Name: Tracy Daugherty, AWS
Uploaded: 2024-10-16T18:56:00.000Z
Duration: 19 min 59 s

Clips
More from AWS Generative AI Analyst Summit

Tracy Daugherty

General Manager of Quicksight

AWS

play_circle_outline Utilizing AI-Assisted QuickSight for Effortless Dashboard and Report Creation

play_circle_outline Simplifying data preparation and visualization for users, improving efficiency and ease of use

play_circle_outline Connecting to on-premise and cloud sources, expanding data source capabilities

Info
Transcript

Tracy Daugherty, AWS

Tracy Daugherty

General Manager of Quicksight AWS

Tracy Daugherty, General Manager of QuickSight at AWS, explains the business intelligence platform as a tool for creating visually appealing dashboards and reports. QuickSight features a Q assistant for natural language processing to simplify data preparation and visualization, catering to users of varying technical expertise. The platform aims to bridge the gap between different user personas within organizations. Tracy highlights challenges in data processing and emphasizes the importance of context in providing accurate answers. QuickSight integrates with ... Read more

explore Keep Exploring

What are the different aspects of Amazon QuickSight and how does the Q feature help users in building dashboards, reports, and visualizations without requiring technical knowledge? add

What are the primary ways in which the product wire together the intent of what the person wants to build with the data? add

What are the different types of data sources that QuickSight can connect to and how does its cloud-first approach set it apart from other tools in the market? add

bolt Powered by CUBE AI

Tracy Daugherty, AWS

>> Okay, we're back here at AWS On the Ground. Dave Vellante with Rob Strechay. We've been going deep dives all day here in lovely Seattle. Tracy Daugherty is here as the general manager of QuickSight at AWS. Tracy, good to see you. Welcome.

Tracy Daugherty

>> Well, thank you. It's good to be here as well.

>> So, could you describe the Qs and where you're focused, Q for QuickSight, QuickSight for Q?

Tracy Daugherty

>> Yeah. Yeah.

>> Help us understand that.

Tracy Daugherty

>> Yeah. So, the way I would say it, is QuickSight as a whole is our BI or business intelligent offering. So, think of it as building dashboards, reports, sharing data from databases of data, where-

>> Viz too, right?

Tracy Daugherty

>> Yeah. All the visualization to really make data assumable by the average business user that doesn't know behind the scenes the technical part of it. The Q to your question, there's a Q in QuickSight that enables various aspects of the product to be better. So, think of it as the assistant within the product for the given persona. So, if you're the person building dashboards or reports, Q helps you do that. Through natural language, you can ask it a question and it'll build a visualization. It'll wire it to the right data, and put it on the dashboard, and help you produce that, or edit or change it. You could also do it for complex calculations which that person would do. If you switched over the persona to the business user, the person who consumes it, it's very much of, "Hey, how do I ask questions of my data that I have no idea how it's stored, what it is, but I want to ask a question and get an answer that augments my dashboard or report." You do that. Or we have a third thing about creating a story, where you can actually... Think of it as writing a presentation or a narrative from your data that's in a database or a data warehouse. And that's for the end user to do that. So, even in the same product, we have different flavors.

>> A very wide spectrum of capabilities.

Tracy Daugherty

>> Yeah.

>> So, for the hardcore analytics person, they can dig in. You've got sort of what I would call essentially a low code through natural language processing capability. And for the business user who doesn't want to know any of that stuff, just tell a story with data.

Tracy Daugherty

>> That's right. Yeah.

>> And it differs from the other Qs. Because there's Q for developer, and Q for business, and Q for QuickSight that... And those are really more like focused agents for each of those personas, right?

Tracy Daugherty

>> That's right. Now, there is some cross-

>> So, I think the Q for developer, like its name suggests, is very much for a particular type of persona that helps them with that?

Tracy Daugherty

>> The Q for business ultimately has some crossover, and you may as a business user experience both. You may have QuickSight and you use the Q in it. You may also use the Q in Q business. And going forward, we'll try to blend the lines between these things to make it most valuable for that end user of what they're trying to do.

>> And when you look at Q for QuickSight, again, when you said... Again, it's the business intelligence using artificial intelligence, all the intelligence. And when I look at that, the different personas that are involved in creating dashboards, creating end of quarter, end of year types of reporting and things of that nature, it's pretty complex. You have a lot of data engineering in there. You have some maybe platform engineering. How does Q in QuickSight really simplify that for customers?

