How Enterprises Lose Control as AI Scales

Transcript

Nick Earle: Hi, my name is Nick Earle, and in this week’s episode of IoT and AI Leaders, we’re talking to ProArch. They are a very interesting company, offices in the US, based out of Atlanta, but they’ve also got offices in India. In fact, they’ve even got a presence here in the UK. And we’re talking about AI, not just the normal way, but how it is being adopted so quickly, and it is so cheap, and it’s so available, and the marketing hype around AI is so big that actually we’re creating chaos. and this is a company that, you’ll hear on the, recording, you could argue that their, motto is, “Slow down and define your processes before you get started. Don’t do what you’re about to do. There will be chaos.” And we talk about, what that means in terms of business processes and workflow definition and security.

We talk about what that means in terms of IoT devices and how that’s going to [00:01:00] potentially make the number of data sources 50 times bigger. And, and we talk about it in terms of, the responsibilities at the board level to understand this because of the chaos, reputational damage, security, issues that, that could come up by just, encouraging people to adopt AI because that’s seen as being the right thing to do ’cause everyone’s talking about it.

It’s a very interesting podcast with two very knowledgeable, people, the CEO, Santosh, and, Jim, who’s the director of cloud strategy and, AI enablement. And, these guys, are working with clients, all the time and, across different industries, and they even– they talk about case studies as well that they’ve found, and the lessons that they’ve learned.

So with that, I’m gonna hand you over to the podcast recording with Santosh and Jim, and I hope you enjoy it.

So, Santosh and Jim, welcome to the IoT and AI Leaders podcast. Great to have you.

Jim Spignardo: Thank you so much.

Santosh Kaveti: Thank you so much. It’s great to be [00:02:00] here.

Nick Earle: We’ve only done a few where we’ve had a double act at the other end. So this is more like a chat show format, late night TV.

I’m sure this will be very interactive. Santosh is the CEO of Proarch, and Jim is the Director of Cloud Strategy and AI Enablement, and that’s what we’re gonna be talking about, is the AI enablement and how your company… What your company does and how it’s changed and how it will change as IoT gets sort of blended into AI.

And then a prediction section, which we always finish on, with predictions for the future and, what- what will work and what might surprise us and we need to watch out for. So with that, let’s, get going. Maybe I can just start off, with you, Santosh, and then you, Jim, just to give a little bit of background on the company, when it started, and, how you would describe what you do today to the market.

Santosh Kaveti: Yeah, I’ll kick it off. well, we are celebrating our 20th anniversary this [00:03:00] year. So it’s been a great journey, and we’re actually excited about our future. Historically, we started off as a product engineering company. later on built a strong partnership with Microsoft to build cloud data and security practices. today the best way I would describe ourselves is, we help our customers operationalize AI. I know that’s a broad word, but that’s really our core value proposition. we’re at the intersection of, cloud data security, and AI. and what we bring to the table is a full stack. and- and these conversati- conversations are collapsing fast, especially in highly regulated complex environments in energy, healthcare and manufacture.

Nick Earle: before I ask Jim to talk about what he does, maybe I can just jump in on that last point for the listeners. why are they collapsing [00:04:00] particularly fast in regulated industries? why is that?

Santosh Kaveti: So historically, conversations around the security or sometimes even data where either with the IT folks or with the security teams, and maybe there is a data team involved in there. right now, when we look at majority of our conversations, they are with the business folks. Business functions are beginning to realize that they can now take advantage of AI. And as a result of this conversation, what’s happening is, of course, to operationalize AI, to build a chatbot is one thing, to deploy Copilot is one thing, but to actually operationalize AI involves embedding AI into your workflows and redefining them. And now, all of a sudden, all the conversations are collapsing, [00:05:00] where you have the, all the key stakeholders coming together and saying, “How do we unblock?” And the blockers typically are in the operational operating discipline, which boils down to their data security, quality, governance, and in general, the security posture.

Uh, and we’re seeing a lot of these conversations collapse very fast, whereas before they were siloed, focused on one function.

Nick Earle: And of course, those issues to do with security and governance are, are very prominent in regulated industry, so thanks for that. Jim, your role at the company

Jim Spignardo: Yeah, absolutely. I’ve been with ProArch, over nine years now, mostly in consulting capacities. my current role really is around helping clients modernize their environments, moving to more cloud-first, solutions. and then most recently is around AI enablement and adoption. and typically when I get a client, they’re at the point where they don’t even know the right questions to ask necessarily. They know that- Yeah … they’re supposed to be doing [00:06:00] something. They may be getting pressure from their executive team or their board. “Well, I saw a podcast the other day or read an article.

