How Databricks Revolutionize Intelligent Enterprise AI in ASEAN with Patrick Kelly
Fresh out of the studio, Patrick Kelly, Senior Director for Digital Natives, Startups & Enterprise and Commercial Sales in Southeast Asia at Databricks, joins us to discuss how data intelligence is powering enterprise AI applications in ASEAN. Beginning with his career journey from network engineering to tech leadership across Asian markets, Patrick explained how Databricks pioneered the Lakehouse architecture and integrated generative AI into enterprise workloads. Emphasizing the critical role of data quality in AI success, he showcased compelling customer case studies from across ASEAN and revealed striking generative AI trends in Asia - notably that 85% of organizations lack proper architecture to support AI workloads, reinforcing that clean data remains foundational for effective AI implementation. Patrick concluded by sharing his vision of what success looks like for Databricks in Southeast Asia.
" We did a survey with The Economist globally which obviously included Europe and APAC as well. And we asked the question, 'Does my organization's current architecture supports the unique demands of AI workloads.' Basically 85% said, 'No. We don't have the architecture to support it.' Some partially does, but it needs lots of modifications. So we can still feel a lot of people are still in the early stages and that data point ties back to: 85% of GenAI [proof of concepts] has not gone into production. I think that another interesting point is, 'Does your architecture connect AI application? -your relevant business data.' which is probably nearly even more important for me. Again, it was still about 80%- 'We don't have that.' Because that business data is all over the place. Without the clean data, you cannot get good AI." - Patrick Kelly
Profile: Patrick Kelly, Senior Director for Digital Natives, Startups & Enterprise and Commercial Sales in Southeast Asia at Databricks (LinkedIn)
Here is the edited transcript of our conversation:
Bernard Leong: Welcome to Analyse Asia, the premier podcast dedicated to dissecting the pulse of business, technology and media in Asia. I'm Bernard Leong, and I often inform the decision makers in businesses that data is important for artificial intelligence to work. How do we ensure enterprise AI applications power the businesses in Southeast Asia? With me today, Patrick Kelly, Senior Director for Digital Natives Startup and Enterprise Commercial Sales in Southeast Asia from Databricks to discuss this subject. Patrick, welcome to the show.
Patrick Kelly: Thanks, Bernard. Great to be here.
Bernard Leong: I should mention, we were ex-colleagues and you are probably my mentor and boss when I first joined AWS [Amazon Web Services] and really guided me through the launch process. So thank you very much, Patrick, for doing this with me.
Patrick Kelly: Yeah, those were fun days. We can talk about that today for sure.
Bernard Leong: Sure. Of course. Without doubt, we always start with the origin story of our guests. So with your origin story, how did you start your career?
Patrick Kelly: Surprisingly I'm in sales now quite a long time, but I did start in technology. I was a networking engineer when I first started my career working a lot in enterprise networking, Cisco, Juniper, doing banking systems and stuff like that. That was pretty interesting. I was really into that. Then I moved into telecom. So I joined Ericsson, which is the big Swedish telecom equipment vendor. I'm Irish, so that was based in Ireland. But then I joined the professional services team, global services.
So I did projects all over the world. It was fantastic. Like in Brazil, in Jordan, in Australia and then landed up in Japan. I went to Japan on a four-week assignment, and I ended up staying five years. In Japan was interesting because I went from engineering into consulting. So network consulting for SoftBank. Back then LTE was just launching around 2009 and SoftBank had the exclusive rights for the iPhone. And the iPhone was a new beast on the network. All the signaling and we didn't know how to handle it. So we did a lot of consulting work for SoftBank, how to manage the load in Tokyo and really high densely populated areas. So that was pretty cool.
From there, I moved into sales, selling the services, selling the hardware, selling the software. And then from Japan, that brought me over to Singapore because in [the] telco [business], Japan and Korea are always number one. They're first with 5G, first with all the new technologies. When we started doing virtualization on telecom networks, we brought that over to Southeast Asia. We worked with SingTel and Telekom Malaysia and brought that all around the region as well. From there, after about nine years, I joined the startup world, so I joined a startup doing IoT, which was Jasper at the time. That was a really interesting role. And that then led me into AWS, where I spent about five years doing different roles.
Started with the IoT business, then we did the analytics business together and machine learning business, Bernard, and then took on the ISV business, which was our B2B software sales.
Bernard Leong: We collaborated a lot during that period in AWS.
