Consumer Industries Transformation: Databricks AI Strategy for Retail, CPG, and Travel

Summary

• Leaders from Danone, Albertsons, Virgin Atlantic, and Unilever share how they are using the Databricks Data and AI platform to move beyond AI-driven efficiency and reimagine core business models, against a backdrop where falling token costs are paradoxically driving higher enterprise AI spending.
• Albertsons tackles the locally optimal, globally blind problem across 2,200 stores by using graph neural networks, causal modeling, and explainable AI to optimize the four Ps of merchandising in a way that is interpretable in merchant language.
• Danone's 90,000-person transformation links data strategy directly to corporate mission using data products and Genie's Talk to My Data capability, while the forum identifies 2026 as the year CFOs demand proof of AI business value realization.

Consumer Industries Transformation: Databricks AI Strategy for Retail, CPG, and Travel

As AI capabilities accelerate and token costs plummet, enterprise spending paradoxically increases. This forum explores how leading consumer companies use Databricks to move beyond AI-driven efficiency and reimagine their business models entirely. Hear from Danone, Albertsons, Virgin Atlantic, and Unilever on data foundations, governance, team building, and measurable ROI at enterprise scale.
Rob Saker (Databricks) introduces Jevons Paradox and 2026's AI audit, where CFOs demand business value realization. Dee Fitzgerald (Danone) shares how 90,000-person transformation links data strategy to corporate mission. Karthik Iyer (Albertsons) reveals how graph neural networks and explainable AI solve the "locally optimal, globally blind" problem in unified merchandising across 2,200 stores. Louise White (Unilever) and Rich Masters (Virgin Atlantic) discuss governance, team evolution, ROI measurement, and the future of agentic commerce.
🤝

Chapters

• 00:00 Welcome to the Consumer Industries Forum
• 02:33 Rob Saker: Enterprise AI Vision and Market Context
• 06:31 Two Types of AI: Efficiency vs Business Reimagining
• 14:39 2026: The Year of AI Audit and Enterprise Requirements
• 19:34 Dee Fitzgerald: Danone's Data-Driven Transformation
• 23:06 Danone Renew 2.0: Strategic Transformation and Growth
• 28:06 Data Products and Talk to My Data with Genie
• 36:18 Karthik Iyer: Albertsons' Merchandising Transformation
• 39:43 The Four Ps Problem: Locally Optimal, Globally Blind
• 41:20 Sense: Making Sense of 200 Billion Rows
• 42:24 Custom Deep Learning Models vs LLM Limitations
• 44:16 Graph Neural Networks and Causal Relationships
• 47:49 Explainability: The Promotional Molecule Concept
• 51:03 Explainable AI in Merchant Language
• 52:07 Databricks Full Stack: Data, Models, Genie, and OpenAI
• 54:33 Structure Over Model: Why Architecture Matters
• 57:31 Future of Work: Panel Discussion Begins
• 01:00:09 Data as Growth Engine and Foundation Requirements
• 01:07:35 Building Curious Teams for the AI Future
• 01:15:39 Balancing Innovation, Governance, and Token Growth
• 01:18:50 Measuring ROI at Enterprise Scale
• 01:23:38 The Future: Agentic Commerce and Human Empowerment

FAQs

What is Jevons Paradox and how does it apply to enterprise AI spending?

Jevons Paradox describes the counterintuitive phenomenon where improvements in resource efficiency lead to greater total consumption rather than less. This video applies the concept to AI: as token costs fall dramatically, enterprises paradoxically increase overall AI spending, meaning CFOs in 2026 are demanding proof that this spending translates into measurable business value.

How does Albertsons use AI to solve its merchandising optimization challenge?

Albertsons faces a locally optimal, globally blind problem where optimizing any single dimension of merchandising—such as pricing, promotion, placement, or product—can create suboptimal outcomes across its full portfolio. This video explains that Albertsons uses graph neural networks and causal models trained on 200 billion rows of data to capture these interdependencies, wrapping results in explainable AI output described in merchant language as promotional molecules.

How is Danone connecting its data strategy to corporate mission?

Danone's Renew 2.0 transformation links its data and AI investments directly to the company's strategic mission across 90,000 employees. This video describes how Danone uses data products and Genie's Talk to My Data capability to put data-driven decision-making in the hands of business users across the organization, not just data teams.

What does the future of work panel discuss in this video?

The panel featuring leaders from Unilever and Virgin Atlantic explores how enterprises are building curious teams for an AI-driven future, balancing innovation with governance, measuring ROI at enterprise scale, and preparing for agentic commerce where AI systems take autonomous commercial actions. This video frames the transition from AI as an efficiency tool to AI as a mechanism for reimagining how consumer businesses operate entirely.