It seems that we should be doing something in this space.” So I really help them set the foundation and, the building blocks to, ensure that they’re successful. So that includes things like establishing usage policies and a center of excellence, champion programs, getting them familiar with frameworks to establish use case development, but also in having a clear and honest discussion around data risk. and coming in and actually helping them, review their current environment to expose where some of those risks might lie- As they begin to think about piloting some of these tools initially, so that when they wan- wanna go to scale, they’re doing it safely and securely

Nick Earle: I always try and relate [00:07:00] it to, recent events. And, your point you made about they don’t even know what questions to ask I think is a really good one. I was at a dinner, last night in central London, and round table, and people were talking. And when you are involved in AI and you are at the sharp end or near the sharp end , then sometimes it’s easy to forget that, that an awful lot of people actually are, are- don’t know what questions to ask, as you said.

Yeah. And one guy around the table said, ” we’re in the Donald Rumsfeld, era of AI. there’s no known known… There are no known knowns.” Yeah. I’m familiar with that one. some, some people know what they don’t know, but most people don’t know what they don’t know. Oh, yeah. And I think that’s ex- And as, as you say, a much, a much more succinct way of describing that is they don’t even know what questions to ask.

Which leads into the question for you is that does, has your company, sort of given that and the pace of change- Does that mean that, are you a, solutions company? You, you talked about Microsoft. You, you’re- Mm-hmm … a [00:08:00] consulting company.

Are you becoming a strategy front end company, and then consulting, and then- Yeah, it’s a- And then, yeah. What are you?

Jim Spignardo: It’s a little bit of all the above. we still have managed service practices, both in the security and support space. But I would say as we think about the next three to five years, we really wanna kind of reposition ourselves as a company that, is building systems, and our own intellectual property around AI solutions. and, and all of the other things that go along with that as far as services to support organizations in data governance and security. And so w- we’re, I, I believe, ahead of the curve, which is good. that doesn’t mean that there aren’t others who are also thinking the same mindset as we are. Right … but we’re looking to kind of transform the way our, our, the way our business is seen to be more AI-focused. And also realizing where is, where is the revenue streams [00:09:00] as well, right? Where, where are the next the revenue streams that we wanna chase? So…

Nick Earle: both for you and for your clients, because the ROI on AI, there’s a lot of talk about, well, you don’t have to hire as many people, potentially even reduce people.

But- Yeah … and you can get more work done. But, but at the end of the day, if from what I understand, you’re looking to, yes, you give them advice, but you’re looking to s- sort of get into work- strategically position yourself as getting into their workflows so that you- Oh, yeah. Absolutely … help, help them transform, and that’s, that’s very different to people just using ChatGPT- Right at their, at their desk, right? That must be a- Yep … conversation you’re regularly having with people.

Jim Spignardo: Yeah. And, and really the focus becomes more about outcomes, right? And looking at why are we bringing Broad in, and, and what are we going to help them with, a-and how are we gonna demonstrate that we’ve actually, improved or, helped them in some, some material way? hopefully in some sort of quantitative [00:10:00] way. Although there’s definitely qualitative ways to, to assist, but, that’s really come, become the expectation, I, I believe, in, in this particular, business environment now. Especially when if you’re using AI to develop solutions, it’s more of, well, you’re not just developing the solution for me, you’re developing outcomes that I want to actually… you can actually guarantee that I’m going to be able to recognize, and that’s very different. And then also trying to price your, your products and solutions around, being able to deliver those outcomes.

Santosh Kaveti: I think if I can add to what Jim said, Nick. So for us, our goal has been to help our customers operationalize AI.

And to do that effectively, we have to consult, we have to strategize, we need to build. We’re not there to just tell them, “Hey, here is what you need to do.” We’re there to actually execute, so we [00:11:00] need to bring our own solutions and IP. And at the end of the day, everybody asks the question, “Well, what did this do for me?”

To be able to answer the CFOs or CEOs and the boards, we have to show the ROI, and that’s auditability. the delivering or measurable auditable outcomes is becoming super important for us. So essentially, the worlds are also converging the consulting strategy and implementation, again, thanks to AI. and that’s where we are at the intersection of all of that right now.

Nick Earle: Has that changed, Santosh? the, When I’ve spoken to, some companies that, that do similar things, as you say, there’s a lot of companies doing… Are in this space just because of the opportunity being raised by AI.

