AI in Mining: From Pilots to Productivity

Thank you, Tim, and welcome. It's a very niche audience we have this morning. Hopefully, a few more join us as, as we get going, but we'll, we'll forge on nonetheless. So again, thank you, Tim. Thank you, Jessica, for making this all possible. Uh, we really look forward to talking about this topic, Pilots to Productivity.

Why have we called it that? I, I think we're living in a remarkable time, I think, which we're all experiencing, uh, on what A-- what's possible with AI. Many of us experiencing it in personal productivity. But the one area we're not seeing it very much right now is in the bottom line of mining companies, and that's what we wanted to explore today and explore why is that and how do we shift that? How do we accelerate that?

I'll be joined, uh, in fact, maybe come up to the stage already now, uh, by three colleagues. Ferran Pujol, a, uh, partner from our office in Santiago in Chile, where he leads a lot of our work in AI and mining. Please take a seat anywhere. Thank you. We have, uh, Akilan Kapilan from Microsoft- Hello. -where Akilan works across energy, uh, and resources sectors, and a lot of his work is translating frontier technologies into bottom line impact f-- uh, with companies.

A, a few quotes from Satya Nadella, which, you know, when we're thinking about Microsoft being here on the stage really resonate with us. One is, uh, "Change the work and the workflow with the technology," and this concept of a, a hill climber machine, and these will be concepts that we're going to want to explore with you, Akilan. Also, at the far end is, uh, Ravi Malladi, uh, advisor to Freeport. Uh, before that, uh, Ravi was actually also with McKinsey for a period of time. Before that, he was with Barrick Gold and GE. So Ravi's got a long history at the intersection of, of, uh, technology, AI, and heavy industrial companies. Thank you, Richard.

We're gonna share a few insights from a report that we've published called The State of AI. There's copies of the report printed out at the back of the room, uh, which you're of course welcome to take, and we're happy to talk about the report here or one-on-one, uh, before or after, uh, the, the session. You'll also see at the back of the room there's, uh, copies of two different books, uh, the Rewired first edition and the Rewired second edition. Uh, both of them are, are books written by McKinsey, uh, around how do we actually help get, uh, digital and analytics to really transform large corporations. It's not specifically about mining, but, uh, one of the case studies in both of the books is a mining company, Freeport. So please do pick up a copy and feel free to talk to us about that as well.

So as we start going into this, uh, first point is that AI is scaling. The data we're looking at the screen here is around AI being used in at least one corporate function i-in large enterprises. And we can see over the last five, six, seven years, that really is starting to scale up. This is all industries. If we then look at a bit more detail, you can see all the industries across the top, functions, uh, down the rows on the left. We can see for sure there's some, some hotspots or some blue spots, I should say.

Unfortunately, you can see just, uh, halfway across or so is energy and materials, and within that lives our mining sector. And you can see that that is low compared to everyone else. In fact, it's the second lowest in terms of adoption of AI. The lowest is another sector which we depend on heavily, which is engineering and construction, which is where our EPCMs and the like all live. So this is a real challenge for us.

If you look at where, uh, functionally, uh, AI is really being adopted, again, there's the usual suspects. You look at IT, you look at software engineering, you look at service operations, heavy adoption. But where's the money actually made in heavy industry? It's the bottom row, manufacturing. And unfortunately, when we look down in that cell, we, uh, are more at the seven percent adoption type number. Even more so, and Akilan was reminding us, me of this just before this session, this definition we have here is energy and materials, and if you compare the rate of adoption of AI in the power sector compared to, uh, mining, it is far higher in the power sector. So mining is actually gonna be even lower than this.

So let's shift gears, uh, and think about where actually are we starting to see progress in mining, and then we'll open up to the panel and talk about how do we actually achieve an unlock in this sector. I'm gonna hand over to my colleague, Ferran Pujol, take us through that.

Thank you, Richard. So this session's name is AI in Mining from Pilots to Productivity, so let's define what's pilots and what's productivity. So pilots is adopting tools, uh, better tools. And productivity is very well put by Satya Nadella, is, is change the work, the workflow with the technology. So changing fundamentally the way we work.