Full transcript

[00:08] Hi everyone. I say I know a few people are still trickling in, but I just really want to thank you for coming out to our consumer industries forum. This is um no it or not, my first data and AI summit. Is it anyone else's first data and AI summit? Okay, I'm not alone. This is great. Um I want to hear about your
[00:24] feedback, but um first let me introduce myself before I get really excited about this forum. My name is Sarah Duffy. I'm on the industry marketing team focused on consumer industries, retail, consumer goods, travel and hospitality. And um I started
[00:39] at Databricks about 6 months ago, and one of my favorite things that I have quickly realized about this community is that it is a community. I can't really tell you the impact that it has meant for me to come in and be able to connect with consumers, with customers, with
[00:55] fellow people in the data and AI space, and just learn from one another. And that's really what we're here to do. So, we really have an amazing agenda set for you. But first, before I can even move forward, I really need to thank our sponsors, Deloitte, for this session. We could not be here without them. They are
[01:11] such valuable partners to us and partners to a lot of our customers in the room as well. So, thank you Deloitte for helping us to put this on and to make it such a great session. Um speaking of community, I do also just want to plug a little personal note of
[01:27] we started a great LinkedIn channel for our consumer industries. Please take a chance to follow it. You might see your photo pop up on there because we've been taking pictures during forum. But this is really where you're going to hear the first things that are coming out of consumer industries for Databricks. Um
[01:44] we are trying to keep this up-to-date with all the new product launches, what this means for you, and how other folks in the business are taking advantage of Databricks' new products to really innovate in their businesses. So, we have a great agenda set out for you. We
[02:01] have some speakers that I have gotten to know personally, and I can't wait for you to hear from them. So, we'll have two keynotes, one from Dee, she's from Danone, and then Karthik as well. You guys might have heard from them either already this week or in prior summits, but they have great talks coming up for
[02:17] you. And then we have a really exciting panel at the end where we will talk about the future of work with AI and what that really means. I think a lot of times we've been talking about how AI is transforming our businesses, but what is the future of those businesses actually
[02:33] look like when we put AI to work. So, I'm really excited to have that conversation. But, I can't go forward until I introduce Rob Saker. So, he is our lead go to market um leader for consumer industries, and he is going to come up and kind of give you the lay of
[02:48] the land of what we're doing at Databricks, what it means for you, and then how we can move forward. So, Rob, up for you. Sarah's awesome. She's been here 6 months. It feels like she's been here
[03:03] for years. Uh well, in Databricks terms, that's probably right. Um thank you for all attending. Hopefully, the conference How's the conference been going so far? Like I've been here 7 years. More announcements this year, I think, than any other year. Is it pretty exciting so far? Yay or nay? Okay, good.
[03:20] Good. Um I want to put some context into why we're doing things. Cuz I think you saw the flood of announcements, and it's hard to make sense of kind of what's the underlying mission and motivation for what we're doing there.
[03:35] We saw over the past year this massive increase in the spend in AI. Uh how many people had their leaders say token maxing? We should be doing token maxing. Everybody invest in AI. Uh everybody, you know, don't worry about the spend. Uh
[03:51] that's quickly come to coming to a a heading point where I think that we're seeing a lot of investment, a lot of expense in the area, but the value part has trailed. And I think there's a couple reasons as we think through
[04:07] this. Here. First of all, AI capability is accelerating. The foundational models are reaching parity in terms of what the capabilities are. We know Open AI, you know, 8 months ago was the leading model. Opus 48 is
[04:22] the leading model now. Gemini, all these other models. The Chinese open source models are starting to reach parity. The Llama models are reaching parity. So, we're seeing this real advancement in the capability of AI. Um but despite the decreasing token
[04:38] costs that we've seen a 90% reduction in the cost per token over the past year, we're actually seeing the actual expense being spent on AI increase. And the reason for this, there's this theory called Jevons paradox. It's from 1865 and it's still relevant, but it
[04:55] explains every major innovation that we've had over the past 200 years. And what it says makes perfect sense. Says that any major technological innovation that reduces the friction will not reduce labor because it makes it so easy to solve problems that unmet latent demand will
[05:12] actually over uh surmount the benefits that we get from this efficiency. So, what we're seeing is it's so easy to solve problems with AI that there's pent-up demand in the business that they want to go and take on. This is why we saw software developers where we had predicted the
[05:27] end of the software developer. Hiring of software developers was up 14% in Q1. Analysts, oh, we're not going to have analysts anymore. Analysts, the average analyst is working 10% extra hours now versus a year ago because AI makes it easier for them to go and answer
[05:43] these questions. And when you add this in aggregate, the net result is while per question the cost may be going down, the number of questions that we're answering goes up. The challenge is if you compare this to what's been happening in our industry, what's been happening in the past 5
[05:59] years? Real revenue is up, volume's flat. We've been maintaining the growth of our industries in every single one, whether it's travel, whether it's consumer goods, whether it's restaurants, whether it's retail, by focusing on increasing prices. And if you compare this to
[06:15] what's been happening with consumer discretionary income in every part of the world, we're at a point where we can't continue to do this. We have to find a way to get to profitable revenue growth. And I think AI actually provides us an opportunity to get there.
[06:31] Now, to that point, when we think about AI, only a third of companies are actually thinking about how do they reimagine their business? How many people here have AI use cases where the primary driver of AI is efficiency gains in the organization? Yeah, a lot, right?
[06:47] I think this is the first-order problem. So we certainly want to go in and use AI to drive efficiency. I'm not suggesting that we don't. But AI allows us to do a lot more. AI allows us to be smarter about what we do. It allows us to augment and act. We should be thinking about AI use cases
[07:03] not just in terms of driving efficiency, but reimagining things, being smarter, so I can focus on areas that drive additional revenue, drive margin improvement, to be more relevant with customers, improve customer satisfaction. Not just efficiency. So to bring this to life, the average
[07:19] store associate only spends 20% of their time meeting with customers. Think about it. If I have an 8-hour shift, only 20% of my time is being spent actually talking to customers. The rest is largely administrative activities that we're spending, checking
[07:34] inventory, back office, going through things that we're doing because there's SOP, not necessarily because there's a condition that needs to be fixed, but just check. There's a lot of inefficiency that's actually built into that store associate on a daily basis.
[07:50] If I can drive 48% of 48 minutes of improvement in their day, I can get a 5 to 14% improvement in revenue. If I can find go from 20% to 26 or 27%
[08:06] of their day being spent with customer, I can do this. So, how do we do this? Well, first I want to talk about how AI sets the stage for doing this. AI does a lot of other a lot of great things, but I think there's three things that it does. By the way, individually,
[08:21] every one of these three is disruptive and potentially bad. The first thing is AI allows you to take advantage of new types of data that you previously haven't been able to tap into in your organization. The Forrester stat, 10% of companies' organization is structured and available to data warehouses. 90% is in
[08:38] unstructured forms, images, text, video. But it's the 90% of that information that is the richest about the context that happens in your organization. Internally, it's PowerPoint, it's documents, it's emails, it's Slack messages.
[08:53] We all use transcripts now, AI transcripts, and that information from that transcript is a prime opportunity to understand what happened in the meeting and why. But externally, it could be social media, it could be video, images, sensors, and other rich information that helps us to better understand. But by
[09:10] itself, more information is noise. The second thing that AI does allows us to do is to identify patterns in information even when it hasn't seen those patterns before. An LLM is fundamentally the relate identifies the relationship between two
[09:26] entities and the proximity, the distance between two Emily and these and those similarities. So, when you think about different data sets, it's looking and saying which data elements are closest to each other that help us to better understand the context of what happens? Now, by itself
[09:42] on a limited set of data, that gives us a false sense of confidence. And then the last thing that AI allows us to do is to augment and even automate the way decisions are being made. Now, we work in a world where there may not be high trust in automation of AI,
[09:58] but I think augmentation, having AI take a richer set of information, find the patterns in that, and suggest to me opportunities that are not just about efficiency, but about better ways of doing things. This is the real opportunity for us.
[10:18] Now, the nice thing is we're seeing a massive explosion in those front-line signals. We're seeing in the retail space, shelf sensors, IoT, CCTV. We're seeing in the travel space, a wide array of partners. I was with a travel company last week
[10:33] and we're talking about Uber partnerships. We're talking about other things to better understand the journey that you take. Your journey on an airplane does not begin the moment that you get on the plane. It begins weeks before when you're considering things. It begins that day when you get in your car and you go to
[10:49] the airport. It doesn't end until you return. So, everything in between. How do we think about what's happening and using that information to better understand? And you think about what are those sources? I like frameworks. So, I would encourage you to think about
[11:05] there's data that you control. This is your enterprise data, your endogenous information. These are things that you're going to have from your internal systems as a result of running your business. This could be point of sale. This could be loyalty programs. This could be booking systems. This could be your back
[11:21] office type of systems. The operations data. You think about the front-line work. Call centers and other types of activities. And you think customer. And customers are where we start to get interesting because there's information that I have from e-commerce and loyalty and mobile
[11:37] applications, but I start to do third party with clean rooms and enrichment. But the real opportunity, I think, is with environmental data. Everything else is about what we control, and environment is about what controls us. And we all operate in a world that is primarily driven by environmental
[11:54] activities. It's driven by holidays, by weather, by key events. The things in the environmental world that we can understand to better determine where do we want to focus our resources? And this is what AI can allow us to do automatically.
[12:12] Now, I would encourage you. I'm not going to take you through that because we've got a large list of speakers. How many people have gone to the demo floor? Though, on the main it was a madhouse when I went down and looked at it. I would encourage you to go take a look. In the consumer industries forum, we built out some incredible demos
[12:29] with Dataworks app that show you how to do this, but I'll tell you one of the applications that we built in there. Imagine from a media perspective on media buying. If I just think about efficiency, what am I going to do? I'm going to look at what I can do to make it easier to automatically buy ads.
[12:45] That's going to lower my acquisition cost because I have people that I'm taking that labor away. But that doesn't grow my revenue. When you go and look at the demos there, you'll actually see that we've done an analysis and said, "What are the moments that matter? How can we use the external
[13:01] information in terms of events, holidays, activities, weather to suggest here are things that correlate with buying periods. So that media planner doesn't have to go in and do what they did last year. They go in and the AI is saying, "We've have information. We