But one of the questions is: Has the trigger where you get called in changed? In other words, Is it that people say, “I’ve got a blank bit of paper, I need some advice before I get [00:12:00] started”? If you’re gonna climb Everest, you- you’d probably be a good idea to go with a guide, because you’ll screw up.

If, AI looks really simple, and it’s free, and, a lot of people seem to be using it, probably a chatbot not an agent, then you may say, “Oh, I don’t need advice.” But the board would… Then the board asks the question, which is, “What are we doing with an AI strategy?” And then suddenly people say, “Well, I’m not sure.”

And they ask a few more questions about, about security, about compliance, about governance, and about risks. And then it can be, “Well, we need to bring in some people to help.” So, so to what extent is it, individuals or board level? , What’s the, intr- introductions or, opportunities that you see, and what’s the, what’s the trigger, and is it changing? is it a subject that’s being embraced at the board level now?

Santosh Kaveti: So the conversations today as companies move away from or move [00:13:00] from pilots, whether they’re successful or unsuccessful, to scalability and to embed AI into their current workflows, the CEOs and boards are beginning to ask the real…

At, as companies are beginning to allocate budgets, especially for AI-driven transformation, these questions are becoming not just relevant but critical. I think there is a lot of excitement, and there is a lot of activity, but the… When it comes to operationalizing AI and doing it the right way- Right now, the boards and CEOs and CFOs struggle because even they, as Jim said Earleier, they don’t know what questions to ask. we strategize with them, for them to help them come up with the framework on how they should approach [00:14:00] a strategy, what’s important for them, what they should look for. AI should not change the accountability. AI, of course, will magnify the chaos, but accountability becomes even more important. one of the key things that we do, our teams do, is we build that truth layer. We build that op-operating discipline. That’s where most failures of AI occur is because of weak operating model. That goes into governance, that goes into security, that goes into transparency and so on. But that’s what our teams help.

And of course, ultimately, always try to draw that always have your North Star. Why are we doing this? Are we gonna get to that outcome? And if so, how can we prove that we have gotten to that outcome, which is operational efficiency or cost savings or new revenue streams, whatever they are, that CFOs are very interested in.

A-And are we costing everything the right way? so yes, these are even at, at the [00:15:00] top level, especially at the top level, there is more, confusion, lack of clarity on how do we approach AI, how do we create a strategy that takes into account all the factors? How do we allocate the budget? How do we really measure the ROI?

Nick Earle: and Jim, you wanna add to that?

Jim Spignardo: part of what, we see a lot of is, there hasn’t been a tremendous amount of governance within organizations Right … especially the small to medium-sized businesses. And even at the enterprise level, I would argue there hasn’t been this type of, collaboration at a business level, from a governance perspective either because I think AI is now exposing that there, there needs to be multiple teams involved and working together. one of the biggest things, which gets exposed very Earley on are, well, undefined or, or poorly defined process, and ownership and accountability, and that can [00:16:00]really, have a serious impact on the ability to, realize some of the potential of AI. but the upside is, that if you can recognize those shortcomings and address them, you’re gonna come out much better on the other side of this. you’re gonna have a better, more well-defined, better working teams, with automation and AI layered onto that

Nick Earle: again, there’s always, echoes of, of the past in IT. we tend to go around the same loops, every few years, but we’re using different technology and different terms and, certainly I was thinking as you were saying that, when we went from, MRP, manufacturing requirements planning, to ERP, people were saying the same thing. If you don’t have the processes defined, you’re just gonna get basically crap quicker. and, a lot of people had to press the reset and say, first of all we need to map out the processes, and then of course, companies like SAP in particular, Oracle as well, and others, start saying, started saying it’s all about the object layer, the [00:17:00] workflow, the consistency, the governance, the fact that whoever, whoever uses this system, you’ll, it’ll be auditable, it’ll be traceable. you’re not gonna have arguments a- around the, around the management table where people are saying, “My system says this,” and, “My system says that.” And it seems that with AI, the potential for that chaos, with everyone experimenting and the fact that they don’t need IT approval to use, to use AI tools, you don’t.

It’s, it’s free, and even if, even if you want, the, the version with more memory or recall or, or advanced research facilities, it’s maybe 20 bucks a month. It’s not, an IT decision. It’s sort of like when cloud first came, and a lot of people were breaking governance by storing data outside the organization. is that analogy correct? And, therefore is the implication of that that the potential for getting inconsistent processes and, automating point solutions rather than looking at workflows and horizontal [00:18:00]processes, is the potential for that chaos getting bigger in your view?