Tracy Daugherty

>> Yeah, it does it today in really two primary ways. It really does wire together the intent of what the person wants to build with the data. So, if you think about it today... And the product's actually pretty amazingly simple to use already to create a dashboard in a report, but you have to know what data you want to connect to, what chart type you want, where do you want to place it. There's a lot of things you still have to, from a design perspective, because it's individual. But that part of just saying, "Hey, show me monthly sales," and then have a visualization pop, and have it wired to the right thing in the database or the data warehouse, and then pop up and exist is a huge time saver. So, say you do that for 10 different visuals, you have it on your dashboard, you could do this literally in minutes. Now, maybe you want to edit all of them. You want to add some filter to all of them where you ask the question. And so, we give you the opportunity to go back and tweak those things as well. And you can do it through either the natural language or you could do it standalone, the way you've always done it. So, I think of it very much of, one, helping the person that does this job do it faster or more efficient, and also opening the doors up to somebody who may be less skilled at it to get to the end point much faster. And that's more on the technical side. You'd also had the other question on the business user, which is really about 90% of the users that consume this stuff. And the BI space has always had this fallacy, in my opinion, and where they say self-service. And my running joke is self-service meant somebody on the tech side said, "Well, here's something I don't want to do, so I'm going to just hand it to end users so they can do it," which is not self-service. The trick to self-service is getting people to do the thing they wanted to do. And we, I guess you would say, changed the wheels on it when we found end users really looked at a dashboard and report, which usually causes them to have more questions. In the world today, that was, "Okay, now I got to go talk to the data engineer to go find the answer to this question because it's not on my dashboard." We wanted the solution to be, "No, I can ask the question myself and get back this multivisual response that gives me an answer that I can believe in and trust in. And then, I can pin and save the answers so that I could use it later."

>> I love . I mean, I love that vision of that simplicity. But if I zoom out and think about just the complexity of the data pipeline, I've got to ingest it, I've got to transform it. I've got my data quality engineer. I got to... Then, I have to bring in the metadata, I have to bring in governance. I mean you know this well.

Tracy Daugherty

>> Yeah.

>> But it feels like GenAI is simplifying your world and hopefully all that other complexity in that data pipeline. I feel like the data pipeline that we know, and not always love, is going to transform here in the next, I don't know, three to five years. Do you buy that premise and what role do you see QuickSight playing in that simplification and transformation?

Tracy Daugherty

>> Yeah, it's a great question. I do buy it because what you find is 70% to 75% of the time gets spent by organizations in that part of arguably preparing the data. And so, I think you'll see a lot of investments that'll help us make that better and more efficient. QuickSight has its own data prep area. We want to improve that, and change that, and take advantage of that as we go through it. The other thing you'll find is preparing data for natural language is slightly different than preparing it for dashboards and reports. Because a fun fact for you, the average sane question is asked about seven different ways by a human. And so you have to interpret that to get to the same answer. And so, how do you prepare the data so that it can give you ultimately the same query for the different variation of the question? So, there's a lot of work that we can continue to do on that.

Tracy Daugherty

>> I mean, LLMs have gotten so much better at understanding the query, and just in a couple of years. I don't know how. Maybe it's chain of thought or it's prompting itself. However it does it, it's like magic. But very clearly, we were using the example before, Tracy, of how you used to put a plus sign when you were doing a search. And this plus that, you were plussing key -

Tracy Daugherty

>> Right....

>> think about that anymore.

Tracy Daugherty

>> That's right.

>> So, GenAI seems to be going through a much more sophisticated but similar transition-

Tracy Daugherty

>> Yeah, that's right....

>> or evolution.

Tracy Daugherty

>> That's right. And you'll find that even on the side of the business user side, there's even more pressure because in general it's an audience that's not very forgiving. If it doesn't work the first or second time, people disappear and go back to their old ways. So, you have to do that. We've learned a lot in some of our experience of doing this, just the last three years, of how we let people ask questions, how we answer them back. It's quite a bit different than we did three years ago. It's not a ask a question A and get an answer B. That's not really how the human works. And we found users don't really know exactly the question they want and they want to be a more holistic answer. And then, you find some of the common things that are super hard. I'll give you an example, the word "quarter". Financial terms, it means three months. In a football game, it's 12 or 16 minutes. Both are obvious in context. Neither are ever mentioned in the database because they're so obvious, right? So, this is one of these things where you have to really know more of the context as a whole. And the data's got to be, I guess you'd say, massaged a little bit so that you can get really, really good answers for those things.

Tracy Daugherty

>> Yes. I like the word harmonize.

Tracy Daugherty

>> Yes.

>> Right?