[13:16] found a pattern that we think that matters to you. We're just going to suggest content. We're going to suggest an audience that's relevant based on what we're talking about, and to be more thoughtful and suggestive so it's more efficient, but it leads to higher ROI. We've got five demos in the demo center
[13:32] to take a look through. And the next thing is building these out is very easy once you get your foundation on Databricks. Now, what it does, I talked about it signal without reasoning. It allows us to make sense of this information,
[13:48] right? I get a wider aperture of customer information. I'm able to identify the patterns, and I'm able to focus on the augmenting and even automating those activities. And when we translate it back to that frontline operations person, I save time
[14:04] because I'm taking time away from the tasks that they would just normally do as SOP, and AI is telling them only focus on these things that matter. I find that 48 minutes. I reinvest and thus those 48 minutes and cross-sell, upsell. I reinvest in in
[14:20] other types of activities in terms of OSA on the frontline and bring forward. So, how do I take this not just for efficiency and design my applications to be more thoughtful about revenue and and margin improvement activities?
[14:39] So, I think that the net of this, 2026 is going to be the era the year of the AI audit. I think we have reached that inflection point with AI spending where the hockey stick being spent on the foundational models is not sustainable, and CFOs are going to come and say,
[14:56] "We need to have better value realization." Now, the first part of this is we need to make sure that the product design that we're doing is focused not just on efficiency, but also on things that drive revenue and margin growth like I just talked about. But the other part is how do we make sure we have that right foundation for
[15:12] building responsible enterprise grade AI? If you take only one thing away from this week, what I would tell you is that of all the announcements that we have, the Unity AI Gateway, the Unity Catalog,
[15:28] the Genie Ontology, what we're doing to help our customers to better understand the data with things like Genie Ontology and semantic auto generation, with the ability to better manage our models, to log the interactions with
[15:44] those foundational models with Unity AI Gateway. You can't optimize what you first don't measure. And they have the financial controls and visibility within Unity Catalog. Allowing you to take that information, to better manage your information, to
[15:59] better audit and manage your foundational models, that is the most important set of announcements that we've done this this year and that we're announcing at Data and AI Summit. And when you put this in place, it puts you in a strong position to rapidly build and prototype to prototype and build these applications on Databricks.
[16:19] I think to contrast that, if you think about vibe coding and why we need to make this shift. If I took 10 of you and we started building applications just with an end foundational model, all of us would have different context. We would have different semantic
[16:35] definitions. We wouldn't be able to scale this out to an enterprise. There would be no policy enforcement. We'd have no financial control. It's ultimately not a way, and this is what we're seeing with our customers, it's not a way that we can actually sustain and deploy these enterprise applications.
[16:50] What's required to build an enterprise AI application? I think there's five levels. I think first, you need to have that solid foundation. If your information is not stored in an open source format, Delta Lake, Iceberg, lake base,
[17:07] put it in an open source format. If you have it in a proprietary format, you're going to be paying egress fees, you're going to have a lot of latency that's going to make it very difficult to deploy AI applications. I think that governance layer next is critical. Governance is what determines whether an
[17:22] AI application is ready for enterprise versus individual desktop types of applications. We have the ability to build these AI applications and the whole workbench that we have within there. We have the fast execution engines.
[17:39] And this is the ability to use both foundational models from commercial models as well as open-source models. And lastly, what that end user experience is. We want to make sure that you have the ability to whether you're using Teams or Slack or any other proprietary type of interface with an
[17:56] API interface to your models or to build Databricks apps or to have Genie as a mechanism build it centrally and then deploy it where it makes sense. You've seen the announcements that we had. I mentioned Unity AI Gateway and Unity Catalog of all the announcements.
[18:12] If you take one thing back and you talk to your teams this week, this is the most critical announcement that we've made because this is what's going to set the foundation for you to be able to build responsible enterprise-ready AI.
[18:28] And this is what it looks like when you translate those five areas back to the platform. And this is a an eye chart, I realize. So as this is up here, I'm going to leave you with one last comment. We've done a lot of work with our customers to build the foundation, to build the governance there.
[18:44] And we're now in a place where we can rapidly work with your teams to build these AI applications. My offer to you is reach out to your Databricks account team, talk to us. We've done workshops in a day. We had a session on Tuesday where in the session
[18:59] in 40 minutes, we built an application. I had another meeting with a customer, we sat in the morning for 2 hours and wrote the specification and by 3:30 in the afternoon, we had a functional version of that application worked with the synthesized data. Get your foundation right. Get Unity AI Gateway right. Get Unity Catalog right.
[19:17] Talk to your account teams. Let's work together. Let's go build something. And I'm going to skip through. So, let's go build something great together. That's my session. I want to bring up on
[19:34] stage Dee Fitzgerald. Thank you for coming to the the the conference today. By the way, any of the people that come up and speak, if you have questions, we're going to hold the questions until after the session. We'll stick around up front. But, thank you again, and Dee, where are you? Here we go. Thank you.
[19:58] Hello, everybody. Thanks for having me. Um I think I'm going to repeat a bit of what uh Rob shared, a lot of the information he shared, but with different context. Before I get there, I'm going to talk a little bit about who we are. Uh so, the company is called Danone. It's a French-based company.
[20:15] Here in the US, you may pronounce it Dannon. Um we also have a phrase in in I think in Spain, it's called it's pronounced Danone. So, um we talk about data standards, here we are. Um but, who we are? We are uh a team of 90,000 plus, uh quite
[20:32] strong. We have a complex uh supply chain network, over 150 sites with uh more than 300 suppliers. Um we are about 27 plus billion uh with a like-for-like growth of 4.5 to 5% year-over-year, and we're serving over
[20:48] 1.5 billion consumers uh on an annual basis in 120 countries. Now, um what I am going to also share is maybe bring you to life about who we are, cuz you may not know much about the company. Hopefully, you know something about our brands. Um we have a mission.
[21:04] Our mission is to bring health through food to as many people as possible. That includes the consumers, but that also includes what I would call a context of consumers, patients through health care providers. Um we have a a
[21:19] number one worldwide where more than 48% of our products are in premium uh dairy and plant-based. So, some of the brands you may have heard of Activia yogurt, Oikos, etc. We are also in the waters business. Uh Evian is our uh let's say
[21:34] premium global brand. Uh much of our water is tailored to the to the country. Uh we deliver for, let's say, refreshment, but also for for for the needs. And then I'm going to focus a little bit on the middle sector, and that's where we are um very proud of. We
[21:49] have uh over 34% in specialized nutrition. This is what I was referring to about how we focus on our consumers through patients and health care providers. We have been uh focusing very much on acquiring some businesses. Most recently, we acquired a company called
[22:04] Kate Fods. They became a partner of ours. Uh just to give some context, what is Kate Farms? It helps provide alternative methods of feeding through tubes. So, providing the essential nutrition for people to survive who have uh who cannot eat We
[22:19] take certain things for granted. Um and our mission is to again to help with the bridging that gap as much as possible. If I bring it a little bit to home, you'll see some of the uh brands that I mentioned, but on top of that, we have uh So Delicious, Stoke, uh some other
[22:34] brands again back to home here in the US. Now, um in 2022, we launched a strategy called Renew. Uh and about 2 years later, we launched uh the next chapter, Renew 2.0. Uh what is the
[22:51] overall goal? Again, to bring health through food to as many people as possible, but really to reset us to become a much more growth-oriented company. Uh with the ambition. So, we reset the way, we focused on our portfolio, we re- reset where uh how we manage our
[23:06] financials and our and our cash flow. We re-phased and reset how we manage our culture. Um so, lots of change has happened since then. We continue to to to accelerate in that space. Um basically, if you look at the capability
[23:22] section, that of course involves people, then includes areas in our operations and our supply chain, uh make making efficiencies in our factories and our frontlines, etc. But, also if we call in the digital transformation, that's where we're very uh keen on. And we generate our strategy
[23:39] enabled by AI and data. Now, as you know, we have quite a lot of opportunity in AI. We have a quite a lot of opportunity in how to generate AI. We are being quite focused. We have
[23:55] about, let's say, seven to 10 flagship use cases. And those use cases we want to heavily invest in what we call the transformative area. How do we help enable our growth of our company? Especially in those areas that I talked about in the uh life science, the health
[24:10] and nutrition. Um how can we use the power of data, use the power of AI to target more of our healthcare providers, target more of our consumers who need uh different uh areas of focus? Um one of our key topics in Danone is the gut health.
[24:27] As I mentioned, yogurt is a is a premium product of ours. Uh nutrition all starts in the gut, it correlates back to the brain. There is so much science-based evidence to to prove that. We have invested in um uh research and innovation centers in Paris, in Singapore, just to focus on
[24:44] this type of science-based uh innovation uh supported by data and AI. We also then have our perform and optimize categories. Again, I I uh Rob mentioned this. This is to help us do what we do, but better. Uh unlock capacity to enable our our
[25:02] business to engage more with our consumers and engage more with the customers. We spend a lot of time mining data. We spend a lot of time pulling that those presentations together that have the joint the JBP's with our customers. How can we do what we do better? Make it much more data driven,
[25:18] make it much more AI driven, and then of course optimizing the efficiencies of our of our Danoners here. I think this is a bit of a chart and then we probably can all appreciate it as as was mentioned before foundations are important. Um there I say every day
[25:36] seems to be my birthday. I wake up and there's something new to add to the bottom of this this iceberg. We have sovereignty now to add, right? We have many other topics. So you know, cyber is always there as well. But so many different topics to talk about and to to fulfill while you know,
[25:53] what what our consumers want or what our customers want within Danone is what's on the tip of the iceberg. They want to see the insights. They want to see the action. They want to see the results. They want to see the P&L impact. Um so we have been very, let's say, deliberate
[26:08] in ensuring that as a company we are focusing on the foundations. Our foundations have been derived off of, let's say, a few guiding principles. The first one I'm not going to talk about all of them, but the first one is really about how to make sure that our people we keep people the human at the
[26:25] center of everything we do. Um they are to be the ones who decide. They are to be the ones who leverage this data, who define the data, who then leverage the insights and the actions. And how do we train these people our our consumers and our Danoners to take these insights and
[26:43] act differently. I mentioned before we're being very pragmatic about what we focus on. We get a lot of requests. I want this tool. I want to do this. I want to do that. We really need to we focus on what our priorities are, which are flagships. And as a result of that, by doing that, we
[26:59] have become value, let's say, obsessed, which is our third pillar. How do we ensure that whatever we do, we are somehow creating value, whether it is to unlock capacity, to unlock efficiencies in our manufacturing sites, with with production lines, with IoT, or with
