Summit Live: Data Strategy - Democratizing Consumption and Growing ROI

Summary

• Data strategy importance: Data democratization and strategy help organizations prioritize accuracy, trustworthiness, and accessibility to unlock data value and grow ROI.
• Global CDO insights: Robin Sutara, Field CDO at Databricks, shares insights from her global experiences on building data-driven organizations and consumer enablement.
• Data literacy: Democratizing data increases data literacy across the organization, enabling faster insights and better business decision-making at scale.

Summit Live: Data Strategy - Democratizing Consumption and Growing ROI

With organizations collecting more and more data every day, the accuracy, trustworthiness, and accessibility of data must be prioritized if businesses want to unlock its value and grow ROI. And, that's where data democratization and data strategy can help. Hear from Robin Sutara, Field CDO, on top insights from her global experiences.

FAQs

What is data democratization?

Data democratization means making data accessible and understandable to all employees, enabling them to extract insights and make data-driven decisions without specialized technical skills.

How does data strategy unlock ROI?

A data strategy ensures data is accurate, trustworthy, and accessible, creating the foundation for successful data democratization initiatives that drive measurable business value.

Who is Robin Sutara?

Robin Sutara is a Field CDO (Chief Data Officer) at Databricks with global experience helping organizations build data-driven cultures and unlock value from their data assets.

What's the connection between data literacy and ROI?

Increased data literacy across an organization enables faster insights and better decision-making, directly improving operational efficiency and bottom-line business ROI.