Patrick Kelly: A lot. Yes. I remember in Malaysia we did a lot of good stuff, good case studies as well. And then finally the role was both ISV [independent software vendor], but also digital native as well, so the likes of Grab and Traveloka and all those iconic B2C young customers. Then it led me to Databricks where I'm doing a similar role. In that digital side. But then I also have more of what we call commercial or emerging enterprise: traditional companies, but very big across Southeast Asia who are trying to understand how to use data and AI to solve business problems.
Bernard Leong: How did you actually come to this present role with Databricks?
Patrick Kelly: Yeah, I was thinking about when I was in the process of it, and I thought back to our days when, back in 2019 when we had the DAMM team: Data and Analytics team and Machine Learning. Where we were talking about those problems. How do we solve problems with data? Like customers have a lot of data, but for a lot of them it was in a data lake or a data swamp, they couldn't get real insights into it.
We helped a lot of customers with that and especially with our machine learning solution lab, solving really tough problems through data science. When I was in the process with Databricks, I was thinking about how to get back to something that's really specific and into a certain technology domain. Amazon is fantastic. You have all of the services and all of the technology, but you become super broad because you're selling 200 services. But Databricks has a very defined focus on a data platform. We're building intelligence on top. It's AI as well and that was something that really excited me, and also building a team and building a business again, because I think data and AI now, or even last year, was probably at the same stage as cloud was maybe six, seven years ago in Southeast Asia. So it's in that early stage of customer mind share and transformation.
Bernard Leong: I totally agree with you. The essential parts of AI is actually to do with data, and I find that the integration when we work together between data analytics and machine learning, or even what we call generative AI, tends to now integrate quite seamlessly once the customers actually have a pretty good understanding of where their data structures are. So before I'm going to get to the most exciting part of today's conversation, I definitely must ask you this: From your career journey, what lessons can you share with my audience?
Patrick Kelly: I always think in anything, it's personal over professional. So no one's ever going to remember if you work long hours or long weekends when you retire. So personal goals are super important because your personal goals are going to drive your professional goals down the line. I think that's super important. I know you're passionate around NUS and education. It's a huge part of your life as well. I think with a lot of my team, I've got an acquisition team of people newer in their career, and I always say to them, you don't need to be in a rush. Your time is now the right time. Don't be looking at other people and thinking, "I should be like that person." Enjoy the moment. It goes super fast.
Then focus on your profession. If you're an engineer, if you're a data scientist, you have to be expert in what you do. Coding skills need to be practiced. I think there's a lot of noise around AI changing everything - like you won't need software developers anymore. I don't buy that at all. Software development is an art in how you create software and I don't think AI can get there so fast. And definitely in sales. Sales is a profession. You need to work at it. You need to work on your discovery, understand customer problems. Go deep to understand that technical pain. Translate the business objectives and then show how your solution can actually differentiate and help them solve those problems. You only get good at that by practicing. I think delivering results is super important.
We know that from Amazon. So we start with customer obsession and end with deliver results, and we have all the leadership principles in between, but really it's a trailing indicator of success. It shows that the strategy you have, the tactics you execute on, they worked, and then you get the results at the end, and then finally coming back to the whole personal piece. Just be kind and be a nice person. Karma can come around and bite you. So just be a nice person and I think things will happen.
Bernard Leong: Patrick, you're one of the best people that I've worked with. I'm definitely saying this in public because I really enjoyed how you guided me through some of the processes when we were thinking about how to do sales with the ISVs and the mental models behind it.
So let's get to the main subject of the day: Databricks in ASEAN and in the age of data and AI. To start, can you talk about the total market opportunity of AI and data in the Asia Pacific or maybe even Southeast Asia, specifically for business enterprises and why Databricks is poised to capture this market.
Patrick Kelly: Before we even look at the data and AI, I want to look at the cloud market. I think in Southeast Asia, talking to different people like Gartner and IDC, roughly it's about $20 billion, probably about eight to 10% of that IT spend is in the cloud with hyperscalers: AWS [Amazon Web Services], GCP [Google Cloud Platform], Azure [Microsoft], Alibaba Cloud & etc. For specifically data AI workloads, we think it's about a 1 billion market for Southeast Asia. That's excluding the compute and GPUs. Here we're talking about the SaaS [Software as a Service], PaaS [Platform as a Service] piece and the analytics software angle. So it's a huge market and it's growing very fast.
