's it. These are the capabilities that we're bringing. Uh let me switch back here. So uh so just to recap these three things. So there's the AI powered dashboard authoring experience. And this is for uh the business analysts to more quickly create dashboards. There are the AI answers to questions of data and this is for business analysts to uh for business users. I'm sorry to quickly answer questions of data in order to inform decisions. And then there's the AI assisted data storytelling which helps people to examine this data and then share with a team in order to bring people to a shared decision.

Uh so you can learn more about this at generativebi.com, but I'll give you a minute to take a little photo. Yeah. Uh and uh there's a bunch more going on with QuickSight. If you're curious to learn more about what we're up to a variety of sessions continue. Uh and then uh please, if you have a chance and you're uh curious and interested, please join the QuickSight community. This is a great group of people. People are, are very helpful. They, they really want each other to succeed and they have great advice. So, uh I recommend just going up there, you know, asking questions you have, seeing um you know how you can participate and contribute.

And, and thank you for coming. Please connect to me on LinkedIn if you'd like.