AI Transformation Panel at H2O World Sydney 2022
AI is enabling a data revolution that will transform businesses, creating new opportunities for growth and innovation. Early adopters across industries uncover new and significant breakthroughs using AI by solving complex medical challenges, improving business agility, and predicting human behavior.
The AI Transformation Panel at H2O World Sydney 2022 brings together business leaders and technology experts to share their knowledge and discuss trends on why AI will continue to play a critical role in driving business transformation and what's the application of leadership in the AI ecosystem and adoption.
Charles Sevior, CTO of Unstructured Data Solutions, Dell Technologies
Jana Kapr, CEO, Billigence Group
Natalie Mead, VP of APJ Sales Engineering, Snowflake
Dr. Olie Fleming, SVP of Strategic Transformation, H2O.ai
Fergal Murphy, Managing Director, Accenture Applied Intelligence
Read the Full Transcript
It's really a pleasure to be able to join the stage and moderate the panel for this exceptional group of industry leaders. And I think today is going to be pulling together a lot of the threads that we've heard over the afternoon as we've had some exceptional stories from CBA kicking us off, and then hearing from other organizations as they've started getting to scale with AI. So, before we get into it, as before, if you've got some questions, you can use Slido but you're very welcome to also ask us questions as we go. So let me start. Leadership, as the day started, is so important in enabling enterprise-wide transformation. So I wanted to start; Natalie, I was going to start with you. Can you help us understand what's really important from a senior sponsorship perspective?
Sure. So I think for us at Snowflake, there's two different areas of sponsorship we look at. There's external sponsors like our customers and our partners, and making sure we have the right folks in the room. So when we're doing large scale engagements, we have to make sure we have the right sponsor, because especially in this region, Snowflake's pretty small, comparatively, to our partners in the US where they've been building the data cloud for 12 years. We've only been around in APJ for five years. So we're still building that traction. So if we don't have the right sponsor at the customer level, or like with Yana and Diligence at the partner level, we're fourth in line. So for us, it really is getting the right sponsor and getting the right level of leadership. In your last session, I was thinking about how to drive change and how to drive use cases.
We have the same discussion as Snowflake. Most people do think of us as a data warehouse. We get that all the time. And so how do you spin that to a data engineering and a training model? And how do we think about data sharing and really building the art of the possible? So if you don't have the right sponsor that can see at that level what we're trying to do with them, you're never going to get any further. And then the second piece of that is internally. So one of the things I love about working with Snowflake is that our leaders are driven on building the right models to drive our success. So everything within Snowflake is cloud. I hate using this American term, but we eat our own dog food, and we really do. So everything is driven in the cloud.
Everything is driven out of our own product. And while our account teams, including myself, don't like this, our finance teams are phenomenal around our training models around consumption, revenue. They know down to the second what our customers are going to be using or the models we expect our customers to use. They know our peaks and our valleys, and they use that information to help us drive the level that we need to be at. And so, internally and externally, we both have that really high level of sponsorship that really keeps us driving. While I'd love to say our finance team is wrong almost all the time they're right. We have a large data science team internally in Snowflake looking at what our customers are doing and looking at cloud consumption and looking at what are our customers doing with compute and what are they doing with sharing. And so they drive a lot of what we do. And sometimes I wish they were wrong, but they're almost never wrong.
Planning for AI
Thank you. So then throughout today we've heard getting AI to scale, there's no silver bullet, right? There's not one thing you need to do. It really is coordinating multiple aspects of the organization to enable something special to happen. So Fergal, I'll ask you, what is required to systematically plan for this? And where do you start?
So for us, I think, and some of this is personal experience, but I think as I interface with many of my colleagues across the globe, I think there were three things for me that are probably the primary features of a successful scaling for AI in an organization. The first one is, your business strategy needs to be your AI strategy. I think this is probably, for me, first and foremost, it's not something off on the side. It's got to be right at the core of your organization. And I think for me the great example of that today for anybody who's here this morning. So you had Matt Coleman and then you had Andrew and you had Dan and Sonel, you had a top to bottom business strategy right through to the AI strategy.
That is probably a fantastic example, I think, of an organization who really understands how AI plays in their business strategy. And that's common for all the major scalers, I think in our experience. The second thing for me in that one is about getting your organization ready. Get your people ready. I think so many organizations sort of charge ahead on AI and they don't prepare the ground with their teams for scale. Because as I said earlier, it's a team sport. It takes a village to raise the AI through the organization. You got to bring people on that journey. And that's I think one of the key things. And having not just the data science community, but also all the business community, they've got to understand that they've got to live with it.