For Databricks globally, we're growing like 70%, and in the [ASEAN] region we're actually growing faster than that. So it's pretty hyper growth. For Databricks globally, we're the fastest growing enterprise software company ever. Last run rate globally 2.4 billion at north of 60% growth rate. And really at Databricks, our mission is to democratize data and AI. Helping data teams solve the world's toughest problems.
You notice there, I mentioned data. Data is core of everything. If you don't have clean data, you cannot do AI. And where Databricks came from, we invented Spark, which a lot of people know, which really revolutionized how data processing worked at scale from the Hadoop days. We pioneered the Lakehouse concept, which means you have your data lake and your data warehouse. You put that together, which then drives down TCO [total cost of ownership] and enables self service analytics. So you get analytics from cloud cost storage from S3. You don't have expensive data warehouse to manage and deploy. And then lastly, I think most importantly now we are talking about data intelligence. So we're talking about how customers can democratize insights with natural language, and build AI on top of their data, their own private enterprise data. They're not giving that data away to anyone. We always say don't give your data to Databricks - your data drives your insights on top of it. That will really differentiate your business.
Bernard Leong: Just to help my audience, TCO means total cost of ownership. And I think one of the major muscles was Databricks' acquisition of Mosaic as well for the AI side. So maybe we should just baseline our audience, given that they are from very different walks of businesses. Can you explain the concept of generative AI and data lake houses and how they help business enterprises to actually achieve their business goals? One of the things that you have really alluded to is the use of the lake house concept that actually drives down this total cost of ownership and the self-service analytics very quickly.
Patrick Kelly: Yeah, so with Lakehouse, traditionally before we had cloud, we had on-premise data systems. So we had a warehouse on-premise, but that was very structured data columns, tables. Think Excel. So you could ask questions, go in and filter and say, "What was my sales report for January?" That's very easy to see. So that was the warehouse world. As the internet became prevalent across the world, we had websites and pictures and images and video and social media and all this unstructured data. And the cloud came around and we said, okay, how do we store this cheaply? Now for Amazon, the first service was Amazon S3 (Simple Storage Service). So that storage became very cheap and we put everything in there. We created a Data Lake. So we had a data lake, and then we had a data warehouse on-prem. Then we thought we could put a data warehouse into the cloud. So data warehouses came into the cloud. Amazon had Redshift, Google had BigQuery, Azure has Synapse. So they built all these data warehouses in the cloud for the structured data. But we still had the same problem. We still had data silos, so unstructured data is in your lake. All your structured data is in your warehouse and trying to find some commonality between both. It still needs a lot of work from the engineering team, from the data analyst team. They're still churning through creating reports and doing a lot of manual effort.
So that was a real problem statement, and also the cost of both. You're paying for storage in your data lake, you're also paying for storage within the data warehouse. And we found there has to be a better way of doing this. So the founders of Databricks from Berkeley wrote an academic paper defining how this architecture should look like, what are the key principles around it, and what does it mean for customers. And that's where the Lakehouse concept started a couple of years ago.
Bernard Leong: How does the strategy of Databricks with that current architecture depend on this Lakehouse concept, and then how does it evolve into the AI side?
Patrick Kelly: So Lakehouse underpins everything. So we have all of your data sitting in low cost cloud storage, and then all your analysts and data scientists can work off the same copy of data. So one copy of data. That's underpinning everything. Now, as you know, Bernard, you were talking about AI back when it was called ML. So now with that single copy of data, you can run a machine learning model. You can train it with different data sets, but it's not a data set sitting in this environment or another data set here with data drift and then the model - you're not double-guessing the model. You're not saying, "Is that a good outcome?' because you're not sure if the data is clean or not. So once you have that then you can add intelligence on top of that Lakehouse architecture. We're calling that data intelligence. So GenAI obviously is the talk of the world when GPT came around, but we are not forgetting classic AI. Those great use cases around predicting customer churn, forecasting demand, optimizing customer experience, those are huge benefits for companies today. GenAI obviously will generate content and you'll have smart digital advisors in financial services, robo advisors, which is fantastic. But I think we should talk about AI in the wider sense of generative, but also classic.
Bernard Leong: It is a very good point. Currently a lot of people just focus on GenAI, but they forgot actually there's a lot of very classical use cases that are already being solved by basic AI. In those use cases that you have mentioned, I'm curious - I know that Databricks is a pretty well known company funded by Andreessen Horowitz, and I have listened to a couple of panels planned by your CEO Ali. What is the current business footprint of Databricks in ASEAN?