Full transcript

Hello and welcome back to uh again back with Data and AI Summit 2025. This is day two and this afternoon we have a huge amount of guests lined up for you uh to be able to share the experience with you back at home. And to start off with, I would like to introduce my fellow teammate on the developer relations team. , hello. Hello. >> Hi. How are you? >> Good. Busy, thank you. Uh so, I know you've been incredibly busy in the run-up to data and AI summit because you have been building the chatbot for the app named Brickbot, correct? >> Yes.
Uh, and you also did it last year as well. So, how was that experience? What were the kind of things that you wanted to build into it? What was it that you wanted people to be able to do with the bot? And how was the experience of building it? Frankly, >> uh, building it was very stressful. Um, but the experience was really good. >> Excellent. Good. >> Um, last year what we did was uh, we basically hand rolled the whole thing. >> So, it was all manually done with uh, with Python, bunch of function calls, wrapping all the LLMs ourselves.
Um we did use Emlflow to do tracing. >> Um but essentially it was kind of all off-the-shelf tooling. >> Okay. >> This year uh we worked really hard to integrate kind of the new agentic approach. >> So kind of a slew of new models that came out the tools. >> Uh so instead of wrapping our functions and making all the calls ourselves, we kind of let the agent loose or the LLM loose, patch the tools that do uh the retrieval.
Um and just kind of magic this time around, honestly. Okay. So, you talk about kind of the agentic approach versus the uh previous approach. For people at home that maybe haven't built their own kind of chatbot before, >> what does that mean in terms of the differences? Is it like less manual in terms of like step one, extract the intent, step two, do the thing? Yeah. Like what were the steps that you don't have to do anymore? >> So, the steps that we had to do before were it's exactly what you said.
I would have to essentially classify the intent when it came in. So, hey, what is this person asking about sessions or the expo hall or something like this? Um, then once I have the intent, I would manually go and say this intent matches up with such and such query. Go get this, inject this into the context. So, somebody comes in, asks about a session, I have to kind of figure out that it's about sessions, which could be a dubious thing to do because people use all kinds of different words for those things.
Um, >> and also, I mean, we've got a very international audience here as well. Like, not everyone is a native native English speaker. And so the way that they ask the bot again it it vary wildly because we've got like 20,000 people here. So >> it can get confusing. >> Yes. So once you have the intent you go you retrieve some contacts. So say go get the sessions and inject them into the contacts based on whatever the query was.
Um and in in this years what you we basically didn't have to do any of that. >> I let the LM handle just about everything. So the LM knows that it has a tool to go get sessions. It has a tool to go get the frequently asked questions. it as a tool to go get the expo hall. Um, and so if the user says they want something, the agent actually has a thought process. It goes through its like reasoning stages which aren't revealed to the user.
Um, and in those reasoning stages, it figures out, oh, this person's asking about sessions, therefore call our session retrieval tool. >> Interesting. And you talk about tools. I'm assuming this is this is the very similar to the tools that we've got in Unity catalog. Yes. >> Uh, and so maybe some examples of that would be >> Oh, I don't know. uh like go to the schedule and look up a thing like and is that the kind of thing that you would be using a tool for? >> Yes, it's exactly it.
And not only that, but actually it's powered by Unity catalog this year. So the tool is actually registered in Unity catalog and then the tool also leverages our connections API. So literally data bricks is basically doing everything behind the scenes. >> Okay, wonderful. You also mentioned ML flow as well. So what was the difference from last year compared to this year in terms of the ML flow experience? >> Candidly, last year was a nightmare. mostly not because of ML flow but because uh kind of the space is moving really fast and it wasn't clear how to adopt kind of the old way and the old organization of ML flow the way that the experiments and models and all the things >> because it was very traditional ML before wasn't it rather than a IML >> right so the iteration process around traces just like it wasn't clear to me our team how to do it um so it was a mishmash we made a lot of mistakes and you know it kind of worked we managed to collect traces we managed to improve the bot but it was uh it It was really difficult.
Um, this year we used the new MLflow 3 release and uh, I mean a breath of fresh air. Everything's centralized. All our traces are in one place from whether it's like the evaluations that we run from the live production bot itself from several test environments. Everything's in one place but also can be separated and organized throughout the entire life cycle. We can manage it. It's really great. >> And so my understanding of tracing is it kind of helps you step through like what is the LLM doing here?
Like what tools does it like? So if something goes wrong, you can start unpicking it a lot easier. Yes, my understanding. >> Yeah, absolutely. So the tracing call essentially um starts at the beginning of the request and ends at the end of the request and also captures everything in the middle. So all the timing, all the external calls, the tool calls, the inputs and outputs that go into each tool call. It's just full visibility. >> Okay then.
And um Okay then. So Nick, thank you so much for joining us today. And I really hope that if you are watching this from the event, you are able to use uh the chatbot and give it a go and know that it's hard work behind it. So thank you for joining us. Up next, uh we have a lovely video for you to watch. >> So all over the world, wherever I go, one of the biggest challenges that I see for successful data and AI implementations is the access and governance of secure good quality data.
So companies are siloed. It's hard sometimes to get access to the right data both within your company and across companies. And it's also a challenge to even find and search through the through the hundreds of thousands of columns and table names. If you have something like sales_613 and sales_ne, it's hard to understand exactly where the right data, the right versions are. So where data bricks helps come in are three different areas.
Number one, you have delta sharing which is an open-source approach where you can share uh not just your data but also your AI models, your notebooks, your dashboards. You have data bricks clean rooms which is a safe hosting environment really across companies without the need for costly duplicate information and data being sent back and forth. And the third you have the data bricks marketplace which is really for all of your data and AI and analytical needs. >> And we are back at Summit Live.