Um, so what I'll show you now is a, a couple of pages based on, on research that we are doing. There's two pieces. One, we're following nineteen of the major mining companies, and we're monitoring how they adopt AI and how the, the AI work impacts the P&L. And the second piece is Rewired. So it's basically, um, we did a research across industries to see what, what's the difference between the companies that stay at the pilot stage and the companies that get to the productivity stage.

So those are the domain-- I mean, some of the domains that you see in mining, and it's plotted basically how they are impacted by AI in three dimensions. So left to right, you have the maturity of AI in that domain. That means how long has the AI been around in that domain and how certain we are that, um, AI will make a change in P&L in that domain. So it's basically on the left you have the pilots and on the right you have productivity. Bottom to top, you have the intensity of innovation. So up right, you have higher pace of innovation. And then the size of the bubble is the value at stake with AI in that particular domain.

So if we look at processing, um, it's a domain that is very mature. So actually ten of the nineteen companies claim that have impact in P&L from AI in processing. And how do they do it? They have models that trace the ore characteristics from the block model to the processing plant. Um, they have models that define the optimal blending and models that define the optimal set points. But there's also innovation in processing, so the most advanced companies are moving gradually towards closed loop. So that means in that particular, uh, domain that they are connecting the AI to the control systems, and the operator goes from defining the set points to monitoring the, the models and the production.

So if we move to procurement, I mean, in procurement, we're seeing a Cambrian explosion of agents. So you have many agents that support the, the buyers in negotiation and in finding levers to reduce cost. So a couple of examples. For example, in Minsur they have agents that send directly emails to smaller, uh, to, to suppliers of smaller contracts to negotiate directly. Or at SQM they use agents to improve the productivity of the, of the suppliers.

So moving to... Oops. To the next one, to maintenance. Um, in maintenance, actually, most people talk about predictive maintenance, but what we are seeing in the data from these nineteen companies is that there is value from predictive maintenance, in particular in the mining space, because you have, I mean, lot of the sa-- lots of the same piece of equipment, right? Uh, and in the plant it's different. Um, and they get that impact with these OEMs. But actually, most... I, I mean, even though predictive maintenance is the, is the case that is most, most talked about, the impact is mainly in maintenance planning and shutdown optimization.

So the next one is mine planning and operations. So that's an interesting one because there's, like many tools. But at least from these nineteen companies, they're only seeing impact from one of the tools, which is the, the digital twin. So lots of innovation, lots of things being tried, but at least from these nineteen companies, they are seeing impact from, uh, digital twins.

So now moving to capital projects. So, I mean, capital projects, as you know, I mean, have a very long cycle. They are made out of many sub-processes like permitting, engineering, procurement, and so on. So we are seeing also a lot of development in agents, in agents in that case, in particular in decision-making. So you know that if you make a decision late in a capital project, it could cost you five million dollars. Um, so some agents help you target what's the decision that needs to be made today. There's also other agents that help with productivity of vendors and tracking of progress.

And now moving to exploration. So in exploration, there's a lot of innovation in, I mean, in, in trying to combine three areas. So it's basically sensors, a lot of innovation in sensors, data sources like pu-public or private, and models trying to make sense of these, uh, of these innovations to, to improve discovery's probability.

So this is the what, and what we're seeing is that, um, companies, I mean, or the, the companies that adopt AI for two, three years, they claim that they can get to, uh, ten to fifteen percent of it, uh, impact. And but something is changing. So that's out of the nineteen companies, what you see here is how many of those claim they're having impact in the P&L every quarter. And basically in the last two, three years, you see that the impact has, I mean, o- from one to six companies claim that they had impact in their P&L. But what changed is that this quarter, actually fifteen of the nineteen companies claim that they're having impact from AI in the P&L. So I mean, it's the first data point. We need to confirm it, that's a trend. But it seems that something is changing and really the efforts of these couple of years start to materialize in, in P&L.

Now talking about the, the how. So what do these companies do different to, to move from pilots to productivity? So six things that they do differently. The first one is that they define the roadmap based on P&L. So it's not fifty initiatives, it's like focused initiatives on what really move the needle, and what really move the needle, we've seen it in the, in the previous page. Then in talent, I mean, they follow the seventy-seventy-seventy rule. So it's seven-seventy percent of the tech talent is in-house, seventy per-seventy percent are builders versus coordinators or, or managers, and seventy percent are senior. So it's basically a small group of senior people that are builders.