>> But off of the harmonization and aspect of that, I think a lot of organizations, they look at their BI solution. It's not just all about one database and getting all the data to one data lake. And because you're not the only one in town, how do you look at that integration, not only within AWS ecosystem, but with outside the ecosystem as well?

Tracy Daugherty

>> Yeah. Yeah. Another good question. The thing is for QuickSight, we obviously do it well with the AWS sources. But we also, our next biggest thing, is with what we'd call on-premise sources. So, you can pick your famous databases, and data warehouses have been around for years, and we will connect to them and have a VPC to enable us to bring that in. And the nice thing over time, what people do is they get those off of premise and put them in the cloud. And then, it joins together and it's more efficient that way, but you also... There, I mentioned things in databases, data warehouses, and data lakes. But that's not all the data either. There's more data sources that are out there. And so, we look at that, of how do we expand it more and more. And we can do this. And kind of part to your question of how are we different than some of the other tools that have been around, is since we were built in the cloud first, we weren't built from an on-premise solution, we think quite a bit different. We scale quite a bit different. Our performance is quite a bit different. Because of that, we can actually connect to more things, and do it in an affordable way, and scale it with great performance as we go along. So, that's another piece that our architecture... I would always say, when I first joined the team seven years ago, when I was just launching, I would say this phrase that we were architecture rich and feature poor. Today, I would say we're still architecture rich because you can't change that quickly. But we've really, really worked so hard in all the features and capabilities. And GenAI is just another piece to build on that architecture and expand it.

>> You just gave me an idea. The thought, when you think about the BI industry, is it's huge, probably about 50 billion, let's call it. And it's been built up to help us get through all this complexity of databases and then serve it up to the business. When you mentioned being able to go on-prem and connect to on-prem data, whether it's on-prem or in the cloud, Werner Vogels talks about this massive distributed system that the cloud has become. And we always talk about the fact that a lot of these activities are asynchronous. It's an historical system of truth, analytic system of truth. And we're excited about the day where you can actually have a digital representation of your business, and then I can ask almost in real time a question, as a business user or other, about the state of the business, where you're integrating with the supply chain, and your CRM, and your financial system. Do you see that day in the horizon?

Tracy Daugherty

>> I do. I do see that day, but there's a lot of complexities that come into it. And not very many of the complexities are technical. It's more of who has the permissions to look at what. So, when you take your example, there's usually a set of people that have permissions to one thing, versus another, versus another. So, the challenge tends to be how do you connect the dots on that way to the business user who wants to ask the question. And do they have the access to all those things or not? And that tends to be where more of the complexity comes in. So, things like governance, and lineage, and all these things matter, but there's also a cultural effect of company by company, what makes sense for them. Everybody says they want to get there, but it requires this next level of understanding. But

>> Now, something popped into my head, which not surprising. Like when you bring up the word lineage and then you have AI and you have BI, explainability and trust comes up a lot. How do you talk to people about how they can trust Q on QuickSight to really bring back the right answer? Because they're like, "Hey, it wasn't wired together by my data engineer, I need to believe in Q on QuickSight."

Tracy Daugherty

>> Yeah.

Tracy Daugherty

>> Well, there's a couple of things we do. And one part is there is some data preparation that takes place on the back end by the engineer to go through to do some things to make sure that they can provide an even better tailored answer for their thing. The second thing is we don't let the LLMs do the math, or anything along that. We do the math because it's our data. So, we get that right on that side. The third thing we do though is big on the end user earning trust, is how we represent the answers and the questions. We have a lot of things on, "Are you sure this is what you meant? Do you mean this?" Because what we found in our early , we would be very highly accurate on the answer, but the person interpreted... Their interpretation of the question was different than our interpretation of the question. So, we do more things on the experience side. It said, "Did you mean this? Is this what you meant?" We try to work really hard to clarify what we're answering. And then, we don't give it in a single answer. We give multiple visuals and a text-based answer, so that there's enough information there they can go, "No. Yeah, that makes sense. It may not be exactly what I asked, but it's clearly accurate information to go through and do that." And then, we allow them to pin certain things that they like or not like. We also say, "Hey, were you satisfied?" with a thumbs up or thumbs down. But honestly, only a small percentage of people ever plug those in. So, you have to just do a good job anyways. I'm giving you kind of a multi-prong answer, but we have to think about these different factors to earn that trust.

Tracy Daugherty

>> So, staying on or coming back to the governance question, how do you integrate with governance solutions? Are they AWS IP? Is it third party? I mean, there's all kinds of governance. There's open source governance now and catalogs. And how do you integrate?