Uh thanks for watching and I'm here with and. Uh thank you for coming on. , for those that don't know you, why don't you talk about yourself and the role that you have here? >> I probably have the best role at Data Bricks maybe outside of this role that the two of you get to do every year. >> We'll fight you for that title for sure. Go ahead. So uh. I'm the field co here at data bricks. So essentially what that means is I get to travel all over the world.
I get to work in the Americas with. I get to work in London and AMIA with uh to help our customers think about how to be successful with the data bricks platform. We have great tech and I'm sure you're all excited about some of the technical announcements we're making this week. But there's also more to it that we know uh and we at data bricks know that you need to be able to think about those other aspects to be successful with the platform.
And that includes things like people, uh, process enablement, change management, operating models, organizational design, uh, sort of all of those things that you struggle with connecting the technology to the business to make sure that you're really driving business value. And so I have the amazingly phenomenal job that I get to help customers do that all over the world. >> Wonderful. So you spend a lot of time talking to customers.
What are the kind of trends that you start to see and maybe some of the unusual trends that maybe we might not have heard about until now >> Yeah. Now I have to think that I'm novel and novice. Yes. Uh >> so I think uh there there are some definite trends we see with those organizations that are doing this really really well. Um I got to do a whole talk on this yesterday. So I was really excited to sort of share some of these best practices.
But I think one of the big ones is really thinking about what is the mission of the organization. I I think organizations are feeling change fatigue. I mean, even if you think about the pace of innovation, the number of announcements being made, people are having to change their roles and their functions all the time because tech is now enabling them more than ever before. Uh, and so if you think about that fatigue that your organization is really starting to feel, how do you make sure they're connected to the mission of the organization so they understand why they're doing it?
I think data bricks has a phenomenal mission, right? If we think about helping organizations solve the world's toughest data problems and I don't think there's anybody at data bricks that you would meet that's not passionate about helping people solve tough tough problems across >> and to your point about kind of change fatigue if you don't have that passion. I noticed recently we changed our a lot of our onboarding documentation to talk about everboarding. >> There is there is no more like learn for two months and now you're set to do the job forever.
It is a continual process. >> You're set for two months. Yeah. >> Yeah. Skills are obsolete very quickly. Um, so yeah, I could understand how companies who have been very, >> you know, consistent for maybe decades until now are finding that, oh, but we just did a data enablement transformation piece and now you're telling me I've got to do more. >> Exactly. And the interesting part I think to that, is like you said, how do you ever board, but how do you do it in a way that's very targeted on personas?
Right. So if I look at those organizations that are doing this really well, they're not training everybody anymore on data 101 or AI 101. They're really thinking about what are the personas that exist across your organization. How are you enabling your data scientists? How are you enabling your data engineers? How are you enabling your business users in finance, in HR, in your sales organization, in your services organizations? uh and all of those need very specific data and AI training because how they're using the technology, how they're using the tools, how they're using the platform is very different, right?
If you were to coach a finance person on Python so that they could write open up a notebook and do data bricks, probably not going to keep them for very long, right? And so how do we make sure that we're really really thinking about those personas? And I think those organizations that have cracked that piece are just doing phenomenally well. Yeah. And >> sorry, I was going to say I think also the training piece has changed as well.
Like training once upon a time used to be an incredibly formal thing that someone would do. It used to be like, okay, we're all going to go on a training course once a year for a week and that's that's it. Whereas now it's about I don't want to say empowering because I hate that word, but it's about encouraging people to train on the job and treating that as like, oh, I don't know this thing. let me spend time and find out which is very different from I want to say audience you know um you know attitudes maybe kind of 5 10 years ago where it's like well I don't know that thing and it's your job to teach me and spoon feed me this and maybe that's a controversial take but how much of that do you see?
No, absolutely. I mean, you talked about like what's not going well. I think those organizations that haven't created that almost psychological safety because not everything's going to be successful if you're training on the job. And so, people have to feel like, okay, I'm going to try to do my task this way now. And if it doesn't work, I have the support of the organization to figure it out, right? To be given the time and the space to be able to figure it out.
Because those organizations that strive for perfection before they do anything, >> you will never accomplish anything, right? behind especially in this pace of those organizations that allow their staff whatever role they are to have that safety and the capability to do that on the job training and then how do we build that curiosity I mean >> yeah I've heard the phrase like nonlinear training um as a way >> so we have perplexity licenses at data bricks that thing has saved me more than I would like to admit because I can sit there with a framework that I don't really know and I could be like Hey, I don't really know Rust very well.
Can you explain this to me in very, very basic steps? Um, and it's great. It's a coach. It's something that I'd never get on Stack Overflow. You know, if I wanted formal training, I'd have to wait. I'd pay a few thousand dollars for some training that may or may not have been relevant. Whereas now, I can like instantly in the loop figure something out. And I don't feeling bad at the end of the day because like I've had an AI be really nice to me all day and be like, "Oh, it's okay that you made that confusion." >> Yeah.