In the operating model, I think the, the main thing here is closing the gap between the business or operations and IT and technology. They achieve that by co-location, by empowering the teams, and by funding, funding them over time. In technology, AI, and data, uh, they see data as a strategic asset and they enrich the data. And an example of that is, for example, when you have operational data, you also capture the expert reason of why this phenomena happened. And on data, there's a concept of, uh, that the model needs to earn the trust. So basically, you start with a narrow scope, and you start to broaden the scope when the-- when you can trust the model.

And the last, the last piece is adoption and scaling. So, um, here the companies that move to productivity think about scaling from day one. Um, they change the workflows, as we saw before, and they define one owner that is the owner of a P&L and is accountable for the outcome and for the money.

So in the operating model, so the people in this room also we've seen that these companies give specific roles for each of the CXOs. Uh, so the CEO, as we said, they-- he-- they name an owner that is accountable for the outcome, that has the budget line, and that is responsible from pilot to sustained value. The COO is responsible for scaling, so basically when you have one domain that is proven in one site, scaling it to multiple sites. The CFO is responsible for understanding the value of AI isolated from other effects of other levers. And the CTO aligns the, the specific milestones with the, with the vendors and platform partners.

So what I show you about the, the what, the, the where is the impact and what they do different and, and how is based on the research, as, as Richard was saying, of the, um, State of AI report that is in the back of the, the room, and Rewired and the books are also on the, on the back of the room. So with that, I'll hand it to, to Richard.

Super. So if I summarize, I think we've heard that, you know, AI is really starting to get traction in other sectors, including in energy, in the energy and material sector. Um, unfortunately, mining is lagging, construction is lagging, so we've got some ground to make up. Ferran has taken us through a good number of, of, let me call them individual use cases, but we know that by themselves individual use cases don't create significant bottom-line impact. That's right.

So [clears throat] let's shift gears to the, to the panel and, and, uh, please also think about questions that you would like to put to the panel. There's a QR code on the tables, and I think it's on the screen. Maybe not quite yet, but it's on the tables. Uh, and you can log your questions there, or there'll be microphones passed around later too.

[clears throat] Uh, Ravi, let me start with you. Um, if, if we think about the reality on the ground, we've heard about how things work technically, but can you... Let's start off with a positive story. Like, where have we got good traction in the mining space with AI actually making significant impact?

Yeah. No, great question. I think Ferran made a great reference to the processing aspect on the milling side, and I think that's a great example. But just to shift gears, I think from my experience in the recent times, maintenance and reliability is one of the critical applications where we are able to move the needle. Um, there are very, very good examples where the shift is from calendar-based reactive maintenance, managing the workload, work exposed, to really condition-based maintenance. And I think from a sensoring standpoint, we really are very, very advanced now with telemetry, with, uh, with, with RCM, with a variety of inputs ins-- on, on-the-ground inspections. We are able to leverage the digital twin AI technology in order to really move the North Star KPIs, Richard.

I think end of the day, what matters is two things. You can under-maintain something and then really have, uh, a lot of value leakage due to increased unscheduled downtime hours on an asset class. On the other side, you can over-maintain something and spend way too much. So I think it's really the optimal between, how do I maintain it at the appropriate level so my cost of maintainable hour also stays within limits? So I think that is a great application, and there are some critical trends that we are seeing reactive to condition-based maintenance and a frontline behavior, which is very, very important.

What changes within the operating model? A clear process adherence. Yeah. Uh, I would say that in a nutshell, it is a combination of reimagining the process. Technology is of no value unless you are able to actually move the needle on process adherence. You can take an oil sample. If the oil sample doesn't make it to the lab and the data doesn't get into your system, which is your foundational system of records, I'm afraid AI is not able to move the needle very much. So I think it's a combination of people, process, technology. We keep talking about it- Yeah ... and I think it's absolutely true in this case too. But within the context of maintenance, we are able to see significant gains.

So that's a great story to start off with, uh, an area where there's adequate sensoring, a clear need- That's right ... and, and you're starting to see progress. Um, Akilan, let me turn to you. Uh, you know, we've referred to that, uh, Satya Nadella quote about change the work, change the workflow with the technology. If we think about, if we unpack that a bit, uh, what actually needs to change, uh, within a operating environment, um, decision rights, the interaction between AI and, and people? What needs to change in that whole system to really try achieve this kind of impact?