They've got to go on that journey with you, and that's when you truly scale. I think for us if I look back, the effort you require to do that is pretty enormous. And I remember I was speaking to a couple of ex accenture colleagues over lunch outside and we were talking about some projects we did many years ago when everything in AI was custom built. We had a thousand models interlaced and it was all custom built. And that's a different world today, thankfully. But when you're on those journeys, you got to go the distance. And I think just being able to have the organization, for them it was marketing and it was about the marketing team trusting the decisions that were coming out.
Because for them it was for the marketing team to say, okay, if I've got one more dollar to spend on marketing as Telstra, your instinct is pumping into Edwards or put it online. Whereas for their customer base, the intelligence was saying, you're a bit over index on that. What you need to do is hit the billboard or hit the shop front or things like that. Trusting that and then seeing the impact of that being successful was I think where the team then rallied around the results. It's all about results. So that was number two. Get making a team sport. And then the third one for us is, I think just being able to build trust. And I think that comes from the top and it comes all the way through.
I think we have, certainly in Australia, I think we've got more work to do on making sure everybody's sort of fully embraced and on board. I think we look at the speed at which things are moving globally. I think we're all here, I think to a greater or lesser extent, helping to raise the tide in that trust equation for AI. So I think the way we do that is just it's got to be explainable, we got to be responsible. I think that's something every organization has to do for themselves. It happens from within.
Pitfalls Within AI
Thank you. We hear that unfortunately a lot of organizations aren't as successful as they would like to be. So I wonder what are some of the pitfalls of investment and where do things typically go wrong?
We've worked with many companies over the years who've used large amounts of unstructured data in their models. And of course, I was very pleased to hear three mention unstructured data right at the start of the session today. Because really AI is about interpreting unstructured data faster than many humans. If you think of a simple mind model of a casino or something with thousands of cameras and they're all filming what's going on, there's no human sitting there watching those and saying, oh, someone fell over at the cafe or something like that. It's all driven by alerts, interpretation, AI and that's a very simple mental model to say, how does this scale and how does it work? Now, what we've seen in some of these areas, we work with customers in semiconductor manufacturing, autonomous driving, of course financial services, and risk management and so forth.
And it's a matter of knowing how to scale and planning that. And obviously most of these platforms and most of the data science innovation takes place initially in the public cloud because it's a fantastic environment to do that work. It's access to innovative partners like Snowflake and many others. And to be able to bring those tools together and develop those models, but understanding how that will scale in production and what it will cost as that grows. Because nobody runs the cloud for free. It's not a service, right? It's a revenue making operation. It's an IT rental facility. And so the technology team has to be closely connected with the data science team. They've got to work together. And we sort of talk about, well, everyone's known about DevOps for a while and now it's all ML ops and you've got to have your DevOps and your ML ops integrated so that people are aware of the budget exposure over the next 12 months, 36 months. And unfortunately we have seen some quite large companies, I'm thinking of a ride share company in the US, not the first one you think of, that was ending up with something like an $8 million a month spend. And that gets a little bit out of control of course. And it's a matter of planning the data and the technology landscape to make sure that you can see where you have to get to in the production sense.
Talent is something that's come up a few times today, and I think like many other companies, the gap between demand and supply doesn't seem to be narrowing anytime soon. So Yana, in this market, in this environment, how do you find talent and more importantly, how do you retain and excite it?
I think it was mentioned a couple of times here in the previous presentations from Matt or from Andrew that it's hard. We at Billigence really believe in attracting people from universities, people who are passionate about data science, the cloud tools. Of course, we are partners with H2O and with Snowflake. So we try to offer a really exciting opportunity to learn new things and really look at people who are from university, fresh, really trying to do something great with their lives and and attract those talents. We run different things across the world. For example, in Prague, where I'm originally from, we run an academy for girls, which is called <inaudible>. And this is quite exciting because my Prague operation now has 60% girls and 40% guys, which is a great example to the rest of the world.
That's why the diversity, looking for talent is the way to go. But of course in this talent, I would say there's another opportunity. And we spoke about it as well. Andrew mentioned it about productivity and also I could see opportunities for data engineers to be really trying to have feature engineers. Yeah, we spoke about, you know commercial data scientists, which potentially could actually translate the problem from the sea level to those data engineers which can pick up feature engineering, which is critical, and I think that's also opportunity to give to the people we already have on our teams, those data engineering guys who can pick up feature engineering because that is what will make the NLops successful. And those opportunities and how to retain them, really make the work interesting.