Jim Spignardo: I think the potential is absolutely getting bigger. this is why we’re very clear when we work with our customers, that although these things can be done this way, right? People are able to go out and use some of these free tools. and now we’re talking about the security, portions- of our business. You have to get visibility into what folks are doing, and you have to clEarley define through good policy what is and what is not permitted, establish the guardrails and guidelines, and also layer into your security stack, tools that can place controls, right? You’ve gotta claw back some of this, some of the chaos and, apply some of the tooling that you have in order to understand what is being done. and you do have to go through and set up a process through a governance council or committee, how does something get approved, [00:19:00] right? So, and at the same time, you don’t wanna create a lot of red tape and bureaucracy, and you do wanna still encourage innovation.

So if you have that visibility and you can, you can be, assured that you understand how these tools are being used, you can confidently innovate at a pretty rapid cycle. But without that, you, you really need to, to kind of step back and say, “Okay, these are the things we’re missing. And, we have to, not necessarily put the lid back on it, but rein it in and make sure that we can, we can absolutely control where our data’s going where our, where our data flows.” again, it’s all the combination of all those things, that you have to have in place in order to, go forward with confidence.

Santosh Kaveti: Yeah. I’d say that most companies today are building on unstable foundations and a lot of assumptions. as we’ve said before, you can’t really [00:20:00] bolt, an AI onto the current processes and systems.

Yeah. They are layering AI on top of fragmented data- Yeah … legacy processes, disconnected systems, and really with unclear governance. so they have to shift from… It’s not about how do I deploy AI. The first question should be: How do I prepare my environment to support AI? I think that shift has to happen. We have to go from the tool-first to foundation-first approach. That’s, that’s, that’s what’s gonna matter.

Nick Earle: I can imagine that you guys, and this is my assumption on your business, I have a services background. I run a large part of Cisco services business globally. When we had to deliver tough messages around the, “You need to slow down to speed up. You need to do this first,” and people have said, “No, no, no, I just want to implement this,” which I can imagine is conversations that you get into. We had to have the discipline from a sales point of view to say, “Well, it, you can do that, but not [00:21:00] with us. we’ll walk away because it, it’ll fail, and we don’t want to be associated with that. And our advice is get this bit sorted out first,” which is a consulting led, we call it… Services led is what we called it. “And you have to do that, and you have to get senior management buy-in, and then we’ll do the implementation.”

But a lot of people just wanted to get going with the implementation, and I imagine that’s pretty, pretty similar in, in y- In… I can see you smiling. That, that’s something that, that, I suspect you come across fairly regularly.

Santosh Kaveti: Yeah. So everybody looks at AI, and of course there’s a lot of excitement, possibility.

But the reality is, grounded reality is complex. And, and everybody also thinks intelligence is the hard part. It’s really not. The hard part is the trust.

Nick Earle: Yeah.

Santosh Kaveti: Hard part is asking really good questions like, “Can I trust the data? Can I trust who is accessing it? Can I trust the output? Can I explain how a decision was made?”[00:22:00]

We’ve had time and again our own clients they think, “Hey, we could do this on our own. I think our, our team can do this.” Few months in, they call us and say, “Look, we need help.” Yeah. And the help is really not on technology side. Help is really on, on creating the truth layer. We, need your help to create that trust layer.

Nick Earle: I wanted to pick up on that. You said that, Santosh, Earleier, a few minutes ago. The trust layer, the truth layer, you’ve described it in various forms. Is that a sort of process governance policy that a company needs to implement, or is it also enabled by technology?

Maybe you can just expand on that a little bit, because that’s clearly extremely important in this part of this.

Santosh Kaveti: Yeah. the, the trust or truth layer is where you have explainability, observability, and transparency, and auditability. That becomes super important. The second one is, hey, [00:23:00] how, how, how can I reduce the risk and how can I really protect myself? That’s where the technology comes into play. Again, there are really good tools now these days that even that we use all the time to be able to reactively and proactively help with your compliance and security.

Because AI brings in whole new level of security considerations- Yes … that landscape is only evolving.

Nick Earle: In a previous podcast, just to, for context for listeners, they may have listened to the previous episode, but it was, it was all on security.