[27:15] our consumer, where we believe we're going to change our market share. So, putting ROI at the center of everything we do. Uh this was also mentioned, responsible AI, but also responsible data. So, what is our what are our data governance principles and our responsible AI principles? And then
[27:32] finally, which is the let's say the core of what we have to do is basically just improve and create those those data foundations and those tech foundations. Master data solutions, data solutions. Um uh having data assets available for for all to to consume. So, much of what we
[27:50] are doing now is then correlating to what you see in the pyramid. So, we have institute we create a capability mindset and capability framework. So, data governance as a capability, uh responsible AI as a capability and as a service. All of these core foundations,
[28:06] our data platform very much heavily reliant on Databricks, which has been amazing. We've seen this tremendous progress uh over the last few years as as a result of that. Um but beyond that, how do we create what we were what we call I'm thrilled that we are now calling data products data
[28:23] products. Um I think Rob mentioned you need these structural data assets to be available. We have an inventory of what those top data assets are. We are bringing those in using Databricks, through data governance, etc., etc., to be available enterprise-wide, not just
[28:40] for a single use case, but for multi-use case purposes. Um again, backed by data governance, backed by strong technology. And then on the AI side, what are our frameworks, our models, our libraries to allow us to leverage AI um
[28:55] from a capability perspective, but with the uh initial ability to generate context and to leverage that from a let's say a a local need. We have these capabilities, but context is super important. Uh what someone in in in our
[29:11] sales force in our sales team in let's say China requires is very different than what the the sales team is talking about in the US. And then finally, this is the tip of the iceberg, right? Which of course is all that the consumer sees. Uh we have created our marketplace that allows our
[29:28] AI products and our AI and our data to be available. Um and then we have a program called talk to my data. The talk to my data is allowing our users, our data owners to be able to communicate directly with the data using Genie. This allows them to
[29:45] let's say bypass the the dashboard need, but to be able to communicate, ask questions, receive the context about the performance of the data, understand the quality of the data, and and leverage it as they see fit. Um we're that's just a bit of the
[30:00] surface. We plan to expand that beyond using many more insights in other areas that I mentioned earlier, supporting our uh product life cycle innovation with R&I, supporting our gut health ambitions uh with R&I as well. Um so if you can ask a question about you know,
[30:17] this product we received from our supplier, how do we see the impact on the One Biome initiative? Having that context, having that insights come back with context, with further information is is what we're leveraging for, and we're very excited about that.
[30:35] Now, my also eye chart here, uh we are creating this how as an ecosystem. Um it all starts with the use case. It all starts with what is it, what value we're trying to generate from the bit from our business for our consumers, for our healthcare providers. And then therefore, we have a data first mindset.
[30:52] What data is needed to support those use cases and create that data not for the use case, but for the the company of Danone at an enterprise level. That's then pillared by, let's say the ecosystem of our data platform, of our AI enabled context, our our standard
[31:10] operating procedures, etc. I'm not going to go into all of the details here as you can see, but we are leveraging and I could sprinkle Databricks across all of this, plus some other technologies, but we are leveraging the best of what we believe to help bring that agility, to help
[31:25] bring that scale, and to help bring that trust in our data, in our AI, and our responsible AI policies to life. We are working very closely in I will use the agile term still, working with our product teams in the business,
[31:41] our product teams in the data organization, supported by platforms and center of excellence to ensure that we are trying to deliver consistency as quick as much as possible, but through agility, through scale. This is all about, as I mentioned, people. We support this with We have a
[31:58] heavy training program with at Danone. We've launched We call it the Digital Data AI Academy. We are Our goal is to upskill all 90,000 Danoners, even those who are not directly connected to our our firewall. Um, we're doing that in several
[32:15] capacities. One, how to help them understand what types of questions to ask, what prompting to to put put into the questions, what kind of requirements, what kind of use cases, what should the strategy look like from their perspective in order to bring to life their functional
[32:31] priorities. We are also helping them on how What does it mean to be responsible With data, with AI, we have a mandatory ethics um session and training session that all has to take in order to ensure that they have access to some of our tools.
[32:49] Um and then finally, we have a big program with our executive leadership where we again, we bring them into hackathons, promptathons, marketplace activities where they learn. We're actually helping train some of them to build an app. I know that sounds a bit dangerous, but it's also good for them to appreciate what it means and how to
[33:05] easy it is to do, but also what is under the surface in order to bring it to life. And I believe I am left here with lessons learned. Um I don't think anything here that I'm saying is is is shocking or new. Uh I
[33:21] think the key here is we have deliberately made sure we are pragmatic about keeping the human in the loop. Um I go back to when GenAI came out, many felt that their roles or their jobs were going to become redundant. Um most of our our team members now no
[33:37] longer feel that way, but they do still wonder what will their job be in the next few years. So, instead of them not having a job, they will have a different type of role. So, therefore, what are the skills as mentioned are needed to help enable them? What do we want to do to ensure that they are making that final decision um and acting upon the
[33:54] models that we bring. Um as mentioned, we're really focused and pragmatic about what we do and what we invest in. Um we really are focused on transforming our organization, um but in order to do that, we need to optimize and perform better on our day-to-day
[34:10] processes. Um and then finally, I would just leave with um data readiness. We have created a framework where no matter what program, what use case, uh you must identify what data you need,
[34:26] what level you need it from, and where you get it. And it can be unstructured data, it doesn't have to be structured data. And then our job is to help bridge that and bring it to life using tools like Databricks. Databricks has been absolutely instrumental in the unstructured data
[34:42] you know, let's say exposure that topic. And we've brought that forward quite a bit with with just small steps, but moving forward quite quickly. And I guess
[34:58] I would just leave you with don't underestimate the skills needed for what to drive value, to drive adoption, and to drive ROI. As I mentioned people love to go back to what they're comfortable with. They love to go back to what is needed. So training is not
[35:14] enough. Training but change management, coaching, reverse mentoring, horizontal mentoring, all keys to to ensuring that this value is is realized in the companies that and that you work on. So as Rob mentioned, I will be happy to take any further questions for you after
[35:31] this. This was just a bit of a flavor of what we are going through. We're not where we need to be by any means, but we're starting. It's been a lot of hard work, a lot of blood, sweat, and tears. I think you can all relate to that, but extremely promising and
[35:46] extremely value valuable. And we're seeing and it's quite inspirational to show not only that we're transforming with certain types of use cases, but we're actually transforming the way every denoner operates in the company and how they perform their day-to-day jobs. So thank you very much and with
[36:01] that I give it back to Sarah. Awesome. Thank you so much, D. All right, I have the pleasure now of introducing our next
[36:18] speaker. I'm introducing Karthik from Albertsons. I think with both of these speakers being you heard a flavor from D and then now from Karthik is just what it looks like to truly transform a business, but not just the business itself, but with the employees along with it to really bring it into the
[36:33] future. So, with that, Karthik, welcome to the stage. I'm going to use the old school cheat sheet, if you don't mind. Um So, who is Albertsons? I think I have an
[36:49] international audience here, so um I'll make sure that uh you know who Albertsons is. So, Albertsons is both a brand, a supermarket, as well as a holding company. So, we are the second biggest supermarket in North America, or the third biggest mass market or mass
[37:06] merchandising warehouse retailer in North America. So, as being number two, we operate across 22 brands, so we have 23 brands. Uh some of us in Chicago might know us as Jewel. Some of us in uh Boston might know us as Shaw's. Some of us in
[37:21] Northern California might know us as Safeway. Somebody in the Midwest would know us as Albertsons. Somebody in Texas would know us as Tom Thumb and United Supermarkets. So, so on and on. So, we have 22 brands. We operate across most states in North America.
[37:37] And I lead the merchandising transformation. So, we have four big bit big bits in our company where we think AI will make a strategic difference for the company. I lead one of them. Uh this is my 18th month into the transformation. I was employee number one in the big bets program. So, now I'm
[37:54] a small team. But still, after 18 months, we've been able to transform something that I don't think there has been a playbook on retail until now from what I have learned from the other retailers. I'm also speaking to some of us over here. So, what do I mean by that?
[38:10] So, if you if you know of these four Ps, the four Ps in marketing are very simple. What products do we carry? At what price and promotions do we need to have it at? And where do we need to place it at? Place it meaning, you know, which stores do we need to carry them on and what aisles do we need to display them on,
[38:26] etc. But if you think of every single retail solution that's out there in the market, they address pricing separately. They address promotion separately. They address space separately. They address assortment separately. So, when you're negotiating with a
[38:41] vendor like Danone or Unilever, you got to bring the products in. You got to price them right at the right cost. You got to procure them at the right cost. You got to price them right. You got to promote them right so that we can bring customers to our store. That's the basics of retail. The four
[38:57] P's, how do you balance the four P's? But the connected systems that we have in retail technologies are all disparate. You might have a separate system, as I said, for pricing versus promotion versus space versus assortment. And there's no playbook that exists today.
[39:12] So, the science is about how do you connect these dots between these four systems. And as Rob mentioned, it all starts with data first. If you're pulling in a system from pricing, you're pulling in systems from promotions. We are a heavily promotional retailer. We are not an everyday low price retailer. We are a heavily
[39:28] promotional retailer. So, we have both digital offers and store offers and personalized offers, etc. They all need to relate to each other. If you're calculating gross profit, you need to see the value of each of these promotions stacking up. But the today's state of retail, this is
[39:43] not just Albertsons. I think it's probably true for many grocery companies out there. They are They are localized. They are locally optimal but globally blind. So, when you're trying to predict units revenue and AGP of what if I do
[40:00] a markdown on a certain product at a certain price in a certain location, there isn't a clear way of doing that today. So, that's what we've been building in Albertsons for the last 18 months. Right from establishing the case of what
[40:17] we should build, should we focus on price, should we focus on promotion, should we focus on space, etc. So, we started I'm going to talk about promotion only today. It is the four P's, but I'm going to talk about promotion only today in how we are addressing it. So, what we are building is very simple. For
[40:33] a given item store week, we have 2,200 stores. Roughly, we carry about 31,000 items in a store. We have 624 categories of items that we sell in our stores. So, every store item week, what should we sell?
[40:48] What should we price it at? What should we promote it at? And where should we place our products? Is it on the front of the store? Is it on a traditional aisle? Do you want to do a display outside the store, etc. So, that's what we've been working on. And for Danone and others, it's in the
[41:04] refrigerated section. So, what should we do around these four P's? And I'll talk about promotions today, but that's what we've been working on so that we can project the units revenue and AGP for it. So, I did not steal this slide from Rob. It somehow seems like we are we're