Yeah. Thank you, Richard. Great question again in how AI is proliferating into the industry today. Um, we have been using... A lot of the companies have been using AI to get reports, to get dashboards, to get point-in-time information, which are rather siloed today. And what we're asking is, or what we're seeing is, that you have to redesign, um, the domain workflow in a way that makes sense so they can pull the data if the sensor information is not available. Uh, take, for example, asset optimization or predictive maintenance. You can get information as to when it's going to fail, but you need more information about, you know, what are the parts available? Is there a workperson available? Do we have supply chain? Do we have information to make it happen? And that can be done by redesigning the workflow and connecting the different dots together.

And the other area that we're also seeing is process optimization, right? When you have an ore body, you have multiple different types of ores that is coming in for your process optimization, you have to be able to identify what is the right mixture, what is the energy capability that you need. And then if you redesign the workflow so that you're connecting the dots, the intellectual information, the cognitive information, the experience or the intelligence that is there, now you have a redesigned workflow that will suggest options to you rather than just giving you information. But more importantly, you need the human in the loop- Mm ... to make that judgment and make that decision rather than having an, an autonomous and an automated process that will do.

We want suggestions and recommendations by looking at the different components, by tying the data together. It's not only applicable in your industry processes, also in your HR functions or finance or customer service, and we've seen that. Being able to bring information together and to make suggestions and recommendations is where we're seeing that, you know, if you redesign the workflow, then there is a chance to make it happen.

And how will you do this, right? There need to be domain leaders in place that will own the process and will make the recommendations. You need to set guardrails in place, and you need to make sure that the domain leader owns that business process and is able to make recommendations as well. Um, decision points are important. Uh, you have to set the envelope, and you have to set guidance so that the metallurgist is able to make the choice in the processing and say, "Okay, this is an option we can go with," and then we are able to redesign the workflow. So safety is also another area that we have looked at to being able to help with the risk capabilities. So in order for transformation to happen, you need to redesign the workflow and always have human in the loop, but human is higher in the loop and not just at the tactical level.

Excellent. Thanks. Thanks, Akilan. That's... I like how you're describing domain leaders, how you're describing what the role of human in the loop is. I, I think that's, um, there's clearly a lot of work to do too, though, 'cause when we think about how in years gone past and still ongoing now- Yeah ... uh, when you think about a workflow redesign in ERP context, it, it's a big lift to actually make it happen. Yeah.

Let me turn to you, Ferran, on the same theme about the, the lift, uh, to make this happen. You spoke a bit about individual ideas, uh, um, that can be pursued. You also touched on the concept of domain transformations. Can you explain to us more what do you mean by a domain transformation, and what's required to make that happen?

Yeah. So a pile of AI initiatives is not a transformation. So the, the test is whether the domain runs differently, not if we have, uh, better tools. So a pilot, um, makes better recommendations, but then a domain transformation changes who makes transformations and in which way these transformation, the, these, uh, decisions are made. So for example, if you take processing, the example that, that you were giving, um, a pilot would be a model that sends recommendations on the set points in the processing plant. And, uh, transformation would be a closed loop system where the operator goes from defining the set points to now supervising the model, the operations, and also understanding the model and improving the model.

So the concept is the never just tech, so that you start the roadmap with a P&L- Yeah ... with a few focused areas, um, that you, uh, make sure that operations and technology work together, that the data is usable, and that you de- uh, design adoption from day one. So in the end, that's a test. Am I adding tools, or am I changing fundamentally the way we work?

Yeah. And you, you, you touch on P&L there, which is such an important point. Ultimately, if you're not seeing the impact either in the bottom line or in the main metrics you're pursuing, it's, it's, uh, it, it's not gonna achieve what we want. Mm. And just a reminder, if you want to be starting to put... We've got a few questions trickling in, but if you wanna add more questions, please, uh, use the QR code, and you can just add your questions there as well.

Um, let's talk about scaling because, uh, y- the overall title of what we're talking about here is, you know, from pilot, uh, onwards. And- Ravi, let me turn to you on this. What are the main factors... What needs to happen to go from pilots in a particular area to scaling across a, an enterprise?