Patrick Kelly: Yeah, I'll start with APJ [Asia Pacific and Japan]. We've got really five defined markets across the area. Starting with India and then we've got ASEAN, Greater China, Japan, Korea, and then down to ANZ [Australia and New Zealand]. So in ASEAN, Singapore is our headquarters. So last year when we did our Data+AI World Tour, which was a flagship event in Singapore, we announced that Singapore is now the regional hub for APJ. We had plans to increase Singapore-based workforce working with Singapore EDB [Economic Development Board]. Adding critical roles in field engineering to help customers unlock problems, our professional services, strategy, ops, learning enablement, etc. So we're building out a whole team across all the different functions. Across APJ we've over 800 employees with about 150 who are based in Singapore. Part of that investment is really our commitment to democratization. What that really means is we're going to upskill greater than 10,000 data and AI talent within Singapore. That's a partnership with IMDA, training partners, NTUC Learning Hub, and NUS ICT Academy.
Bernard Leong: You have also done some work with startups as well specifically with GenAI. Can you talk a little bit about that?
Patrick Kelly: Yeah, I am super passionate on the Gen AI part in what I do. So back in middle of last year, we were thinking about our Databricks for Startup program. A lot like the hyperscalers, we invest in them, we give them credits, we give them go-to-market expertise and help them think about how to build a product on Databricks. In ASEAN with Gen AI Fund, when that was set up to invest specifically in GenAI startups, set up by ex-Amazonians as well. You've got Denning, Laura and Kai Yong there. So we know them pretty well and we did a six-city tour last year and it was great. It was great for us because we found 500 new companies that we didn't know about. That weren't in our purview. So that was fantastic. And then we've got a lot of them building with us now. We're investing in them, obviously GenAI Fund is investing equity in them as well. Now, in two weeks on April 10th, we're going to do a startup matching. So the startups we've identified as high priority with enterprises, and we're doing that with EDB in Singapore in with Google on April 10th. So that's going to be a great event to see how these companies grow.
Bernard Leong: I probably also want to highlight that Databricks is multi-cloud. It's definitely not just only working with Amazon Web Services, also Google GCP, and even Microsoft Azure as well. So, how are customers now using Databricks in ASEAN? Can you just share some interesting case studies?
Patrick Kelly: Sure. I think it's super varied. We've got everything from highly regulated industries like FSI and Telco to some of the largest digital native customers like Grab across the region. Grab is a great one because they really worked with Databricks for many years, building customer data platform. As you can imagine, they've got millions of data points coming in across customers, across ride hailing, across food delivery, across all the different signals on their advertising. So how do they manage all those touchpoints and build a customer-centric experience, and then personalize recommendations for their millions of customers? Another example is GetGo in Singapore. You've probably used it before, largest car sharing platform in Singapore that really helped improve customer satisfaction and their fleet utilization. So some key data points - they really accelerated time to insights by 66% for their fleet maintenance. Now they can deliver insights seven times faster, making next business day decision making. They wanted to figure out how customers actually use the car. So analysing booking behavior and refilling patterns, they could actually reduce fuel theft by 50%, which was really impactful for them and their business by minimising misuse of fuel cards and enhancing overall customer trust.
In more regulated environments like GovTech for example, which you work a lot with as well. So they're in charge of the public sector digital transformation using Databricks to empower self-service analytics through our data security and governance, because all the different government agencies need to have access to just the data that they need. They really achieved dashboards that could be created three times faster. They could democratize 50% of data across corporate divisions and actually saved 8,000 labor hours annually, which is massive for a government agency.
Bernard Leong: So they're actually taking the entire Databricks at scale for the entire Singapore government.
Patrick Kelly: Totally. The productivity gains are fantastic because the platform is taking away a lot of that manual tedious work.
Bernard Leong: I hear a lot about multi-regional companies like Grab and then local companies in Singapore. Can you talk about maybe other use cases in other parts of ASEAN? I believe that Siam Commercial Bank is also one of your customers, right?