[41:20] talking about the same thing, which is the sense versus predict versus decide versus explain versus act. What do I mean by sense? We have roughly about 200 billion rows of data. When you collect the transaction history over the last 3 years, and then you look at all our
[41:36] pricing changes, you look at all our cost changes, you look at all our promotional changes, that's 200 billion rows of data. So, you need to make sense of it all before you can do any predictions on that data set. So, when you have this amount of data, you can't give it to a large language
[41:53] model. There is no large language model today. We've been working with Open AI. Sam Altman was the first uh he presented us in the July conference last year in the developer conference of what we're trying to do with them. They don't have a foundational model that can be trained on enterprise data.
[42:08] It is trained on internet data. Whether it's whether it's the GPTs, whether it's Gemini, whether it's Grok, whether it's Claude, they're all trained on internet data, not enterprise data. And you can't do inference on 200 billion rows of data at real time to say what price and
[42:24] promotions that we should execute. So, we took a slip slightly different path to this. Most of you might know that large language models are based on deep deep learning models, and deep learning models are combined with attention blocks. The attention is what Rob talked out
[42:41] talked about, which is how do you find the correlation between these two things based on cosine similarity of how distant they are between an ecosystem. So, what we decided to do is to build our own attention block. So, we run our own deep learning models.
[42:57] We run our own attention block. For those of you who are familiar with the LLM terms, the QKV of an LLM, we are doing this for retail. What is the QKV relationship that we need to learn between the price, promotion, and placement? And especially
[43:12] when it comes to promotion, when I when I promote Coke on 4th of July, or when I promote Pepsi on a Super Bowl week, what would be the impact on Lay's chips? If you don't promote it, what would be the impact on Lay's chips? That's the
[43:29] relationship that we are learning. That's the attention block that we are learning, where we are pre-training a a custom large language model that we have on our own data set, so that the unit revenue predictions are there is zero hallucination behind it because it's our
[43:44] own deep learning network, and we are applying traditional science for calculating reliable reliability and calibration associated with it, so that we can tell our merchants with confidence when we say unit projections for store A, and our lead merchant for Jewelers just sitting here, when we say
[44:00] unit predictions for store A, he believes that the forecast I'm giving him is right. If you go and say my MAPE is 20%, my MAPE is 33%, it's not going to go anywhere. So, for us getting the predictions
[44:16] right, and for us to get the prediction right, we need to get the structure right. How do you learn these relationship, and we use graph neural networks for it. So, all our models that we are running is running on the Databricks platform. Our data is unified on the lakehouse.
[44:34] And now, once the data is unified on the lakehouse, you can run models on it. In the past, we used to run our models separately on a different environment, and data separately on a different environment. The latency associated with that is large, but if you're running a small data set, it doesn't matter. But
[44:50] when you're running it on 200 billion rows, and you need to to pre-train a large language model on this thing, that's not easy. So, that's why unifying the data and the modeling of the science on the same platform was critical for us. So, the sensing, the data part, the graph neural
[45:06] network part, and the predictions are on the same platform. Now, once you've done that, you can go into the deciding bit, because the merchants need to decide. You've got to give them options to say, "Do you want to go for price, or do you want to go for promotion?" And if you want to go for promotion, what is the item discount, what is the must buy,
[45:22] what is the buy X get X, what promotional strategy is going to work the best for you? So, that's the decide part of it. And the decide part of it, now, for most of you who have been used to working on deep learning, or even large language models, you know the hallucination problem. We have zero
[45:38] hallucination. Because we control the guardrails. You know, the traditional way of doing this is you give an LLM a large amount of data, and say, "Can you predict an outcome?" Then you apply guardrails on top of it to say, "How do you control the hallucination?" But because we're going structure first,
[45:55] because we have our deep learning models, we apply guardrails using mixed integer programming. For those of you who are used to mixed integer programming or quadratic programming, we apply the guardrails using that to come up with the deterministic outcome of what promotional recommendation makes the
[46:11] most sense. So, there is inherent ability in a merchant to trust in it. And those those data points can be explained, and I'll show that in a second, and how we explain it. And the act is the agent piece. So, rather than use the traditional
[46:27] reason and act framework of agents, we are using only the act framework of agents. So, the agents still need to take this recommendation and push it into the cost system, push it into the pricing system, push it into the promotional system, push it to the POS system, wherever the data needs to go. So, the agents have a
[46:44] responsibility to act, but not reason. Because we are taking the reasoning out of the agent because of the hallucination problem. So, that's what we've been working on uh for the last for the last year. And as I said to you, this is the graph that we are trying to
[47:01] build. How can you learn relationships between data? And Rob talked about environment data or exogenous data. When we have our internal company data and exogenous data, how do you bring them together so that you can learn these causal relationships? What would be the
[47:16] impact on weather on ice cream? And this is the use case Sam presented last year when he presented it. Uh ice cream sales was not the greatest last year because the weather was not the hottest last year in Northern California. So, how do you learn these correlations between this? Now, weather
[47:32] prediction is a science by itself. You can't predict weather that reasonably, that accurately. So, how do you build this model that allows us to do it? Now, if you think of a graph neural network, this is the graph neural network for retail. That's what we are doing. So, on the center, you see a specific item,
[47:49] like a 12-oz 12-oz 12-pack Coke. The blue items are the complements, the things that people buy together. The orange items are the substitutes, the people that people don't buy together. Promoting another one will actually kill the sales on the center item, the black
[48:05] one that you show there. And the size of the bubble is essentially the strength of the relationship. So, when you're building a graph neural network, you have to explain it to a merchant. And that's the explainability that I'm talking about here. You don't see nodes and edges and all those kinds of
[48:22] concepts that we talk in graph neural networks. You only see a item and the relationship between these items in what we call a promotional molecule. I myself, I do my doctorate on computational chemistry and self-supervised learning. So, the
[48:38] molecule thing is we're taking the transferability of that learning and applying it to retail and saying, "How can we visualize the promo molecule for you?" And for those of you who know this, who sat through my session yesterday, you know, if you know bananas, if you've ever eaten bananas,
[48:53] I'm pretty sure you have, the smell of a banana comes from just one atom. It's just one atom. You pull out the atom from the banana, there is no smell to the banana. It is the same for mosquito repellents. It's the same for tar. Tar is different. You don't You cannot pull out an atom,
[49:10] it'll still be sticky everywhere you go. So, you can learn this science, and that's the science that we are trying to build. What if you did something with Coke? What will happen to Pepsi? What if you did something to Pepsi, what will happen to Lay's chips? Lay's is
[49:25] PepsiCo Foundation is the the same for us. So, that's what we're trying to do. It's the same for Danone, it's the same for Unilever. It's like when we promote them together, what happens? When we don't promote them together, what happens? And as I said to you, nobody adopts a
[49:41] model that is not explainable. Now, the example I was giving last time yesterday was an MRI report. When When report goes from a radiologist to a doctor, nobody can understand it. Nobody can understand it. It says there is a 1.8 mm
[49:59] ischemic uh arithmetic on the second lesion of the of the brain next to the cerebellum. No doctor is going to communicate to the patient like that. But for us, we need to communicate to the merchant in their language. What is
[50:15] the impact on vendor landing cost on dead net cost? How much off invoices allowance will have an impact on D&C? These terms might not be familiar to you, but John as a merchant, that's what he dabbles with. So, that is what I'm talking about here. If I can't audit the
[50:32] model and I can't explain the science behind it, and if I just throw a large language model at it and say, "Can you reason on the fly?" First of all, on 200 billion rows, it's impossible to reason on the fly. You can't do inference on that bigger data
[50:47] set. Second, the explainability will be lost because learning these latent relationships that is not evident in our data set because of the fragmented nature of this 4P's is extremely hard. So, if you see the explainability
[51:03] here, one was the promo molecule I showed you. The other one is the explainability I'm showing here. When a machine is recommending a promotion, why is it recommending what it did? So, if you have a 75 cent promotion that is running, what is the causal impact of it? What
[51:19] are the halo items? What are the cannibalization items that impact the price of an item? And what customer lift do we anticipate? How will this service our customer? And how is this going to help our customer? That's the explainable AI. For those of you who have heard of XAI, it is not
[51:36] uh Elon Musk's XAI company. I'm talking about the science XAI, which is explainable AI. So, that's the explainability that I'm talking about. I talked about mixed integer programming or MIQP programming. That's the result of what you get from MIQP. You don't get
[51:51] this by running a large language model. But all this sits on a Databricks infrastructure. We run uh we run our data, as I said to you, on Unity Catalog. We run it on Lakehouse. We govern it using Unity Catalog. We run our model infrastructure on Databricks
[52:07] MLflow. Most of you know that. So, all our Spark jobs are running on the distributed data set so that we can run this for 2,200 stores at scale. You're talking about running computations worth in billions overnight because you only have a 3-hour window between the East Coast
[52:24] waking up and the West Coast closing. So, we have to run this between 3 to 4 hours. So, it's billions of computations that we run on that data set that's running on Databricks. We use Genie to explain some of this data, the Genie API, to you explain some of this data.
[52:40] And we have this unique partnership with OpenAI to develop their Chatkit platform, which is the enterprise offering of GPT. So, we don't use the consumer-facing version because of the data limitation that I showed you. So, we have this unique partnership between OpenAI and Databricks to co-develop this
[52:57] experience of how can we build this experience layer that's understandable for retail and merchants so that they can make these smart decisions. So, that's the ecosystem of Databricks that we are that we are partnering with all the way from data to models to Genie to the
[53:15] experience side of it with OpenAI. So, that's the full stack that we've been developing. And I should say that, you know, one of the the most easy things about Databricks has been that I have access to many of these product teams. You know, it's not one
[53:30] product team. It's not the Unity Catalog team. It's not just the Genie team. It's not just the MLflow team. It's not just the AI Gateway team. We are bringing these together. And for a company that's been building products, as you know, just like the four P's of retail, there tend to be
[53:45] four different directional product sets. But, I've been grateful that Databricks has been able to collaborate in bringing these different solutions together in a unified voice. When I started, we had things like rate limiting and so on and on and Genie, which was not usable at
[54:01] all. But, here we are now, we are almost ready to open this up to, you know, we are a 300,000 company organization. Not everybody will need access to this, but at least our merchandising team, which is about 2,000 strong, will have access to this. So, that's the journey that we've been