Yeah, enterprise. No, I think it's a [clears throat] great question, Richard. And I think so the many obvious no regret elements is one way we can frame it, but there are certain trends that I wanna talk about as well.

I think co-creating with the frontline is one of the absolute musts when you really wanna move from pilots to scale, number one. Number two, is it causing a cognitive load? For example, on the mill, is it an additional screen that the operators have to look at or does it very seamlessly integrate into the SCADA and the HMI that they are used to? I think those are all the absolute box checking type elements that we will have really no argument that I think that's an important thing.

Stepping back, I think I have noticed a few things. When you... Especially in maintenance, it's very simple to think about, "Am I solving a siloed problem with AI versus am I solving an end-to-end workflow problem?" Mm. So maintenance is condition identification, planning, scheduling, execution. So one needs to think through a solution across the value chain, and I think they have much better chance of moving the needle versus a very, very siloed solution. That's a very, very important element.

And there are certain foundational aspects that I referred to earlier. Data hygiene are the... You know, end of the day, we rely on a system of records. So is the hygiene of the data and the processes that support that hygiene, uh, is there discipline within the organization to do that? And I think that is absolutely one of the critical foundational aspects that will either make or break the transition between a pilot and a, and a more sustainable, scalable solution. There are many others, but I think these are some of the things that are very, very important. I think stay away from solving a siloed issue. Mm. Solve-- Look at the end-to-end workflow. Mm-hmm. Process adherence, very, very important.

And I wanna come back to pro- people process- Yeah ... and then technology backing that- Yeah ... as the foundation, uh, you know, some of the belief that we need to have as leaders when we think about transformation. And I liked the point you're making there, Ravi, about involvement of the front line. That's right. And ultimately, that's, uh, the group of people that often know the most about how- Mm-hmm ... a particular process runs and about data hygiene.

Um, uh, Akilan, let me turn to you. Uh, your, your, your, your colleague, Satya, also says, uh, the most important thing is actually the, the diffusion of AI and doing it fast. Yeah. Yeah. Talk to us about the role from a Microsoft perspective of what, what's the role of the, the enterprise or the center versus, you know, often mining companies are set up with some sort of central group and a s- technology group, and then individual mines, which often have a fair amount of autonomy. Talk to us about how the, the... If you think about how to get the rate of diffusion of AI accelerated, what's the roles of enterprise versus site versus tech?

Great question. Um, I like what Ravi said about data hygiene and data quality. Hallucination is real. If you don't ground it, you'll get all kinds of answers, but every time you have to check what AI is telling you to do. So that's an area. Um, um, in a-- So the way you have to scale is each site has a different proprietary information. Each asset is different. The way the mining operations are different, the ore body characteristics are different for each of the site. Um, you have to be able to optimize and transform that into a, a repeatable process so that you are able to use it across the domain and across the, uh, the regions as well.

So some of the common things that we have seen, uh, in the enterprise-wide standards that we have seen is platform standards, having the data and, you know, having the cloud and AI capabilities that are common across the mining sites. Um, you know, work order codes that may be relevant to a specific site. I'm not saying that each site can be commonized with what is there. There are site-specific in- information which is proprietary and, you know, that's regional to a particular site. But then you have to evaluate and see if the agent that needs to be tested, um, and probably a digital twin as, as well to validate and see whether it is optimal for that particular site.

Product ownership is another area that we have seen, uh, you know, accountable for adoption of your, your workflows so that somebody who owns a specific product within the, uh, the process optimization has to take ownership and say, "This works in this site and this works in this site." So you may have to tweak the, the systems a bit in order to be able to scale that. Um, and also, um, the right model is important, right? The common data model. You have today within, uh, our foundry, I think we have eleven thousand plus models that are out there, large language models that are out there. Pick and choose the right model that is optimal for your operations.

Um, sometimes nowadays in the domain side, we are seeing that SLMs, which is the small language models, are useful because the token usage is less, but we're seeing that it's more useful in the IoT operations for SLMs. The LLMs can do a lot of the heavy lifting and more of the cognitive work that's needed. Um, there is an important piece called the proprietary learning loop, um, understanding the failure modes and what the inter- interventions are needed in order to be able to scale and to make, uh, standards across.