And, we’re gonna come onto security, go deeper on, in this podcast, but you’re absolutely right. it’s not just security of your data, it, it’s the fact that when we talk about IoT, devices are coming in where you’re not in control of the security policy. I know in the previous podcast we went into the fact that when people buy software, they’re now buying software which contains agents, and how do you actually trust, measure, get auditability, governance, [00:24:00] accountability, all those things when you’re buying agents off the shelf?

Santosh Kaveti: Mm-hmm.

Nick Earle: people are making individual decisions- Absolutely … to implement software with, with agents inside, not knowing what the security policy is and the governance policy is of the agent.

It’s a huge growing issue

Santosh Kaveti: Yes, it is. That’s one of the first questions we ask is they have all of these ERP systems in addition to, to AI systems, and every ERP these days says, “Hey, I have, I have a set of agents,” which is great again. But the minute you enable AI in, in any shape or fashion, you have to understand that you’re introducing risk, and that risk amplifies.

When a machine fails in the previous context or the human error, it was a manageable risk for the most part. It could be catastrophic, but it was manageable. But in the future, that will not be the case, and we’re already evidencing that with recent examples. Jin, I’m sure you can add more color [00:25:00] to that

Jim Spignardo: The more we hand over to, to agents and AI to do work, we- the more, the more con, safety measures we have to have in place to ensure that what it’s doing is, is what we want it to do, and the outcomes are what we expect it to have. And also, it’s not all that dissimilar from managing individuals and identities, right?

Making sure that those agents can only perform the work that they’re intended to perform. and that it’s at the end of the day, especially when it’s decisions that can, have, c- consequential impact, that there’s a human at the end of that decision process to be able to, evaluate whether or not this is something we’re gonna continue to- The human judgment yeah … which is

Nick Earle: the hope as to why all the jobs won’t go away, because you do need a human at the end of the process-

Jim Spignardo: Right.

Nick Earle: Yep …

Jim Spignardo: make

Nick Earle: the call.

Jim Spignardo: and we have quite a few, clients in the medical space, and, and one of the things that, is becoming very apparent, as far as responsible and ethical AI for them is there should [00:26:00] never be any decisions made by AI that will determine the outcome of services, for instance, for a patient or member, right? We should never be making, letting those decisions be in the hands of artificial intelligence. They can certainly give input, and they can help with the decision-making process, but ultimately, that’s, those types of decisions are the ones that still need to be, human-directed. So that’s something that needs to be factored in when we start to think, “Well, yeah, we can do that, but should we?” Right? Right. Just because it can do these things on our behalf-

Nick Earle: it can,

Jim Spignardo: But at this point in time, we’ve all seen, AIs do tremendous things. We’ve also seen it fail miserably, and fail, spectacularly. And so, you don’t wanna be on that, that other end of that story,

Nick Earle: especially if it’s a regulated industry, as you said. we’ve started to see a, a lot of people saying, “Oh, well-” coding is dead. Claude Code is… you can do [00:27:00] 10 times quicker, 10 times cheaper, 10 times whatever. Yeah. It doesn’t sleep, it doesn’t get paid, it doesn’t take vacation,

Jim Spignardo: Right.

Nick Earle: unless you have governance, which I think, I think I’m using the right term, scaffolding, but, but unless you have governance around it, then you actually create something really dangerous much, much quicker.

Jim Spignardo: Oh, yeah.

Nick Earle: And, and if you get an audit, which, which w- you, y- your decisions have to be auditable, particularly if, if they have consequences, like in medical healthcare, you could have very severe conse- But most industries are held to account for- Yeah

Like data protection, personal data protection, even if it’s just that. So what you’re doing with storing and using people’s data. but if, if, if, if you… It seems like now people have got tools in their hands that they’ve not been trained to use, and they’re- Right … really cheap, and they can impress their boss by, by saying, “Look what I’ve built.”

Yep. Let’s implement it. That must be-

Jim Spignardo: Yep …

Nick Earle: that must be a, a big… Well, both [00:28:00] an area of opportunity for you as a company, but a big issue once you get in there. You must be finding stuff all over the place, frankly.

Jim Spignardo: And so an example of development work, right? I just wrote an article on LinkedIn, I think it w- got released today possibly, about how AI’s really good at…

Well, it’s, it’s a kind of a rookie at writing code. it’ll get the job done, it’ll do exactly what you want it to, but don’t look underneath the hood and figure out how it got to from point A to point B, because, there may be all kinds of, security flaws or architectural- Yeah … design issues that don’t allow it to scale. we can use A- AI’s really good at finding bugs. Yeah. So it’s kind of… It’s very ironic that it, that in its mission to accomplish the work, it makes a lot of shortcuts, takes a lot of shortcuts. But then when you turn that lens around and tell it to go and examine its own work, it can find all the mistakes it’s make- it’s making. yes, we able, we’re able to build something that accomplished the task, but w- was it done well? [00:29:00] Was it done securely? And, how do we make sure that we build things into that to follow our frameworks, to make sure that it’s d- Right … doing it in a way that ma- is, is, safe and effective?