Patrick Kelly: Yeah. Siam Commercial Bank (SCB) is a great customer of ours. In the FSI space, being a huge bank and having a large data estate. So what we really worked with them on was how to create a seamless and personalised digitised banking experience. But the big thing they wanted to do is create a customer 360, which is AI powered. That really means when you engage with the bank through the website, the mobile app, or in person, all those data points are tied together. So if you access through the web that is all logged and mapped, and you're not duplicating data. A lot of paper is removed from the process. But the real game changer for them was how they can do instant loan approvals. A lot of times today you put in a loan application, you fill in the paperwork, you sign it, you have to scan it, and then you send it off and wait two weeks for it to come back. But now it's a one-click process because SCB have built a profile of you, built a risk profile, know your income, know your spending, and can predict if you qualify for this based on predictive analysis. Out of this, the customer experience was fantastic, but they've seen a twofold increase in approval rates for their digital lending products.
Bernard Leong: So on the AI credit score, I remember in those days when in AWS, we talked about it, but now to see it in action is an interesting outcome. So what are the unexpected insights or challenges that you have learned from your customers in the ASEAN region now?
Patrick Kelly: Customers are the ultimate source of feedback. So one of our principles at Databricks is really obsess over customers. We are founded by academics who build products and are scientists. So when we build the products, we really take in the requirements of the field and what customers are telling us, and then we build that back into the product capabilities and features. So we get feedback on our streaming service, we get feedback on our warehouse, we get feedback on our UI [user interface]. For example, Databricks Assistant, which is our UI which will add coding pieces to tables. A lot of that is feedback from our community and from our customers.
Bernard Leong: That's how you get the feedback. Given now we have proliferation of foundation models and AI agents, how does Databricks think about its position within the market itself?
Patrick Kelly: Yes, the acquisition of Mosaic AI for us was really a game changer in how we think about AI and also the talent that we brought into the company. We have a very deep and talented research team at Mosaic AI who were really solving hard problems at scale. Especially in the science part, deep science of data points. And what that really brought us was thinking deeply around enterprise quality solutions. And when I say enterprise quality, something that has to be really robust on security governance and be able to deliver that at a very low cost to serve. So really our position is that you should maintain full control over your data and your model. You should not give that away to any LLM [Large Language Models] SaaS model or model that is out in the market because we feel that then the data that you're using to train someone else's model if you're in retail industry. Then someone else can use that model and can potentially benefit from the data that you've used to train that model.
So you should really maintain control. The next thing is really production quality at scale. Scale in an enterprise means you need to have that capability, but you also need to manage quality, hallucination, and toxicity. That's super important. A lot of that is refined from your governance practice within the enterprise so you can control that as well. Cost, I have mentioned already - [Databricks] really drive down that cost. Obviously we're a big partner of all the GPU providers, Nvidia as well, so we can help on that as well. And then with native support we've built out the GenAI framework end to end. The idea is that we want to abstract away a lot of technology hopefully so that you don't have to think about RAG and vector databases and embeddings and all those things. A person just needs to say "I've got a business problem, I'm trying to develop an internal knowledge chatbot for my HR." They should be able to just roll that straight away and then the platform will take care of what's underneath. That's the objective framework that we're building.
Bernard Leong: So in the Databricks viewpoint, those layers are being abstracted away from the customer so that they can basically just focus on getting what they need specifically out of their data - could be insights, could be specific kind of analysis. And am I right to say that the large language model you have, the DBRX, is also currently being deployed on the Databricks architecture as well?
Patrick Kelly: Yeah, so DBRX was the best performing model, I think for about 10 days until the next LLAMA [model] came out.
Bernard Leong: It is going to come back again. Every other week, I am getting "this model performs better than the other model," so I'm going to expect a better model from DBRX at some point.
Patrick Kelly: Yeah. Well, the purpose of it was not to show the best performing model, it was to show that you could do it at a cost effective way. Our CEO, Ali talked about it when we launched it, we trained that model from scratch and did a lot of optimisations, especially with the mixture of experts model, especially how you call an expert for coding or an expert for English or an expert for math. We did that all for $10 million, which is pretty amazing at the time.
Bernard Leong: It's pretty impressive for enterprise AI application. I talked to various enterprises who are using Databricks and I think it's pretty interesting that it's what I call a full enterprise-driven AI model, and I think very few foundation models are thinking about that because they're trying to cover all things for all people.
Bernard Leong: So what's the one thing you know about Databricks in Asia that very few do?