[54:16] up to, and I'll close this with, you know, for those of you who are trying to build There is nothing wrong with the large language model and an agent. It serves its purpose. There are There are places and use cases where it makes the most sense. But, for us,
[54:33] understanding the structure of our data and understanding these latent relationships, it's not the model that needs to talk. It's the structure that needs to talk. If the structure talks, the model is easy. If the model is talking, it's almost like having a clever linguist pretend to be a doctor.
[54:50] We don't want the clever linguist, we want the doctor. But, we need to make it explainable back to the merchant. That's my talk. Thank you.
[55:09] Well, I I have questions for Karthik of how he got from chemistry to retail, but we can talk afterwards cuz I'm very I don't know if anyone else is, but I feel like I got a chemistry lesson and a retail lesson all in one. So, thank you. Um I am inspired as much as I hope as you all are um from D and Karthik. And I'm going to take just a brief moment. I
[55:26] know you've heard innovation from them and kind of what they're doing both at Danone and at Albertsons. We also released some um at Databricks every year we do our customer awards. And so, I just want to highlight the winners right now of the customers that won our
[55:42] awards this year. First up is Pepsi. Um this is one that you have probably seen Pepsi. You've heard from them as well here. They're doing incredible things. You had heard Megesh on the main stage. Um just everything that they're doing from Unity Catalog to utilizing
[55:58] Genie as well. Um Databricks apps, custom AI, just really exciting use cases coming from them. Again, I think all of this serves as inspiration for all of us as we are building the future of um our retail consumer goods and travel experiences. The next I wasn't
[56:13] aware of Fonterra before I came to Databricks, but they are one of the largest dairy um sellers in out of New Zealand but having global networks and tapping into Databricks to really bring all of the insights from our global networks into one so they can make really quick
[56:28] decisions, especially from supply chain all to financial decisions as well. So, congratulations to our APJ winner for the excellence award. And then, the final one that I um will lead us greatly into our next session is leader you're
[56:44] about to hear from along with um Unilever as well. But, Virgin Atlantic is doing incredible things. You probably seen if you've been down to the demo area, you've seen um Virgin Atlantic down there as well. Um but, what I love about what they're doing is really not
[57:00] just building a data platform but building a decisioning platform. And I've heard that both from D and from Karthik as well is that is the mindset to take into what it means to really be a transformative business, not just be siloed in data, but to make data the decision layer that then serves everything else that really drives
[57:16] everything from the customer-facing interactions, but also the operations behind the scenes. I'm not going to steal any more thunder cuz I'm sure um that Richard will get into some of that. But again, congratulations to Virgin Atlantic. And with that, I do want to
[57:31] bring up our last session of the day. So, again, what you've heard a lot today and throughout Data and AI Summit is all around transformation. But, when I was when we were building this out, what I got really curious about is how does all of this new technology actually change the way that we as employees and also
[57:47] our corporations also function and what does that mean? And so, we're going to dive into some of those topics. So, I want to welcome both Louise and Richard onto the stage with me to have a little fireside conversation.
[58:13] It's my little forward. I know. It's a little bit right. Okay, so we talked about this being the future of work, but before we get into talking about the future, I want to talk about the present. So, you both have been on a journey with Databricks with data with AI. So, first introduce yourselves, who
[58:31] you are, the companies that you work for, and kind of where you are with unified data platforms, with AI, and kind of what what is the current state of of how you're operating within Burden and then you know, Unilever. So, do you want to start, Louise? Yeah, sure. Um so, hi everyone. I'm Louise White. Um so, I've been at
[58:49] Unilever for 25 years now. Um and I've honestly been through several transformations during that time. Um I can say that, you know, this shift that we're seeing now with AI is absolutely a fundamental um change that we're seeing. Um so, me personally, you
[59:05] know, within Unilever, I lead out our data strategy. I've been on our data platform journey for the last 10 years. Um been working very closely, obviously, with uh Databricks on this journey. Um and you know, my role at the moment is really ensuring that we have the
[59:20] right data foundations, the right strategy around our data within Unilever to really being sure that we can power out the AI initiatives that we're have as an organization to help us to grow better, to help us to win with our customers, and really generate the right experiences for our consumers, for our
[59:36] people as well. Um for me, you know, I I also lead out marketing and kind of consumer XOps. So, I'm kind of right in the pivotal point where it's about, you know, understanding the data, driving the strategy, but then applying that data um into the execution side of things as
[59:52] well in terms of the products that we build. Um I always say, you know, when we talk about AI, I'm super excited about the AI journey, um but I'm equally terrified by it. Uh you know, I lead out a lot on data, so for me, I know there's a lot of dependency on getting the data right,
[01:00:09] having the right structures, having the right governance, and the trust in that data, um which is absolutely critical to make sure that you're getting the right result out from an AI perspective. So, we've been on a journey in Unilever. Um we've got, you know, a really good solid foundation not dissimilar to what we've
[01:00:25] heard um today from um from Danone and from Abacus and terms of um Albertsons and terms of how they built out their foundations. Um but for Russia, it's all about having that connected unified data layer, having that well-governed, um and really
[01:00:40] ensuring that we are empowering our employees at the heart to be able to leverage that data across the organization to drive insight and growth. I love that. Yeah, so I'm Rich Masters, and I'm the vice president of data and AI at Virgin Atlantic. I'm a bit of a boomerang. Um I've actually been back about 2 and 1/2
[01:00:56] years. Um I was there 2018 to 2021. Kind of started Databricks actually in 2018 with the notebook environment for data scientists. And unfortunately, people just using it as a as a Python environment, not using the RDDs, and really it's quite quite expensive at the time um to be using it in that way. Um
[01:01:13] but when I came back, we we actually went really hard on Unity Catalog to to kind of really bring everything together. So, so I'm in charge of of all our data when it for decision support. So, not system to but anything else that comes out of it Uh and creates that decision intelligence, but also AI. So, who do what do we buy, what do we build,
[01:01:29] who do we partner with, what does it mean for us, what does responsible AI mean for us at Virgin Atlantic, and how does it kind of empower our people to do what we need to do, and how do we assure the engines we're using underneath that? So, we don't necessarily build everything. We have a great partnership with Fetchr that works on our dynamic pricing, for example, but we use Databricks to feed the data to
[01:01:46] that engine, assure that, make sure it's operating in the right way, and we can feed more more sources into that at the same time. But, we operate right across the organization from engineering and maintenance through to the customer experience, as you said before, our commercial teams, finance, legal, they're all using aspects of Databricks
[01:02:03] alongside our own kind of open AI partnership as well there. So, we've kept it really kind of streamlined and quite slick there, but it means that everyone can kind of can really get close to data and AI and what that really means for them at Virgin Atlantic. I love it. I love two things that you both said. First of all, I think Louise,
[01:02:19] a lot of people in the room can empathize with being excited and terrified about AI and the potential there, and also both of you kind of hit on the ultimate service of the customer, and the ultimate bringing and making sure that everything that you're building is
[01:02:35] in service of them, and it kind of brings me to I I heard the US Chief Data Scientist. He said what he believes is like the purpose of data is in his role is to unleash the power of data for the benefit of all Americans, and it that
[01:02:51] kind of resonates with what you both have been saying, and and I'm curious of how you kind of bring together your data strategy with the missions of both Unilever and Virgin Atlantic, and what you see is how does data and AI benefit those missions?
[01:03:07] So, I'm I'm as we mentioned a bit before with signal and noise, but kind of I I I don't like data. There's too much of it. I was yeah, an astrophysicist before, and a lot of that was capturing a lot of noise, a lot of static, getting rid of all that and just focusing on the data that's really important. So, we really kind of focus on that ethos. Get rid of
[01:03:23] the data. We used to call them data reduction pipelines, actually. You want to reduce the data as much as possible. That's really important. Um so so we we we really focus on that so that you can really understand it and and then take action off the back of it. So, um we prioritize that in the right way. We actually with our with our people team,
[01:03:39] with our FP&A team, with our corporate strategy team, when we do big data and AI kind of change pieces, they are all involved with me in how we make those decisions so that it's not all about the data. We don't have a separate data and AI strategy. We have a corporate strategy and everything we do is in service of that.
[01:03:54] Well, that. Um and I think, you know, at Unilever, the the way we look at the data and AI side of things is really about the growth engine for the organization. So, again, it has to cut across end-to-end um across all of our operations, across all of our capabilities. Um you know, for us, when it comes to the the kind of
[01:04:11] strategy side of things, it's all about ensuring that we can um navigate the the wealth of data that we've got. We are a hugely data-rich company. Um but I think it's really important to make sure that we bring the right views on top of that data and turn that data, most importantly, into information and
[01:04:27] into knowledge. So, I'm kind of super excited about some of what's been presented here around like ontology cuz that context side of things is super important. Um and that's one of the big elements of our strategy at the moment. How do we bring about those kind of connected views? We've got our solid data foundation, but let's
[01:04:44] turn that into the semantic view so you've got consistency of data flowing into your AI models. Um but then also making sure that we've got the right governance program around that as well so that we can embed trust by design, we can have observability by design, um and really make sure all of that comes
[01:05:00] together. So, it really focuses on solid foundations, semantic views to drive consistency, and really having a well-governed estate um for to power our growth within the organization. Yeah. Can I promote a bit the product? for The gene ontology.
[01:05:16] I don't know I'm super I'm super excited about gene ontologies because there's something in that came out the onto rank kind of concept. This this idea of putting some extra data in there that we might not have thought about before, which is really who is accessing it, what is their role in the organization and waiting that accordingly. Now, it might not solve it fully, but as a as a
[01:05:32] kind of a step change in kind of how you think about navigating graphs here, that's going to be I'm really interested to see how that as the new semantic layer and actually makes this practical for us to go, well, actually I've got all this stuff connected together now, but how do I filter again another layer of noise in this massive graph we have?
[01:05:47] Um but to what's really important, how we waiting that and it's really interesting and innovative in the way you're considering different data points now to to do that. So, really really excited about that. Yeah. Well, that's that's a real kind of selling at the moment, but like yeah, I'm I'm genuinely very excited about that. Well, you're officially a Brickster freezing super excited. So,
[01:06:03] thank you for that. Um you both kind of mentioned things that, you know, I think maybe two-ish years ago, a lot of corporations were seeing the boom in AI and wanting to like bring it in and asking everyone to bring AI into their
[01:06:18] strategies. What are you doing with it? But as you both probably both know intimately, AI isn't magic. Like it doesn't just automatically improve the experience. It doesn't automatically bring, you know, a strategy from A to Z. So, when you think about your AI