Tracy Daugherty

>> Yeah. Well, there's some of both, but DataZone is own offer that we have. And so, there's integration with DataZone. There's continued work that we're doing on that front. So, that gives you both the data catalog and the governance pieces to go with that. A lot of companies do their homegrown. They make their own solution and they try to wire into it. And we have some big customers that do that. I think that's mainly due to they have a demand, but maybe the thing that they want to get out there isn't quite there yet. And so, we're really no different than most other BI tools in that we're dependent upon a separate catalog and governance, but we wire into them and go from there.

Tracy Daugherty

>> And so, DataZone, that would be the metadata catalog. Is that the operational and the business meta?

Tracy Daugherty

>> It's more the business one. We have the glue-

>> Glue-...

Tracy Daugherty

>> . It's more the-

Tracy Daugherty

>> Technical-...

Tracy Daugherty

>> metatechnical one. And so, like in DataZone.

Tracy Daugherty

>> Do you need-

Tracy Daugherty

>> And they're closely together, so-

Tracy Daugherty

>> But you need access to both, right?

Tracy Daugherty

>> You do, yeah.

>> Yeah.

Tracy Daugherty

>> You do.

>> You'll connect-

Tracy Daugherty

>> And we connect to both. Yeah.

>> Okay. So, I don't have to worry about that. You're going to do that connection?

Tracy Daugherty

>> Yeah. There will be a data engineer on the background that'll worry about that.

>> Yeah, so- Oh. Yeah.

Tracy Daugherty

>> But yeah. For sure, because that data is... Again, the trick is if it's in a data warehouse, or a database, or a data lake, I mean it's not... How would I say it? It's not willy-nilly data, it's serious data, right? Somebody is managing this, making sure it's accurate. To your point, there's pipelines coming in and monitoring and all that. Our job is to represent it in a way that it can be consumed by every user in the company, if they want to.

>> But the example you gave before was an interesting one. Is a quarter a time period? Is it money? What is it? And so, come back to that. Again, we like the word harmonization.

Tracy Daugherty

>> Yeah.

>> Are you doing that harmonization? Is AI doing that? Is there some-

Tracy Daugherty

>> It's a point in time answer. In the beginning, we were doing a lot of it. And we have this concept of a topic that you can create. And so, you can do things, like synonyms, and names, and all that. So, you could define in there, "Hey, quarter for this data means X, and it means something else for Y." As the ALUMs get better and science gets better, they have more knowledge and insight to go down that path. And so, it evolves. The thing I will tell you though that we found, the most obvious terms tend to be the challenging ones, because they're obvious in different industries, but they're not the same. So, we found it from sporting analytics, the terms they use, some common terms, similar to financial but means completely different things. You could take manufacturing and medical care and you'd find these similar things. So, it's really interesting and challenging.

>> It is a big challenge because, as you reach into those systems that are either on-prem or in the cloud, those apps, business apps, et cetera, supply chain, ERP, those definitions that are-

Tracy Daugherty

>> Yeah. And the nice thing though is they're obvious to that industry.

>> Yeah.

Tracy Daugherty

>> Though it is something that somebody can massage it pretty quickly and go through that, it's sometimes not obvious that they need to because it such an obvious term. But then, you talk to folks and go, "Hey, how do you define a quarter in your database?" Like, "We don't define it. That's just obvious." But then, when you say that, they're like, "Oh, this is a trivial thing to do and go through."

Tracy Daugherty

>> We're out of time, but re:Invent's coming up. Any glimpse you can give us?

Tracy Daugherty

>> Well, I can guarantee you it'll be exciting. How about this? It taps into a little bit what you guys talked about. And I'll just speak to it at a high level. It is about how do we take advantage of more data, more different types of data in this one thing. And then two, as you guys would know, BI is very much about doing analysis on data. Well, how do we take that to the next level? And there's efforts in both of those fronts that I think will be pretty exciting.

Tracy Daugherty

>> Raising the bar at re:Invent. I'm excited. Tracy, thanks so much for coming on the program.

Tracy Daugherty

>> Awesome. I enjoyed it. Thank you.

Tracy Daugherty

>> All right, that wraps up our On The Ground Day here at AWS in Seattle. Rob Strechay and Dave Vellante. Go to siliconangle.com for all the news. The cube.net is where you'll find all these videos on demand and live. And don't forget to go to thecuberesearch.com. That's where all our deep research is. TheCUBE AI, RAG. Go ask the question, and you'll get an answer. And you'll get some short clips. Thanks for watching, everybody. We'll see you next time.