Patrick Kelly: I think obviously the technology is there. We present at our events, we show the tech and we work with customers and partners. But I think one thing is we are an extremely diverse company with people from lots of different cultural backgrounds. For my team, I've got obviously Singaporeans, Indian, Thai, Vietnamese, Indonesian, French, Irish - very diverse background of culture, which is fantastic for the team. But then from company experience, we've hired people from all sorts of backgrounds. Of course we've got people from hyperscalers, myself from AWS, people from GCP and Microsoft, but also we've got a lot of people from application vendors like Salesforce, Workday, observability vendors like New Relic. We've got people from system integrators who bring experience of building solutions end to end and implementation plans. I think that diversity really helps us learn from each other and how to serve our customers better. For example, a lot of people may have been selling CDP [customer data platform] solutions and know all those use cases inside out, or someone has come from an SI [system integrator] and implemented really complex on-premise data warehouse migrations and we can learn from that as well. So it really helps us be a very holistic team around delivering the data intelligence platform for customers.
Bernard Leong: So one curious question I really have is thinking about the trends in generative AI and data, what are the trends that are globally or even locally that you have seen becoming important for business applications, thinking about their enterprise AI deployments?
Patrick Kelly: The main thing with GenAI now is of course, agent. Agent is a new buzzword.
Bernard Leong: It is the new buzzword. Agent AI.
Patrick Kelly: We'll call it a buzzword. But we're seeing actual deployments. So MasterCard have just deployed an agent framework for their digital payment assessment after doing like 300 POCs, which they openly talk about. Because it was the last 18 months of testing, seeing what works. Our last report with the Economist found about 85% of GenAI projects were still POCs and didn't go to production. But we're seeing the first stages of actually production now with agents. Because agent is end-to-end, where GenAI at the start was very point-specific. Okay, it's LLM and I'm just doing this piece. You need to have the whole end-to-end framework. Another trend we're seeing, and actually I was speaking to a big GSI [global systems integrator] yesterday about this - a lot of customers want to access all their data under management. So they want it all in one place. They've got data everywhere. It's in the data lake. It's in data warehouse, some is in SAP, some is in Salesforce, some is in different platforms, marketing platforms. They want it all in one place to see how they can actually get better insights. And that's something we're working hard on, which is something called Lake Flow Connect, where we connect into all these systems and pull only the pertinent data to get the answers that the business wants.
Bernard Leong: It is basically the data connector layer very similar to what Anthropic now called Model Context Protocol, MCP or something like that.
Patrick Kelly: Yes. And then super important is governance and security. That is key. That is at the start. It's not a bolt-on. If you don't have your governance set up on your data to figure out who can look at what, what is the lineage, who can access what at what time, and be able to trace that back to your system, especially in enterprise, in a regulated industry, if you have any breach you're going to be in trouble. So we think that governance is such an important part, and it's non-negotiable. We always said that at AWS as well: Security is job zero.
Bernard Leong: Governance is very important. A lot of the senior executives when I train in NUS focus a lot on how to think about guardrails with the data. That is probably one of the key concerns for enterprise customers. A lot of companies who are serving these customers don't really think about how important that element is. But how about locally then in terms of trendlines for generative AI and data?
Patrick Kelly: I think there are different stages of adoption. So I think really advanced big digital native businesses, they built data platforms over many years and they built AI stacks - variation of self-built or built with cloud providers, or built with Databricks as well. So they really help us challenge us on delivering better performance, better price-performance, better outcomes, challenge us on the agent frameworks, which is fantastic. Again, giving us feedback about our product as well. I think on the enterprise side, a lot of them are really fed up with their legacy on-prem warehouse. It's a real cost on the business and getting data out is super laborious. So a lot of them are looking to migrate to the cloud, build their lakehouse on the cloud, which will then drive those insights. I think a key part of how we're helping customers there is we did announce an acquisition of BladesBridge. BladesBridge is a data warehouse analyser tool. So we'll analyse your on-prem data warehouse, look at what that would look like, and then we'll be able to convert the code into cloud. So it's a really great way of moving into a cloud-based lakehouse. And then as we move to the mid-market, SMBs really just want access to their data to drive their business without having a heavy lift. So they're looking for cost effective BI solutions. How can they really get insight to their data, allow self-service analytics, ask questions of it, and there we're providing our GenAI assistant, which lets you ask a natural language question and then enables you to build dashboards. So we call it AI BI. So we're kind of rethinking the world of BI where it's not just dashboards - it should be dashboards and ask questions and you should be able to draw a dashboard and pull in data points that you like.
Bernard Leong: So essentially this AI BI is kind of the fusion of the earlier iteration of business intelligence to be able to get your analytics, but then now having the AI to power the insights on top of that. Is that how I understand it correctly?