That's one thing we love with and you as well. Like there's like brick bites internally where it's like you or you or myself and many others just succinctly talking about you know the the core of everything and then there's longer training and then there's like academy training and then there's certification and then there's like the organizational personas and then different from the organizational standpoint what your maturity level is like a lot of customers are still like you know I love Gen AI I love agentic systems but we don't even have the data foundation yet.
And then there are others that are great foundation and they're ready to accelerate, put their foot on the pedal. >> But don't think that you have to have all those foundations in place to start enabling your people to give them I mean just like. I think I use Perplexity every day and it's not necessarily on the not as tech. way more technical than I am, but right I mean you can use it for things like help me understand this market space, help me uh understand this customer that I'm getting ready to have a meeting with. like really sort of pulling those things together to make you way more efficient.
And all of that comes with having the freedom for us internally to be able to sort of play with those things and be willing to say, "I don't know. I don't understand." It's sort of like having your little mentor, right, where you're not having to admit that you don't know something and that you can give be given the time to figure it out. So I think those organizations that are putting that into the hands of all of their employees, not just thinking about it from your data teams or your IT teams, but thinking across the entire organization, from your CEO to the right to the store manager to the person on the manufacturing floor.
They all have a value ad that they can get from these sort of capabilities. And you don't necessarily have to have every aspect of every piece of data migrated into data bricks today to be able to do that. start with what you have and there's so much value that you could attribute out of that >> paralysis by analysis as we say >> in sport. >> Exactly. >> So that's uh that's very much kind of the education piece of it.
Um in terms of kind of the AI implementation story or journey that organizations are going on. We've talked a little bit about what to do. Can I ask you a slightly controversial question? Are there any things that you've seen out there that you're like oof that is a complete misstep. I would not recommend this to anyone. uh against kind of anti-recommendations for people in their strategy. >> Yeah. I think um tell telling yourself that you're going to chase that I a uh AI project, right?
Just because it's a shiny toy and you want to play with it and be able to deliver. I think 18 months ago that was all the conversation we were having. Everybody was like, "Let me talk to my data. I' give me a chatbot inside the company." Every CEO I was talking to said, "Just let me talk to my data. That's all I want to be able to do." So lots of organizations created 57 different chat bots across their organization and now they're realizing what's the real value that we're trying to attribute out of that >> and sorry from a practitioner perspective though I was speaking to people it's like oh I want to do this AI project I'll be like okay but like what's the business value it's like oh the board has told us we need to do an AI project and I'm like okay but this like this isn't going to work long term and they're like I don't care I just need to do the thing because the board told me to do the thing >> and I think now they're starting to go back to say what are those metrics of success what's the CFO is saying what's that return on that huge investment I gave you all a year ago to do this AI but the other aspect to that I think is failure is to be expected in AI I think everybody thinks every AI project is going to be hugely successful and you've put all this influenced revenue around and all of this productivity gain that you think that you're going to achieve those organizations that are doing really well at this and those that are not doing really well at this have figured out what are those intermediate metrics that we're going to set in place so we know that we can deinvest, right?
That we can pivot resources and not invest anymore. And I would I would say for those that um you know are really struggling with budgets or articulating that business value, go back and look and say, have you created those milestone metrics to say it's a lesson that we learned on what worked or didn't work within our organization? Not a failure. It's a lesson that you've learned and now how are you going to take that forward into the next AI project? >> Yeah.
It's also about kind of when to pull the plug quickly. So we just had kind of on talking about Brickbot and he was saying how last year there were some things that were so difficult to do. The technology wasn't quite ready at that perspective and I'm sure we're going to have a whole host of innovations and I don't think AI can answer everything. Absolutely not. And so you know maybe maybe something is a bit too ambitious in terms of what it is that you want it to do.
Um you know it's okay to kind of pull back and be like oh no we've got to integrate with all these source systems and that doesn't work and like actually we thought the users wanted this thing and they really don't. what they actually want is something different. So like being able to kind of like very quickly say, "Okay, yeah, thank you for the learning, but we're done. Goodbye." Um, really important. >> And following up on that, like it's the whole maturity level.
And we all travel around the world sometimes. We meet up in London. You gave this amazing 5,000 person presentation, but we were recently in Nashville and you were talking about this uh account team maturity assessment. Want to make sure you have the time to talk about that. >> Yes. Yes. So anybody who's interested, we have done some pretty significant work. I talked about how do you identify those different personas across your organization so that you can make sure you're building the right enablement, the right capabilities, the right governance structures, etc.
Uh our teams inside of data bricks have done some phenomenal work on what we call uh the maturity assessment that can actually help you start to look across your organization. You can look across the entire organization. You can look at it by domain. uh because if you think about it, you're just like an AI project, you're not going to solve all your c cultural issues across your entire organization in a day. So, how do you start to prioritize your limited resources to be able to go and deliver that?
And so, we think with this maturity assessment, you can really get a good point of view of who understands our organizational data strategy, who understands the operating model that we have in place, who's uh aligned with the governance structures, etc. , and help you sort of create that heat map that allows you to then invest your own resources. So if you're interested, follow up with your account team. We would love to work with you on that.