We try to make sure that our team works on different projects. We go and do some good type of projects. We've done really great work for AIS here in Australia also using Snowflake now, hopefully we introduce them to H2O as well. And really looking at maybe optimizing how much these guys are training and what's the right level of not having injured athletes, overtrained athletes, how we identify the future Olympian using all these. So lots of that work will be done pro bono because I'm finding all also these types of interesting opportunities because number of those guys want to give back. So that right balance, really keeping it interesting, keeping the opportunities to do different things and really listening to them.
What's the most challenging role to recruit for at the moment? On the science side, the engineering side?
National data scientist, because there's not many people who could really articulate well and simplify. Because at the end of the day, it's all about outcomes for the business, for that sea level stakeholder. In the past we were doing like a past reporting, we were trying to give them beautiful dashboards, you know colorful and simpler, et cetera. Now we are basically trying to give them future trends and it's critical decision making information. And as Matt said early in the session he said, you need some subject matter expertise as well. It's got feel is why, to be able to have someone who could actually translate what the leadership wants to the low level teams is so critical. And I think this is still in demand. Not many of those, those unicorns exist.
Quite. Thank you. And for those in organizations where machine learning is not like their primary focus, where have you seen success for organizations upskilling the overall business?
It's a good question actually. Yana talked about it from the perspective of hiring smart, and then also finding the right partners to work with. I'll just tell you Snowflake, hiring a team that understands training models or moving from a traditional model of learning in SQL versus Python was really hard for us. We've got a large percentage of our team that came from that data warehouse background. And our customers were like that. And that's really where we dove in to start. And as we started to look at data engineering and different models and learning and with partners like Villigance and looking at other environments, we said we have to shift. And so the same thing that we found is with our smaller organizations, we've helped them shift. We said, okay, looking for a commercial data scientist is really hard.
Let's figure out another way to teach them to fish quickly. And so what we found from the ground up is finding the right partner that can start you down that journey and then hiring somebody that can see your business outcomes. Because what I've noticed in the two plus years I've been at Snowflake, when we hired pure data scientists that only wanted to build training models that didn't understand the outcomes of our customers, we ended up failing. So we had to say, okay, how do we look at this a little differently? We looked at a lot of our SI partners had very, very smart data scientists that were looking at business outcomes. And that's what we did. And then what we would say to our customers in the end, we said, hey, look at hiring in this direction. I like to call them tinkerers, look at people that like to tinker and look at a training model to figure out what business problem does this solve. So what we've said is find the right partner and then once you find the right partner, I don't want to say hire smartly, but just be creative around how you hire.
Upskilling ML and Data
Thank you. And then if I step deeper into the business, what are effective ways you've seen the business get upskilled with the necessary ML and data side of things?
At Accenture we did some research sort of pre and post pandemic and we'd seen before the pandemic that about 12% of organizations had really got AI ML, like they'd really got there, they were practicing. And about another 25% were in the game a little bit back in 2019, and there were like 63%, whatever the balance is, were just nowhere. And that top 12% were generating 50% of the profits around the place. Like it was material what impact that AI ML was having. And then we went back and reconstituted, okay, well let's look now post pandemic, what's happened, what's changed? And we surveyed, I think the first one was 1200 clients globally. This is 2000 clients globally. What was found is that more and more people are moving into that top bracket, so that 12% is going to be about 27% by 2024.
So there's more people getting it. And they're all on that journey and all competing for that top talent, and nobody wants to be left behind. And I think what we did then was to look at, okay, what does it take to get there? And I think the thing we discovered, I think interestingly is whilst AI ML is a very scientific kind of academic background, AI maturity is actually an art form in an organization. It's actually not a scientific process. It's actually a bit of an art form. And what does it mean to get there? I think it absolutely starts from the top. I think it's got to be laid right from the top. That's probably the factor number one. I think the second thing is probably the culture you create around that in how you bring it.
And I think some things are cultural. We work with Snowflake and I think some of the commentary, I think Snowflake is part of that ability to move quickly and nimble. I think a lot of what we had in the past was mindset oriented. The restrictiveness was how we approach things, not necessarily the technology. I think we've got platform players, like H2O and Snowflake are starting to help us now get that mindset and culture going. Yes, we can do this, we can get it done, we can get it done quickly. So that's been a huge shift. And then you bring it down, there has to be investment. I think maturity doesn't happen unless you make the investments.
And these progressive organizations they're investing at all those levels. They're investing in their leadership, they're investing in good platforms to go on that journey with. It requires that investment. And I think because people need the right tools to do their jobs properly. And I think you can't hold back on or skimp on that investment and expect to get there. So that then comes down to the upskilling process. And I think one of the great examples was, again, back to bringing people on the journey. I think the video shown earlier today with Dan with his bake off internally, I think that was a fantastic example of how you bring something like enablement for the organization to life. I think that really resonated with me.