[01:06:33] strategies, I'm trying not to use data since you don't like to use it. But what have to be What are the enablers for that? So, when you think about the conversations that you're having with your executives and they're wanting to bring in AI into your strategies, what has to actually be underneath that? And
[01:06:49] you mentioned some of them from this conference, but I'm curious as you thought about your whole strategies, what was really underneath there that actually could then power Yeah, the future of as I mentioned within our strategy, I think you know, that the for us it's absolutely about the critical data foundations themselves.
[01:07:04] Um you know, we I said I've been on this journey for 10 years. Um, we've spent a lot of time in Unilever making sure we can bring our data into a unified platform um, and really have that connected. So, for us, you know, the the must-haves from an AI perspective has absolutely been about having the right
[01:07:20] data at the right time, serving out the right models. Um, that's been absolutely critical. I think, you know, when you look at the the foundation side of things, it's not just about tech though. Um, so we've spent a lot of time in Unilever really empowering our
[01:07:35] employees, demystifying the whole AI side of things. We've trained over like 40,000 um, employees to date and and we have continuous um, programs that are going on, which is really about trying to make sure people understand that yes, AI is not magic and
[01:07:52] certainly some of our leaders do think it is magic. You know, you can um, point AI at anything and it will magically deliver what you need. Um, but it's about really helping people to understand when to use it, how to use it, the limitations of it, and what you need to do to kind of enhance the
[01:08:08] output. It's all about the outcome that you're trying to achieve. Um, it's not, you know, the the AI is a tool and a capability, but ultimately it's about you the human guiding it in the right way to get the right outcome from the organization. It's a little bit magic.
[01:08:24] a little bit magic. I think it's a little bit magic. It just the ability to So, the the step changes in kind of that little bit of magic that comes in, right? But ultimately it fundamentally completely agree depends on that having that core foundation to then govern and control the step changes in capability that happen that suddenly can then make
[01:08:39] things kind of run a bit rampant. So, you know, earlier in this year the ability for people to build things now and building meaning a very different thing now changes your strategy of what you do acquire and how you start to work with different vendors, etc. in there is a real is a real consideration point for us, but that building has to happen in
[01:08:55] the right environment in the right way and you don't want to stifle ideas, but not every idea is a good idea. Um, so you kind of need ways to be able to kind of distill that as well. So, it is part of our AI strategy, like I said, linked to the corporate strategy is around here, what is the outcome you're going to say? What's the action that's going to happen? How does it fit with our own
[01:09:10] our own brand and what we are what our purpose is? As you know, our purpose has been the most loved travel company. So, everything we do should be in service of that. Optimizing EBIT is part of that. But, um but then there's also net promoter score and risk avoidance cuz we're very kind of regulated and operational organization as well. So,
[01:09:25] actually that there's a real understanding at our leadership team level around the opportunity for AI, but also that real understanding of risk being an airline that really helps us then then understand what are the right things to be targeting, which naturally then helps with cost and the right engagement with the teams as well. Uh
[01:09:41] and yeah, with with the training programs we've got an uh an academy in but really a champions network as well um that really then helps kind of seed um those mindsets in the organization, not just the skill sets, but the mindsets of this. And then as different use cases come in, pennies start to drop in other areas as well, which has been really
[01:09:57] interesting to see. That That brings up a really good question, too, of like a different type of enabler. So, like you just talked a little bit about the the data enablers, but then in the foundational. But you both also mentioned people as a fundamental enabler of data and AI. I'm curious as you think about the
[01:10:13] evolution of your teams, um whether it's when you're mentoring, when you're hiring new people, what are you looking for and the people that you bring on or in your own teams that can make not just your organizations and not inside of Virgin
[01:10:29] and inside of you know be successful, but in general, like what kinds of people are going to thrive in the next I don't know, say 3 to 5 years within in the business transformation. For me, number one is curiosity. 100% curiosity. If you're not curious in this
[01:10:45] world of kind of fast-pacing change, that continual development of yourself, really wanting to understand, you know, we we encourage our teams when they're doing some of the the magic with AI to get under the covers, to have a look at the code, you know, to educate yourself.
[01:11:01] You know, you might not be especially for those non-technical people who may not be doing the code writing themselves, but have a look at it. Have a look behind the the covers and have a look and understand how is, you know, the prompts and the that you put in actually translated in the background.
[01:11:17] But yeah, from a people perspective, I do think that there's so much focus now on, you know, how do you we enable our people to be more effective, more efficient. So we're really looking for those people who are going to come in and be wanting to know about the next thing, that curiosity,
[01:11:33] that that mindset of, you know, I want to know what's happening in the future. I think the other thing for me, the second thing for me would be that kind of translator role. So how do you ensure that you can talk the business language and help to translate that into the outcomes that
[01:11:50] you're driving as well. So understand the business that we're working in is another the key element that we look for within our talent. And really somebody who's just, you know, really going to be there in the moment and be able to embrace all of the tools, hands-on experience, get, you
[01:12:06] know, get close to them, really start to work with them. So, yeah, they would be the key things for for us when we're looking for talent at the moment. I love it. Anything from you, Richard? I completely agree with curiosity is kind of number one that's is a fantastic one. A couple more I'd add is there's this
[01:12:22] the the how to understand if something's right. So how do you yourself, in varying degrees, measure the impact of it, also understand like is can I verify that output? Is it doing what I need it to do? Because especially when you think about kind of talking to a chatbot, an identical process,
[01:12:37] when you ask someone else to do something for you, if they're working from home on the other side of Teams, you have to understand whether that's right. Is it is it is it a good leadership principle to understand like how am I going to get trust in this? I've got to go and defend this somewhere else. I'm getting this information from all over the place. So at all levels, how can I do better at understanding is
[01:12:53] it right and translating that for my audience at the same time? So, the skill sets, the different tools now to be able to do that is really important for for team members. And the other one is presence. So, actually being present and physically there with people to then work through a problem because again,
[01:13:09] you can get the the engines now and the ability to build um full stacks off doing it on its own that rather than sit there and watch it for 15 minutes like, you know, we tend to do when you you write etc. You can now work with someone else, work on design, collaborate more. And again, if you're
[01:13:26] just behind the keyboard somewhere kind of remotely on this, again, that's an agentic process. That will become more and it So, actually being present, being there and working with people will mean that you can actually as it come up earlier like rethink your organization, what products, what new markets can we operate in, how can we innovate around
[01:13:42] what our business needs to be, and physical presence will really help you do that as well. So, people that are willing to do that are really important. Yeah. And and I think that for me, it's about the expectation from AI is not that it will kind of do the job for you, right? There's still an element of the
[01:13:57] human side that needs to come. And it allows you if you get the the AI to do the lion's share of the capability for you, it gives you the opportunity then to bring your creativity, to bring your problem solving, to really challenge the outcome that's coming out and, you know, not just 100% trust what's coming out of
[01:14:14] the AI, but challenge it. You know, ask yourself questions. Is this what I was expecting? Is this aligned up to where I thought it was going? What else could I be looking at to kind of build further trust in the output that we're getting here? So, I think that has to be something that as, you know, an individual, you look at how can AI make
[01:14:33] me more efficient, more effective, drive better, faster, you know, more trusted results at the end of it. Yeah. And what you both are talking about too is these aren't just tools and and um needs from the technical folks, right? Like, these are skills that need
[01:14:49] to be across the organizations. And you mentioned Richard earlier, too, like, you are working across from customer experience all the way to the technical teams, to the marketing teams, like you mentioned in working with. So, there's a lot, I think, of excitement that AI is coming to those non-technical roles, but
[01:15:05] there's still some learning, like, I sit in marketing and I am definitely not super technical, but I'm excited about, for instance, Genie, like, being able to talk to my data and tap into that. My follow-up question to that is, how do you think about, with all of this excitement coming in about tapping into
[01:15:22] the magical power of AI, um, that, of course, causes what we have all heard of agent sprawl and token maxing and really bringing in this, okay, now everyone is using AI and that's exciting. How are you thinking about
[01:15:39] keeping that in check while also not stifling creativity and the ability to get excited about building and deploying products? I think well, that having that governed environment, having the observability of what people are doing, so is is just super helpful to be able to control that
[01:15:55] in the right way. The token maxing thing's really interesting. We're not really seeing that. We get the odd spike of someone just using kind of the wrong model or something like that. I think when when you're an Amazon or an Uber, yeah, you might kind of blow through. We'd be interested to see what anyone else is really struggling with that. But actually, um,
[01:16:11] we kind of struggle to meet our commitments sometimes in terms of just getting more and more use cases in and you we try to optimize as much as possible, but actually, what we're what we are what we're what we're seeing is just kind of, yeah, the odd spike there, not real token maxing. We don't encourage it, either, um, to be fair, but, um, but really having
[01:16:26] that kind of core platform to observe. And when you do get some of those spikes come in, it's very easy to then kind of whack-a-mole and and knock them down. And then you actually kind of really iteratively build on your governance and control framework, cuz no one's got it solved at the moment. No one knows whether going from a 5.4 to a 5.5 is really going to improve productivity at the moment. I just know that actually
[01:16:43] um multi-turn versus single-turn is having a different effect on a little bit of a cost spike. Actually, I can then start to optimize that across the platform. Uh and again, you know, what we saw earlier with some of the um was announced today through what's going on with Genie code, that's really kind of a really exciting step in being able to
[01:16:58] add another layer of control there to even just giving someone a budget and helping them kind of tier their models and move through. And it will help the teams really understand, well, what does intelligence really mean here? What does it really mean to help me kind of build in the right way? So, I think some of those aspects coming in are going to be really helpful as well. Yeah, and it's finding the right
[01:17:14] balance. It's the right balance between having that flexibility from an innovation perspective to having that kind of governance control in place as well. Um so, very similar, you know, we look at it through that observability side of things, making sure we really understand what people are doing, having the right guardrails in place, and kind
[01:17:31] of principles in place, so that we can guide people, so that we're not doing the token maxing, so that we are um you know, using the tools for the right thing. But, we want to encourage people You know, it's a it's a really exciting time. We want people to get hands-on, be able to really work with the tools that they've got in front of them. Um and