Uh it's completely free. Uh I love when I say free because people just love that. And so we really want to make sure that we are supporting organizations on their data and AI maturity and that journey that they're looking to accomplish. >> And talk me through the process roughly kind of like how long does it take? Like is it a get everyone's head together in an afternoon type kind of thing? >> Well, so it just starts with a 30 question survey that takes about 15 minutes and we can deliver it to as many people across the organization as you want.
Okay. We've had some pretty significant sized organizations where we've started just in one department or one business domain just to get a point of view from them and they got so much value out of it that we're now expanding into other businesses and other parts of the business so that the CEO and the chief data officer etc can get a point of view of their entire organization. So relatively low-level list on lift on the individual uh and then all the heavy lift is on data bricks to pull it all together to give you a point of view and a comparison against some of the data and AI market leaders that we see. >> Well, great.
Well, we're getting the signed. Thank you so much. >> Thank you. >> Always a pleasure to have you come on, but we're going to now toss a video. the sessions. Uh the keynote is amazing, always so informative. There's so many things that I've already seen and heard that are coming out. I think there's some new announcements that are going to come out around the free edition that I think everybody's going to be able to get access to.
That's going to be really exciting so that people that are in school, people that are outside of data bricks today can get access and using it. So the best advice I can give you is start using it. start exploring the features like you know do not hesitate to you know play around with the with the C uh with the SQL editor over there or like or DT or like AIBI like there are so many things out there just start playing with it what fits with your team and with your organization right really uh you'll know until like you you really test with it >> yeah if you're starting first uh there are unlimited possibilities but first thing is to know take your first initial step so don't hesitate explore all the resources you have from data bricks and partners likes us and then start your journey.
Everybody's there to help you. >> Hi, I am here with Ford Direct. , it's so great to see you again. >> Yes, it's good to see you too. Um what's been 6 months since you came to our uh annual meeting, spoke to the employees. uh got everybody fired up about where we're headed with um our our data position as a company, what we're doing with um analytics, what we're doing with customer journey measurement and uh innovations in AI to sort of action that data for our constituents and our dealers. >> Great.
And it was incredible to come out and meet the broader team, see what you're doing. And you spoke at summit last year, so thank you for doing that. And you spoke about all the great things that direct is doing with uh data bricks with conversational interfaces. So one of the great things was the uh dealer co-pilot that let's start with that. What what's that about? >> Well well first of all um Ford Direct is a joint venture between Ford Motor Company and all of the uh dealers and retailers in the US and in Canada.
Um we are like the primary we we essentially sit in between the OEM and the whole dealer network and we are the primary uh digital marketing advertising and dealer systems provider and trusted advisor um to help ensure that customer or that sure that dealers can um attract, engage and convert customers as they shop you know essentially across the brand experience what we call tier one the region advertising communication that happens with dealers and uh customers and when they're either shopping or in service at tier 2 they call it.
And then at the dealer level where customers are interacting with the dealerships either via the website or they're responding to advertising, they're responding to oneonone communications and things like that or they come into the dealership and the dealer starts to um work that sale um and essentially transact. So, we're in a position at Ford Direct where we have access to all those channels through partnerships, capture all the data from the customer as they do all those different things either as a shopper or as an owner.
And we basically have consolidated that and stored it all into our lakehouse that's provided by data bricks. And from there, we've built a essentially a customer journey platform um which is essentially all those customer signals stitched together in a time series fashion. We built a metadata layer and a data science machine learning layer on top of that to get that strong profile information about the customer because you don't want to know every single thing they're doing at all times.
But all that data is very rich for data science. And now our latest phase um in in evolving that is um adding basically an AI environment on top of that metadata layer so that we can um allow different ideas and solutions that we can take to market talk to that data so that they uh we either launch those as separate products or we enhance our partners' products with that information to create a more personalized experience for the customer as the dealers interacting with them. >> Well, it's incredible.
I I I loved having gone out there but talking to the entire organization. It's very innovative. That's what I love from the top down, from the bottom up, you know, great practitioners. Uh but then also the leadership understanding to get the best information in a way that's actionable and there were also some dealer leaders that were there and they just love being able to get really actionable insights from what your team is doing.
Yeah, I mean we're still in sort of a pilot mode with the what we call the dealer co-pilot right now. It's probably going to be renamed and branded as something different. Um the dealers that are using it are seeing its value. It's essentially an AI forbi tool. The dealers interact with anywhere from 9 to 15 dashboards uh that they have available in the dealership and they interact with about you know 10 of those a day.
Bogs them down uh to get information. So, we're trying to bring all that together in a mobile uh voice enabled assistant where they can answer quick questions either right there with the customer or on their feet or they can sit back and analyze what's going on operationally in a dealership. Um, let's say in terms of optimizing inventory mix and things like that. >> Well, awesome. Well, thank you for sharing. This is incredible.
One of my favorite uh customers that I talk about all over the world where I travel. So, thank you for coming in person and doing this. >> Well, you're welcome. It's good to see you again and it's been great to be here. You too.

Learn more about the Databricks Data and AI platform.