Patrick Kelly: Correct. So you could just put in a prompt and say, "give me the sales report for the last 12 months and graph it in a bar chart" and boom away you go. That's all your data, on your data only.
Bernard Leong: I was looking at some of the reports out there from Databricks. You talked about some data relating to ASEAN specifically on the current architecture and also how Asian countries are viewing GenAI as strategically important. Can you elaborate about that or maybe provide more colour about the market?
Patrick Kelly: We did a survey with The Economist globally including Europe and APAC as well. We asked the question, "My organization's current architecture supports the unique demands of AI workloads" and basically 85% said no, we don't have the architecture to support it. Some said it partially does, but it needs lots of modifications so we can still feel a lot of people are still in the early stage and that data point ties back to 85% of GenAI has not gone into production. I think another interesting point is "Does your architecture connect AI applications to your relevant business data?" which is probably even more important for me. And again, it was still about 80% - "We don't have that." Because that business data is all over the place. As we talked about at the start Bernard, without the clean data, you cannot get good AI.
Bernard Leong: So it is the same infrastructure question that hasn't been really addressed.
Patrick Kelly: I would agree, yes.
Bernard Leong: I think more business owners should talk to you then.
Patrick Kelly: Yes.
Bernard Leong: How about the perception of ASEAN countries in terms of the strategic importance of GenAI?
Patrick Kelly: I thought this was super interesting. Along with UK and Japan, ASEAN is up there at around 70-80% saying GenAI is critical to their long-term strategic goals, which is really interesting. Whereas compared to Korea which was at about 65%, which is a little bit surprising for me being Korea, a very high tech country. So it really shows that in ASEAN, a lot of companies are thinking about how to use GenAI to really differentiate the business, grow the business, and actually drive growth for my company. But I think it also drives a lot of growth for the countries as well. It will drive GDP, we'll have startups, we'll have new businesses, we'll have training. So the societal impact is going to be pretty huge.
Bernard Leong: We talked a bit about the large language models from the perspective of which new model comes out. But let me flip the question a little bit - from your perspective, what should businesses be thinking about with Foundation AI models?
Bernard Leong: For example, we talk about Databricks DBRX, and then how do they actually think about the agent AI workflows as well? I think there's still a lot of education that we need to help decision makers to think about these workflows.
Patrick Kelly: There's always going to be a new model, right? We launched DBRX and the whole intent was to show a leading LLM could be trained and tuned for $10 million. Of course then DeepSeek came out, which kind of rocked the world a bit, which is fantastic, showing how you can innovate with technology limitations. And then we have Claude Manus, which is showing an agent working across multiple different platforms, which is super cool as well. But our belief is that you should have a platform with clean data. You should have access to all of these models which we provide. So we've got an API gateway, which you can then decide, okay, I've got my data set here, I want to run a GenAI application. Should I use GPT? Should I use Claude? Should I use DeepSeek? We provide you that capability. We think you should choose the best model fit for that use case.
Bernard Leong: I'm sure that now we are going to see everything go through the applications there. So what would be your advice for business owners in thinking about implementing AI applications in their organisations?
Patrick Kelly: For the AI application, it really depends on who the product team is in your company. I think a lot of companies now, if you think of the digital natives, they have teams. They have a chief product officer and product managers, probably a bit less so in the enterprise where it was probably sitting under an IT team or some application team or customer service team. I think they need to really think about what is the product and what is it doing. And with a product design, then you'll start thinking about what the technology maps underneath.
Bernard Leong: So it sounds like you think there's been a lot of proof of concept over the last two years. How much is actually in production and how can we drive them towards the production workloads?
Patrick Kelly: I think it was 85% from the last time we saw from the Economist Report, and I'm seeing that as well in the field. We did a lot of POCs last year on internal knowledge chatbots, external customer service assistance, content creation for marketing and things like that as well. A lot of that has actually gone into production, which is interesting. Sales augmentation, preparing sales outbound and messaging - a lot of that is done by AI today because I think that's super useful. But I think the real external facing GenAI use cases we haven't seen at scale yet. But I think it's coming there because people were still concerned and customers were concerned about hallucination and giving the wrong information. We had the Air Canada case and the Chevy case giving wrong information from the chatbot to the customer. But I think with guardrails and security and being able to train the model, but train the system what you should say and what you cannot say - again, that's all tied in your enterprise governance and your security posture. I think that's where we can help a lot of companies get to production.
Bernard Leong: So what is the one question that you wish more people would ask you about Databricks?