So I think these are right the way through, these are the dot points and I think we've happy to share that paper and all those steps to get there. But I think for us, it is really that art form of how to get to maturity and I think thankfully we've gone from a set of principles around AI ML a few years ago to now having real evidence of how to put them into practice. So yeah, happy to share that with the audience.
Productionalized Model at Scale
Great, thank you. I'm going to step into the technology side of things. So we've heard it's difficult, a lot of organizations kind of get stuck in the POC world rather than operationalized at scale, impact being realized what's required in order to go from data to a productionized model at scale?
Yeah. As we've been talking in the panel, obviously leadership and executive sponsorship is critical because the company has to move forward. And we heard from Matt this morning, and the focus that ComBank has on being data driven and being an AI first organization is fantastic. From a technology point of view as I mentioned before, it's about scale. And at Dell Technologies we're really focused on infrastructure and having an open and very scalable infrastructure solution. Whether you're talking about storage whether you're talking about data protection, data management, compute there's a lot of technology resources there, and that can exist on-premise, in the cloud, in a colo, we manage all of those resources. One of the things that we announced back in May at Dell Technologies World was a partnership with Snowflake.
And that was reaffirmed at Snowflake Summit also in Las Vegas in June. Basically what we are doing there is providing the ability to have very large external tables that are connected to the Snowflake data environment. So fundamentally, to achieve an outcome in analytics and to leverage AI, you need a data lake or a data lake house. Data lake house is the term of the month because that combines both structured and unstructured data. And snowflake's a great partner for data Lakehouse solutions, H2O bolts onto that. It's a really nice synergy, and Dell fits in in terms of that very secure scale out protected storage. Because as we all know some of this data might hold personal and private information, and it has to be carefully managed and staged.
So this is the way that a POC can start in the cloud with a theory, it can develop and build into a working model that's producing sensible results. And then when you start to load serious amounts of organizational data, again, you want to start relatively small before you invest in infrastructure. There was a question, I think one of the last questions to focal in the previous session about the big three year project and all that sort of thing that under delivers and runs over budget. Nobody wants that. and Dell works closely with their customers to also follow that scale out principle. So the idea that infrastructure can start on a pretty small footprint, it can scale simply by adding nodes, adding nodes of storage, adding nodes of compute to be able to grow as the organization embraces that information flow.
And to work with a partner like Snowflake, where at the same time as you are scaling your external tables, you are building your model and you are taking advantage of the Snowflake analytics platform for that. Likewise with H2O. So the key term I want to leave you with is scale out and design up front. Understand how the POC will build, what are the success criteria to move to the next stage. And then you've got your working prototype that's functioning within the organizational data flow. And then how will that scale?
Thank you. And then Yana, I'd love to dig into the build versus by piece extend on that. So historically, I've seen a lot of data science teams and technical teams wanting to build things from scratch, but increasingly there's amazing tooling out there that can accelerate this space. So it shifts the lens to organizations perhaps needing to become expert procurers in this space. Do you observe that in your work?
Absolutely. And you are correct. Three years ago, if you would ask me the same question, I would probably answer differently. And today, a hundred percent, there are very advanced AI tools and H2O is one of them to really go for your life with AI and the scalability which you can achieve as we achieve with moving data to the cloud, having data like houses and snowflake we can achieve to scale basically AI to lots of I guess users and users. and I think there's probably two main benefits, why I would say eight to 20 rule now, buy rather than built is because it's just all about time to market and having it accessible by more people and scaling the capabilities to the really those right audiences. Because if you think about it we are here in data analytics to give the decision making information to those end users, and now it's becoming given customers. We probably will have virtual banker in our app in the future and it will be driven by AI. And the only way to achieve it is to go through that, having that scalability, having those models well trained. And there is no way that even as any consultancy would have or any internal like a scale of staff that you would be able to achieve it without those smart tools, which we now have access to.
Decluttering Complex Systems
Thank you. I see there's a question up on Slido. My eyesight's not brilliant, but I'm going to give it a crack. Big orgs are complex and have legacy of traditional tools and systems. Some of those now offer AI solutions too. How do we declutter, remove bureaucracy and move ahead? Anyone fancy it?
I think probably the simplest answer is back to leadership. If you have a great top-down leadership, you could declutter, and I've seen it in some insurance companies, one of them being our clients who is retiring the whole legacy and fully going digital, fully going cloud, and they're going for it. So it's top down. It needs to be exactly what CBA has demonstrated today with all the leaders being committed behind it. Otherwise it'll never happen.