We want AI to meet the people where the work and the human ambition happens, and that's what we're looking at. In the sense that you have an ambition to produce more gold or more iron or more for your business. You have the intelligence that is there- And then for that intelligence to work, you need the trust of the intelligence and governance that happens. Mm-hmm. Right? Today you have Work IQ, which is how's the way we work, and we pull that information, and you have Web IQ, which talks about gathering information from research and other documents that are out there. But you need to ground it, all the intelligence, with a foundry so that the information is grounded and not, you know... The data quality is an important piece there- Yeah ... so that you don't run around, have these agents run around and be rogue in there.

And of course, um, the last piece is more in terms of the Fabric IQ. Your data platform, your data quality needs to be in place so that you're able to ground it and provide the information. So with that, you should be able to scale, um, at different sites and also keep the proprietary learning information going at your site as well.

Akilan, a follow-on question. Are you seeing that what you're describing is, is a step forward, in my experience, compared to where, uh, many of our mining, uh, clients and, and mining companies are. Are you seeing that in other industrials? Are you seeing energy companies?

Um, yes. So we, we work with oil and gas companies and power and utilities as well, and we are seeing that they have different site locations for oil rigs- Yeah ... and also for different locations and ore body characteristics. One, uh, similarity between mining and oil and gas is blending of the ore is similar to blending of the oil, uh, crude oil that comes from different sites. So we're seeing that how do you optimize that process? How do you pull the, uh, necessary agents to e-evaluate the process optimization so that, you know, we can, you can use it in both domains. And so that's where we're seeing, uh, improvements.

Wonderful. Yeah. I think it's a, it's a compelling vision. I really like this vision of this proprietary learning loop. Uh, many- Mm ... mining companies I speak to are, are trying to also figure out how do we maintain an advantage, and I can see if you adopt that approach- Yeah ... actually you do maintain a, a, an advantage. Yeah.

Let's turn to a question from the audience. Um, a question here: How is the understanding of boards and CXO, uh, on the application of AI in mining? Ravi, you've probably got a good perspective on this. Uh, what's, what's your thoughts on that?

Yeah. I would say, I would say two things. The inevitability of it is very clear in terms of it's not if we have to adopt this technology, it is when. Yeah. So I think there is a lot of focus that I think this is really the way forward relative to the way we work, and so that acknowledgement is, is, is there across the board when you think about the board and the CXO. But the real challenge, and I think this is a challenge that the CFOs and the, uh, CXO continues to feel, is when does the cash register ring relative to here's my roadmap, and I think positioning the AI roadmap in an appropriate way, I describe the end-to-end solution aspect. I describe the foundational element on the data. What are the process adherence steps?

I think those are very, very important, and I think if you're able to put that roadmap where you are seeing monetizable value, and I think that's the challenge that everybody is trying to solve for. Mm. Uh, those are the two things. But it is no longer a question of whether we need to adopt these technologies. It is when and in what pace- Yeah ... can I do? So I think it is no longer a technology problem to solve. It is how can technology help to solve a process problem and a governance issue that goes with it, and I think that's really the triangle that we need to, uh- Yeah ... cross-pollinate across, is what I would say.

I think that sense of inevitability is, is, uh- Mm-hmm ... is, is for sure there and, and I think you framed it up nicely. It's not a technology problem, it's how do we adopt this? That's right. That's right.

There's a follow-on question here. Ferran, maybe I can ask you to tackle this one. Of the 19 mining companies you're following, what are the different AI models you are seeing? For example, a centralized AI improvement model versus a decentralized model. Ferran?

I've, I've seen... We've seen them, we've seen them all. Um, we've seen models where it's completely decentralized. Actually, the data science are embedded in the operations. Um, that's very good for adoption. Sometimes they, they, they lose focus though. Um, we've seen models that is completely centralized. For example, that was initially, Colquitt had a, a centralized model, uh, but completely centralized. Um, and that's... I mean, that has the difficulty of the adoption obviously at the sites. And then a model similar to what Akilan was saying, where you have the data foundations, the reusable, uh, elements like, uh, skills, the platform centralized, but then the, the sites, um, own the, the value, the, the budget. So it's three different modes.

Okay. We've got a few minutes left. I'm gonna close us out with, with, um... Ooh, gosh, suddenly we've got an interesting, another interesting question up here. Let's, let's, uh... Akilan, maybe this is a good one for you, and then we'll go to a close out question. Right. What is the most valuable data set that no miner is really tapping today with AI?