Patrick Kelly: I'd love them to ask me what is data intelligence?
Bernard Leong: Then what is data intelligence?
Patrick Kelly: Well, people always ask me what the lakehouse is, but we say the lakehouse is the architecture. But with data intelligence, as I said before, it's democratising access to your data to derive intelligence. So it's data and it's your AI. So with data intelligence, you have clean data and you know where it is and you know what you can do about it, but you can derive intelligence from it. So you can ask a question about your customer, you can ask a question about your operations, and a lot of it is very natural language and being able to get answers immediately. And that for us is data intelligence, and for the smallest company in the world to the largest enterprise, we think everyone's going to adopt this approach.
Bernard Leong: That's so concise. I'm going to try to use that tagline too.
Patrick Kelly: Please do. Please market it for me.
Bernard Leong: So my traditional closing question, what does great look like for Databricks in ASEAN from your perspective?
Patrick Kelly: Great question. I think we alluded at the start - I think we're really at the start of something special especially around AI. It's like a once in a lifetime opportunity to change how people work, how people live, how people connect with each other through the concept of data intelligence. It's for the smallest company, for the largest enterprises, for regulated industries, for government - to really unlock a lot of data and solve a lot of really hard problems. We're really here to solve hard problems. And then for our people, it's all about creating a career defining experience. We're the fastest growing software company ever. We're growing north of 60% and we're really investing in our people. We're investing in resources, we're investing in the tech to really deliver on the data intelligence strategy. And then super important for me, especially in Southeast Asia, is building that ecosystem with our partners. You mentioned before, we work with all the cloud providers. We work with system integrators, we work with ISVs, we work with country associations like IMDA, Malaysia Digital, and Vietnam. How do we really drive the ecosystem to upskill? I said 10,000 people in Singapore, but millions of people across Southeast Asia - that's going to be great for us.
Bernard Leong: Wow. That's a very good way to conclude this. So, Patrick, many thanks for coming on the show. And of course, if you are recruiting I strongly recommend anyone to join you because Patrick has been a very fair and great boss to me when I was working with him and taught me a lot. So, in closing, I have two quick questions. Any recommendations that have inspired you recently?
Patrick Kelly: I was thinking about this. I travel around ASEAN a lot, and Singapore, we're super spoiled with how we come in and out of the airport - face recognition and everything else. And usually every other airport I need a visa because I have an Irish passport and I need to apply before and get a stamp and all this kind of stuff. But I got to Jakarta two weeks ago. I was going through my normal queue, and the guy comes over and said, "No, Auto Gate." I'm like, "That can't be true. Auto Gate for me?" He says, "Yeah, go up." I had my visa on my QR code. I scan, facial recognition, and I'm through. It was the fastest time I've ever been through Jakarta Airport in my life. All of that is underpinned by AI - facial recognition, mapping back to the data where I have my QR code mapped to my passport number. So that was super amazing for me and that's an innovation that's just going to grow tourism for Indonesia like crazy.
Bernard Leong: I had the same experience in KLIA Kuala Lumpur as well recently, also the same thing. Auto Gate, just go right through. You don't even need to get a fast track special discount to get through. So my final question - how can my audience find you?
Patrick Kelly: LinkedIn is best. You can get me on LinkedIn or patrick dot kelly at databricks dot com. Welcome any conversation you have, anything you're looking for around Lakehouse data intelligence. Happy to talk.
Bernard Leong: You can subscribe to us everywhere from YouTube to Spotify, it's all in video now. And of course share with us your feedback and give us a five-star rating from any of the podcast platform. Patrick, many thanks for coming on the show. Really enjoyed this conversation and I wish you all the best and definitely we will talk again soon.
Patrick Kelly: Yeah, Bernard, one last plug. Data and AI Summit from Databricks on June 9th to the 12th. If you're interested, please register. It's going to be awesome. We're looking to have more than 20,000 people this year. Great forum to learn, great experiences, all sorts of customers, industries, etc. So we would urge you to join that.
Bernard Leong: Definitely. So, if you're interested, go ahead and take part in this event. Patrick, many thanks.
Patrick Kelly: Thanks, Bernard.
Podcast Information: Bernard Leong (@bernardleong, Linkedin) hosts and produces the show. Proper credits for the intro and end music: "Energetic Sports Drive" and the episode is mixed & edited in both video and audio format by G. Thomas Craig (@gthomascraig, LinkedIn). Here are the links to watch or listen to our podcast.