[01:17:47] it's that kind of freedom within a framework. And, you know, as uh Richard said, no one's really got this right yet. Um we're all finding our way, but I think it is about trying to find that balance. Um so, that we can at least have that innovation versus the governance as well. I think the
[01:18:03] the struggle for us at the moment is the change management piece, right? It's the human change management piece of like, we've got we've got all these great controls and got this platform and built a thing that that is optimized and you know, it does what it's needed to do, then how are we then really changing our processes and our behaviors around that? What is How does that then feed
[01:18:19] back in? And what more could we do off it? But, we've we've got some big wins there, but I think that's a piece that there's still this big gap that that I think people could really focus more on as well. Yeah. Yeah. And do you think that leads to to the conversation around, you know, we mentioned token maxing, you mentioned
[01:18:34] EBITDA before, executives are now looking for what's the ROI of the data strategy and what is like when you're bringing in new products and new platforms, like how are you seeing ROI? That's sometimes really hard from a data side because you stretch across so many different functions and you're empowering all of those. So, as you
[01:18:50] think about that change management and then also leveling up and showcasing how powerful these solutions are to your executives, how do you think about like I guess is it easy or hard to show that ROI? What's really easy, what's really hard? How are you seeing that? But we see it um relatively easy on a
[01:19:07] use case basis. So, when we're doing our pilots, when we're looking at the the smaller use cases, I think the the ROI is easier to measure um because you've got your proof points. I think where it becomes more challenging is when you get to scale. Um so, that's the bit that we're kind of looking at now is, you know, how do we
[01:19:23] ensure that we can get the right level of ROI when we are doing the the scale implementations and within our organization, you know, we're constantly looking at, you know, for the 3.7 billion consumers that we hit every day, you know, for the all of our employees that we've got within our organization,
[01:19:39] when we're looking at, you know, the all of the I don't know, 300,000 kind of customers that we're engaging with. Everything that we do, even a small change, that scale gives us a capacity advantage. So, whenever we're looking at any of our things, I think we we do have a really good kind of methodology around
[01:19:56] the return on investment for the smaller scale activities, but I think that that large scale, we're still kind of working that through at the moment and saying how can we truly measure um that return on investment. But again, it comes back to just ensuring you've got the right structure, the right guardrails, the
[01:20:12] right governance in place, and the measurability around that so that you can at least have some of that uplift. Yeah. The building of the capability is definitely getting, you know, a lot easier and that's not never a real problem when you're estimating this. The approach we take is we take a very domain-focused approach and change that
[01:20:28] domain one at the time. Because then you focus on all the different processes around it. So, our customer contact areas when we're focusing on the moment our maintenance area in there. Um in our marketing we've we've done some big stuff in kind of Q1 on that to help with our sole launch, which is fantastic. Um and what we do there is we
[01:20:44] we go very granular on the value pools that are available. So, is it IVR containment? Is it agent coaching? Um how we deal with kind of EC261 claims, etc. That's all automated processes there. But really then bucket those in terms of what's the real opportunity for that value value to come out. How is that going to happen?
[01:21:00] Is it attrition rather than disrupting the team and understanding like how we better deploy people to to do um sales versus kind of sat down on a 45-minute call. Um is it kind of outsource spend? Is it software spend? Really really going granular on where that cost is. What are the constraints of unleashing
[01:21:16] that cost when we build a thing? Is it contractual? Is it going to be to do with kind of what we've got coming up in the rest of our with our fleet when we've got new premium seats coming in. What's that going to do to kind of demand in it? So, we do a lot of that up-front planning to then really understand when that value is going to be released. So, this kind of J-curve certain value, we've kind of been really
[01:21:31] up-front around open and when that's going to happen with the leadership team so that we know that everyone's on board with this is the pace of how the change is going to happen even though we're going to build it really quickly. Um and then we have to kind of keep adjusting our actual cost base because some of these things become very easy to build and then you got to try and then move more value into that as you go. So, you
[01:21:47] got So, the the agility comes from really trying to identify value along the way even though up-front you kind of plan this as well. So, you can get even more ROI as you go. Being really conscious of that as as you do it not being stuck to this original view of what the value is going to be. And for an airline when everything's been kind of waterfall and fitting out planes and
[01:22:03] stuff, that's been a huge kind of mindset shift for us. Um and the other thing there is that it's the vice president that owns that outcome. That owns the has the accountability for realizing that value. I'm going to build the thing. My team's going to build the thing in the right way, but then they've got to work with their team and work with my team to change that process to
[01:22:19] realize it. And a really challenging thing there is normally when you go through machine learning data science and analytic with analysts and managers, you're supporting decisions or changing or helping automate decisions in some degrees. For VPs, you're really taking more decisions away potentially if you don't do this in the right way. And that
[01:22:34] disrupts kind of fundamentally what their their role particularly in operational spaces what their role is. So, our VPs have been fantastic in embracing that change and having their mind-sets in the process of going, "Well, how can I really change how traditionally operational function can now can now be different as well?" So, that's been a real benefit um
[01:22:50] uh at Virgin. And I think it is those levers. It is, you know, when you when you're driving any of these initiatives, it's not just one thing that's going to change. There's several levers within there that's going to change. Um and I agree with what Richard said, you know, for us it's all about it's not just about activating the change, driving the
[01:23:06] initiative through AI. It's actually about the adoption of that. And for that, we also have a really good kind of top-down approach. So, we have a lot of sponsorship um behind the AI initiatives that we're landing so that we can ensure that we can get them properly embedded into the teams, we can get them adopted, we can get them used, and then we can
[01:23:22] start to see the real value um come through from that. Yeah. I I want to bring this in because you guys have mentioned so many things that have have shifted the focus to like the future of like you mentioned like reshaping what the airline looks like.
[01:23:38] And we we were even talking yesterday about how do we change the mindset of folks to stop just fixing problems one after another, but fundamentally changing business so that it looks different and it's better for our customers in the future. So, as you
[01:23:53] think about the future of consumer goods and travel and hospitality in 3 to 5 years, what does that look like? Like what are you seeing in in in your people, in your technology, in your services? Like what is going to be the future of of our consumer-facing
[01:24:09] industries and how we serve our customers? So, we future absolutely kind of total re-imagination of terms of how we as humans are going to operate in that environment. You know, we're going to have a lot more capabilities at our fingertips. We're going to have a lot more trust in the the data, the outcomes
[01:24:27] that are coming out of the initiative. So, I think there's a you know, for me as a um when I look to the future, there's definitely that human component of we're going to be able to use our brains, use our creativity, use our problem-solving a lot more, and do more value-adding activities. That
[01:24:43] human-in-the-loop component becomes even more important as we go into the future. Um obviously, there's the whole agentic commerce, you know, our our employees, our data, everything has got to learn how to talk to consumers as
[01:24:59] well as talk to machines. So, there's a real transition that we've got to do to make sure that we've got the right components out there, the right data points, the right metadata to really ensure that our data can talk to the the the machines that are going to be making those agentic decisions on behalf of our
[01:25:14] consumers. Um so, that's going to be a huge shift. We're already seeing that shift. Um and that's only going to get more and more um as we start moving forward. You know, it's not even humans to machines, it's machine to machine, yeah? So, it we're seeing that shift already. Um and that's going to continue as we move forward. And I think, you
[01:25:31] know, from a data perspective, I think that trust side of things and the governance, you know, governance has always been an afterthought whenever we've been dealing with data. Um that's becoming more and more important, that ownership of the data, understanding um you know, what we're doing with it, who owns it, how we use
[01:25:48] it. That shift is happening from an organization perspective and is resulting in a lot of changing how we work, how our um leaders drive the organization as well. So, a lot of emphasis on ensuring we can win in that agentic world, machine to machine. Um
[01:26:04] making sure that we've got kind of that employee fully empowered with the activities run through AI, but that human component being able to drive the outcome that we need from an organization and to to win in this world. So, I see dramatic shifts particularly for our employees
[01:26:22] and exciting. It's going to be a very exciting journey. The way we kind of think about it is making our making our business more accessible to agents. So, you don't necessarily know need to go and build a bunch of agents, but you need to think about all the integrations and the tools now available to where
[01:26:38] this may be and how this is going to evolve more into agents and harnesses and meta harnesses and all these things that kind of go around it. We were the first airline to have our app in chat GPT for example, which kind of blew people's minds a little bit weirdly all the LinkedIn experts kind of messing me. Well, all that we did was wrap our search in MCP, right? There was there
[01:26:54] was no new agent or app we built there. We just did something in a day. And it was great to be there and it was great and it's helping us learn about that sort of that side of things, right? But making ourselves accessible to these new platforms that come out, I think it's going to is it's going to be more and more important and not overthinking
[01:27:10] or overcomplicating what that what that really means and what's what's what's being built there. So, we've got kind of that ethos running through to make that accessible. The industry's got a big change coming in as well in airlines this modern retailing concept of being able to simplify the whole kind of booking process and the basket and the different products one might want to put
[01:27:25] there really gives us that guiding light for how we want that business to be accessible to and how AI will kind of support that as well. But then ultimately putting our people first here with Virgin you have this kind of magic touch of our crew and that's where we have our big NPS spikes is when you talk to our people.
[01:27:41] So, we try to encapsulate some of that in for example in our AI concierge when our people can't be there with you. So, if we distill that into that we keep adding capabilities in there that's really a nice way for us to kind of keep evolving that process and then when we can you can meet our people they're then augmented by AI and apps
[01:27:57] and our connectivity now that you get with things like Starlink you can then really just make everyone connected and really tied to the to to all the different ways you can now have a great experience on on a trip. So, that's going to help us do that. I love it. You guys are giving confidence just like the encouraging did well that human is always going to be
[01:28:13] there and we're getting more empowered through the technology and I am super excited for the future of all of these industries. So, thank you both for having this conversation with me, with us all today. Thank you all for joining us. Um I think I might have one more slide that'll tell you to take a survey.
[01:28:28] So, I'll throw that up for you. Um but thank you guys again and really appreciate the conversation. So, thank you. Thank you.

Learn more about the Databricks Data and AI platform.