Oh, wow. That's a, it's a good question. Um, most valuable data set. Um, I mean, we've seen common patterns that I've... that we know about is the, the ore blending, the ore processing. You know, signal optimization is another one that we have seen. And then, uh, predictive maintenance and asset optimization is an area that's... it's across. Um, the, the most valuable is in the exploration space, that being able to use seismic information To determine the, the quality of the ore that is in there, and that's an area that we have seen in both domains, both in oil and gas and in mining as well, that they can improve, and that's a valuable data set to identify exactly where that ore is using seismic imaging- Yeah ... interpretation. And today we have the technology, the HPC, as well as the tokens and LLMs, that have the ability to pinpoint the actual location either of the ore or of the oil reservoir that's in there. So that's an area that we're seeing that will definitely add value, uh, in exploration.

Excellent. Another interesting question. Uh, uh, the, the, the questions are getting more and more interesting, so- [laughs] But I know we have a hard stop at 8:00, so we'll squeeze in one more. We'll squeeze in one more. Is there an ecosystem of innovation designed for mining? Uh, how many startup companies, if any, are trying to develop solutions for mining, and how receptive is the industry to try them out? Hmm. They're all looking reluctant to answer that question. [laughs] Maybe it's the time of day.

No, I think, I think it's a, it's a very rich ecosystem. So ex- for examp- the exa- the clear example is exploration. You have, like, companies, like, super innovative that actually, um, put the equity on the, on the exploration, um, and, and they are very innovative. I think in different regions you have different receptivity of the companies. Yeah. So for example, I think in, in Australia there's a good, uh, relationship between universities, the startups, and the companies. We just did a study in Chile with Fundacion Chile. And Chi- and Chile is a, is a, is a disconnect. There's no, um... Even though the mining companies say, "Those are the teams we would like startups to work on," uh, it's very difficult for the startups to find the right decision-making. Mm. There's a lot of funding problems, a lot of problems for- Yeah ... for startups to get to, to implement at scale at, at mining companies.

I would also point out the rate of innovation in mining in China is really something quite remarkable now. When we look at, um, autonomous vehicles, for example, 5G deployment, or many other, uh, dimensions, um, electric vehicles in mining, there's a real uptick, uh, in, in China of late, so...

A close-out question to all of you. We'll have to keep it brief, but if you were to give one piece of advice to the, a senior executive from a mining company, a CEO, CFO, CTO, over the, the, for something to focus on, something to do differently over the next year, what would that be? Ravi, let me start with you- Sure ... and we'll- Yeah ... we'll give that a go.

Y- Y- Y- Let me say, I think I said a few things about end-to-end solutioning and the foundational aspect as two very, very critical aspect. I'll add one more thing, Richard. Embed the tech teams with the frontline- Mm ... is one very, very critical advice. I think there is always the corporate versus site type dynamic- Yeah ... that you see in many mining company. So find an operating model to embed the tech resources who are doing the stuff that we are currently talking about really at the, at the, uh, sites. You know- Yeah ... at the rock face, if you will- Superb ... to the extent possible, and I think that is gonna move the needle a lot more rapidly than, than other paradigms. So combined tech, frontline teams- That's right ... site-based. Form the core parts of teams that work together.

Akilan? Yeah. For the frontier transformation I, I don't think there is a single process that would take it together. Have a workflow ownership. You know, evaluate and trust infra. You build that evaluation harness. You know, have a common platform that we talked about. Um, the P&L will show up, you know, when the downtime is avoided, when the recovery improves, you know, and there's energy consumption that's, you know, optimized for you. So it's, it's a collection of, uh, processes that will help you get to that state.

Thank you, Akilan. Any closing words, Ferran? Yeah. So I agree with Ravi that the gap between operations and technology is where a lot of the AI in mining dies. Yeah. So I think, uh, one of the solutions for that is selecting a single owner, uh, that has one line of budget. It's a BLN owner. No, it's not the, it's not an AI COE, it's not IT. And that helps g- you know, connect to the P&L and close that gap between- That's right ... operations and- That's right ... technology. That's right.

Super. Thank you, panelists. Yeah. Thank you. Much appreciated. [audience applauds] Thank you. Thank you. Thank you very much.