Nick Earle: And, yeah, and, and, and it comes back to, again, and do people know what questions to ask, and they almost certainly don’t. Let’s pivot if we can, and, get into, IoT and, the- The, the theme of this podcast series, as regular listeners will know, certainly since the beginning of this year, has been how IoT and AI are coming together.

The basic premise being, in case anyone’s new, that, there’s 50 times more data that’s gonna be generated at the edge by things than is currently has been used to train the models, in terms of the, data, the sound, the video. The industry says train the models. The, the people who own the content say stolen data to feed the models.

But either way, that’s flattening out [00:30:00] in terms of, we’re not creating content as fast as the models can absorb it. But now you’ve got this 50 times at least more data being created by things at the edge. and so, the opportunity is clEarley huge. The idea of almost creating a digital twin of something that’s really complex, like a jet fighter or a manufacturing process or a product or whatever, and, and to say, “I’m gonna collect data from all these things, and I’m gonna use AI as a layer,” notwithstanding the problems that we just talked about.

But AI is a, a sort of overall layer to produce this enterprise dashboard, to, show what’s exactly going on- Mm … so I can turn all my processes from reactive to proactive to preemptive, which is it didn’t break because I made an adjustment before it broke. It, it, it’s very attractive for companies. but it’s… are you starting… So the question is, are you starting to see [00:31:00] people ask about that, maybe make similar mistakes of, of getting started and screwing things up? Is that becoming a, a growing part of the work that you do?

Jim Spignardo: Danthash, you wanna take that one?

Santosh Kaveti: so y- yes, it is possible to- To preemptive, prescriptive, or proactive and pres- and become prescriptive and even become reflective. but I think at the moment there is still a lot of ROI to be gained just to, to be able to do proper predictive, analysis.

Nick Earle: Yes.

Santosh Kaveti: we have, one of our customer, segmentation or segment is power, power plants. We help a number of power plants, whether they are renewable, fossil, hydro, gas, natural gas, in the U.S. here. So what we’re beginning to do there with some of our customers is build what we’re [00:32:00]calling as an equipment AI hub essentially we take a type of, equipment.

We, go through the process of, creating the truth layer, and after that we get to a point where we’re looking at, high risk, high value, use cases. let’s say we identify certain equipment that falls into that category. we’re now able to do predictive, analysis much more accurately, and also connect to the IT side of the systems.

So we’re able to say, “Hey, look, this particular equipment, based on all the data that we have gathered so far, and looking at their optimal performance, their manufacturing curves, the drift potential, what should have happened here, but this is what we’re noticing, and therefore these are the possible reasons, and these are the possible recommended actions based on all the access that we have to the, the trained [00:33:00] data.” and then also take it to one extent and say, “Hey, by the way, one of the recommended actions could be, A, just you’re better off replacing this particular equipment as opposed to scheduling maintenance window, and these are the reasons why, these are the costs.” But in order to do all of that, we’re, we’re having to connect systems, across the point, meaning IT and OT, to be able to get all the data and, and also the trained data. of course, as, as we, we continue to say, human in the loop here is very important. but detecting these anomalies, Earley on could save six figures easily or more.

Nick Earle: we had a, guy… in the oil platform business-

Santosh Kaveti: Mm-hmm …

Nick Earle: He said there’s something like 8 million things get measured on an oil platform at sea, which I found completely mind-blowing. Right.

Santosh Kaveti: Yeah.

Nick Earle: But the… If you think about it, when something fails, if, if, if the, if they, if they have to turn the well off, in the, in the most extreme example, and they need a part, which they didn’t do what you just said, they didn’t [00:34:00] predict that, the, it, it was behaving and therefore there was a, there’s a pretty good chance that within X hours or days it’s gonna fail and it’s critical.

In their case, they, they, they pro- There’s not enough space to have spares for a c- second rig on the rig, so it’s a helicopter. Mm-hmm. It’s a helicopter, which just means it’s two, three days.

Santosh Kaveti: Mm-hmm.