Thank you. So we are coming towards the end of our panel discussion. I wonder if there are any other questions that folk in the audience have, you can of course put it on Slido, or if you would like to use your mouth, we can get a mic to you.
I think we scared them.
Taking AI Abroad
Audience Member 1:
Can you repeat your question, I couldn't hear you completely.
Audience Member 1:
This is much better. CPA is part of the 12% of companies that adopted AI successfully. Do you see a good benchmark abroad, outside of Australia?
I think that was my stat. I think yes. I think what we're seeing is for many of these industries, I think that AI is actually becoming platform businesses. I think it's probably one of the key things we're seeing. And I think the nature of sort of how you can create IP inside an organization like a bank and then actually monetize that externally, I think that's the future. The boundaries are breaking down. We're no longer geographically and physically constrained. So I think what we're seeing amongst that top performing set is their ambitions no longer constrained by their geographies. And I think, yes, you've got the digital native businesses who are using AI, they'll be some percentage of that 12%, but a material part of that 12% are going to be global organizations in banking, in telco, in natural resources who are out there putting a AI ML to work. And that's going to be the difference for them. I'm happy to share examples more broadly, but I think you'll see, and many of these companies are on the public record to demonstrate what they're doing. So I think you'll see more and more of that.
To add to that, in finance and banking, banking especially, the companies that I worked with in the US, they were driven around that. And we talked about being the data mesh. A lot of them are actually putting their own data scientists in their.. like they have a data scientist and data engineer for their group. So customer might, customer 360, the marketing team, the reason they're doing that is so they can accelerate. And it really allows them that agility to change their model as need be. When you're relying on one team under IT, it's super hard to get those resources and figure out how you can shift quickly. And so what we're seeing, especially in the larger banking institutions, they're just actually going to this mesh strategy because they can move faster and then we're seeing it in other regions or in other industries as well, especially in banking. It's just critical.
I've got time for one last quick question if there's... Yep. Please go for it.
Desirable Curriculum for Undergraduates
Audience Member 2:
You mentioned on talent, looking at universities. So kind of to the whole panel is, what do you think would be most beneficial for industry going forward? What type of curriculum would you be looking at from these future undergrads and graduates and postgraduates coming out of the university system?
There is already, Sydney Uni, for example, is teaching classes like econometrics, econometrics, I think data, science. I think anyone who wants to enter this world understanding the statistical modeling, math degrees are absolutely vital in this space. And they don't need to call that's all learnable. And being able to present and simplify. I think those are probably the key skills, what we are looking for. And any engineering degrees if you have really good electrical engineering, mechanical engineer, you probably even touch some simulation models in the past. Absolutely. These are the skills, and I think lots of those students are already thinking that way, but the combination of being able to communicate as well as be able to understand statistics is probably the best combo we are looking for.
I'll throw another one in there, perhaps more on the technology side, but there's a lot of really exciting stuff happening in the world of digital twins, which moves into the metaverse, which is all sort of future. But the digital twin could be a physical digital twin. It could be a smart city social digital twin. It could be a financial model, currency, exchange rates, digital twin, and yeah, it's all based on statistics and mathematics, but understanding how to represent that and visualize it is critical.
Can I just pop in on that one? I think for me I think there's this all or nothing proposition with sort of learning and data science. You got to be a data scientist or something else. I've got a 15 year old daughter and she's …
Does she want to be a TikToker? Cause that's all mine wants to be.
I'm doing my best, right? And a case of saying those she wants me sort of in medical or something. I'm saying it doesn't need to be in our proposition. I think we have to make this, it's kind of a bit of a personal fashion, we've got to make this an end proposition. It's got to be in all those industry disciplines to say, you need to take advantage of this. It's not going to be just the data scientist that needs to be more democratized than that. And I think for me, the role we have to play is to make this accessible to more people.
And probably that's something subject matter expertise will never go away. That's why you need to have people with that specialist, otherwise you can't check either that the model is doing the right thing.
Absolutely. Thank you. Now there's all this evidence out there and examples of organizations that haven't managed it, but some that have. So I think the opportunity for new organizations going on this journey to learn so they don't repeat mistakes that have been made in the past, hopefully paves the way through to experimentation at scale and value creation through AI. So I'm personally excited about the future. So I'd like to thank all of you for your attention. Thank you very much. I'd like to do a massive thanks to this panel, so maybe a round of applause. And the last group of people I'd like to thank are all the team here who have helped pull today together, and I've been running around making all these sessions a massive success.