Nick Earle: Two, three days with no oil flow, it… , six, seven, could be 8 million. I think it’s- 8 million. yeah. I know, it’s… And so that is… I think you’re right. I think that is absolutely the low-hanging fruit. and there are certain industries where, like energy, like you say, oil. There’s arguably most industries have their own version of why that is the biggest low-hanging fruit to go after first.

Santosh Kaveti: our goal there is we’re beginning to start with certain equipment. Again, we’re choosing the high-value use cases. And but our, our goal here is to cover all of the equipment that goes into production [00:35:00] of electricity, and then create a, an ontology on top of it. At that point, it’ll be easy for us to build operating agents or data agents. the good news is that, we partner with Microsoft, of course, but good news is the technology is, is now available to, to do this. But again, at the moment our journey is we’re beginning to go equipment af- equipment after equipment, and beginning to really understand the data, understand the workflows, understand the nuances to say how do we really predict accurately?

Nick Earle: Phrase popped into my mind, it’s almost like mapping the physical genome, not the human genome, The millions and millions, and they all use different interfaces, and the data is all can be interpreted a different way.

It’s… People think connectivity, just getting the data is, is simple, but it, we know from being an IoT company that just, connectivity is one of the most complex things ’cause all the devices are made differently. In the cell phone [00:36:00] area, there’s sort of six or seven cell phones that make up something like, 97, 98% of all cell phones, and how they behave, and it’s all documented and whatever.

But every product, every piece of equipment, every sensor is kinda different, and the company didn’t make it. they bought it. Mm-hmm … and so, it’s all using different standards. A- and so to map that… Jim, I suspect, that as an ambition to map the ontology, that’s gonna keep you busy, isn’t it?

Jim Spignardo: Yeah, for sure. I do believe that if we’re… You’re gonna get the maximum ROI, those are the types of, activities that these, these organizations will have to undergo. and, I think it extends beyond just, the, in- industry types of organizations manufacturing p- power generation utilities. And we go back and have that conversation about companies that are in professional services, medical, Yeah … legal, whatever. It, it, the, the opportunities are too big not to be, not to really kind of [00:37:00] clEarley define how work gets done in this place, right? Yeah. And for so many, for so many organizations, the, the organization actually survives and somehow, even thrives to, to a certain extent, without a full, recognition and a full understanding of, of the function of the business and then how things are done.

Yeah, absolutely.

Nick Earle: Most people have no idea.

Jim Spignardo: Yeah …

Nick Earle: I always like to reference it back to previous podcast guests. We had a guy on from Turkey actually who, was trying to do it for a hospital. You talk about healthcare. Well, he was doing it for a series of clinics and hospitals and healthcare groups, and, basically he said- they don’t know what their processes are.

If you say to them, “What is the process for coming in for a hip replacement?” Nobody knows. Nobody. The end-to-end process.

Jim Spignardo: Well, they know their piece, yeah.

Nick Earle: Anyway, and, and most of the, the delays are to do with inefficiency between, parts of the silo.

So he basically was… put a little, sensor, a Bluetooth sensor, on every piece of equipment. He put a bracelet [00:38:00] on, every, ev- every, patient when they came into the hospital, they got a little bracelet so they could track them, and he put a tracker on every physician, and-

Jim Spignardo: Mm-hmm …

Nick Earle: and so then you would then have this map of people, staff, and physical assets, and just collecting that data and putting it into this high level would actually give you your first pass at what your end-to-end processes are. Yeah. And you could see that, that your biggest issue is the fact that i- it…

the, the inf- and he said the example was once people have been diagnosed and the doctor says, “Well, this is the medication you need,” you then go back- Mm … sit back, sit back down again, and you can wait an hour until your medication is available just because there’s a very inefficient process from that department to the, dispensary- Yeah

or whatever. And so, so just having a visible dashboard can make a massive, change, and that’s exactly what [00:39:00] you’re

Jim Spignardo: Yeah, 100%, because, again, if, if, if you’re not understanding the upstream and downstream effects of what you’re trying to solve for, you may, create some unintended consequences.

Nick Earle: Yes.

Jim Spignardo: Someone, someone makes a change to a system. They didn’t know another group depended on a specific field within a data set, and now they change the… changed it for their benefit, and now it breaks it for somebody else, right?

Nick Earle: Listen, we’re getting towards the end of our time, so I wanna finish, if I can, for both of you with a, a question. We always do this on the pod. It’s sort of called the, the look forward, section. So, I’m gonna… I don’t mind, maybe one of you wants to take the lead, or you both chime in, but essentially the question is let’s imagine it’s two years from now.

E- e- everything that we talked about, is happening, but it’s also accelerating, and who knows? They say AI’s getting better at 4X per year, not 40% like Moore’s Law. So we could have 16… Potentially, some areas could be 16 times more effective than they are today. [00:40:00] So, so, what are the, I think we’ve seen the opportunities like mapping out the ontology, connecting everything, identifying the processes, optimizing the processes and the workflows. What are your concerns that are big i- industry issues that we haven’t yet solved in this space?

Maybe you could each have a crack at that

Santosh Kaveti: Jim, want to take the lead?

Jim Spignardo: I think if we go back to the initial, conversation that kind of kicked this all off, there has to be a reckoning in most organizations about how they’re handling their data, and the quality of that data, and what they want out of the data as far as understanding the impact it has to their business, so that they can, apply this technology appropriately. and that includes looking at their security stack. I would almost guarantee the vast majority of organizations out there today do not have tooling in place that is ready to deal [00:41:00] with, managing and controlling, putting controls in place, for AI systems. And so, that’s not something we can hide and ignore from anymore, right?

It needs to be top of mind. let’s assume that they don’t even go down a path of implementing AI, which would be almost ridiculous, but, there, there’s still benefit from going through that exerci- that exercise. And I think the, the organizations that can get there quicker are gonna be able to take advantage, faster and see a competitive advantage over their peers.

Yeah. those that continue to struggle or continue to drag their feet because it’s just not fun or not exciting or it’s the dirty work or we’re gonna have to hire- Yeah … different resources to make that happen, those are the ones that are gonna start to be left behind.

Nick Earle: And, and Santosh, that would mean at the board level, it’s not just that they’re left behind.

Something could go really wrong, and you’ve got reputational damage and brand damage, [00:42:00]and it’s really, it’s g- potentially really hard to recover from those things if they don’t do what Jim, Jim says upfront, if they don’t pause and, and focus on this upfront, right?

Santosh Kaveti: In the board level, I would say three risks, Nick.

One is obviously the security and compliance risk. AI is continuing to amplify that risk- and it will explode, as my really concern what keeps me up at night. The second one is at the board level, the business model risk, because AI is disrupting so many business models, commercialization models, that everyone will have to reimagine the value that they’re adding to their own customers and come up with a new way of, managing the customers, including commercials.

I think there will be… we’ve seen that with SaaS companies now. Almost consulting companies. Many of these models, legacy models, historically were very strong, are, are completely [00:43:00] being disrupted right now. The third equation is human capital. I think that’s probably one of the big disruptions that boards will have to think about is how will the, the team look like in a company two years from now?

How and how will an AI native… Almost everybody is now– There’s a demand and there’s a requirement for every individual, whether they’re technical or not, to kind of become AI workflow engineer.

Nick Earle: Yes.

Santosh Kaveti: So when there is that expectation that everyone is gonna become an AI workflow engineer, what does that mean? how do we train somebody? How do we get to them, get to that state? So these are the three risks that boards will have to worry about. Security risk, compliance risk, just the entire business model risk, and the human capital risk.

Nick Earle: I really want to, say thank you. I think we’ve had a really good, conversation. certainly I’ve learned a lot. there’s a huge opportunity, a lot to do. but if I was to take one [00:44:00] takeaway, from it, I think it’s you’ve got to slow down before you get started, and, think about it, and take advice. to do this on your own without working with a partner, such as ProArch, is, as we say over here, a brave choice which is an English way of saying, “That’s a really dumb thing to do.” Yeah. Working with a partner, it seems to me to be essential because suddenly we’ve democratized access to AI, and everyone is innovating.

And as you said, and just on that finishing point, now we’re telling people, we’re even measuring people and doing their performance reviews by how many tokens they’re using, right? So we’re actually encouraging them to do the opposite- Mm-hmm … of what, of what we’re saying here. We’re saying everybody has got to start innovating and do vibe coding and use Claude, and it’s like, “Whoa, whoa, whoa, whoa, whoa.

Stop. Stop.” You’re walking around a, a dry forest throwing matches everywhere. This is not gonna have a good [00:45:00] outcome, right? Right. so, Jim and Santosh, thank you so much. I think ProArch’s doing some, very important work, and I’m sure you’re gonna be very busy going forward.

Thank you again for being my guest on the, IoT and AI Leaders podcast.

Santosh Kaveti: Thank you for having us.