Matthew Kleiman ADT podcast cover
Episode: 129

Matthew Kleiman - Leveraging AI & LLM without falling for the hype

Posted on: 04 Apr 2024
Matthew Kleiman ADT podcast cover

Matthew Kleiman is the co-founder and CEO of Cumulus Digital Systems as well as the author of the recently released book Work Done Right.

In our second conversation with Matthew, we talk about making use of artificial intelligence, and in particular large language models, without falling for the hype surrounding these technologies. We revisit the concept of the "splashy technology syndrome", discuss how to identify those solutions that bring real value, and take a look at some examples of how Cumulus leverages AI.


Links & mentions:


"How are existing companies integrating AI into their existing products to make them much, much better? And I think that's where the initial major value is going to be created in the SaaS software space, not so much new applications, but the companies that can successfully make their current applications much, much better."

Welcome to the Agile Digital Transformation Podcast, where we explore different aspects of digital transformation and digital experience with your host, Tim Butara, Content and Community Manager at Agiledrop.

Tim Butara: Hello everyone. Thanks for tuning in. We're joined today by returning guest Matt Kleiman of Cumulus Digital Systems. We already had a great conversation with Matt recently. It was such a great one that we decided almost immediately after recording that we should schedule another episode. The first one was about doing work right and the importance of work being done right, especially in industries such as construction and maintenance.

And this time we're going to talk about a very hyped up topic, artificial intelligence, but to have it have a special twist, as opposed to all the hyped up conversations surrounding the potential of AI and, you know, just all that, we decided instead to focus today on how we can best leverage AI and large language models without falling for all this hype.

So Matt, welcome back to the show. It's really great having you here again with us. Shall we just jump right in before anything else? 

Matthew Kleiman: Sounds good, Tim. Great to be here. Thank you for having me back. 

Tim Butara: Yeah, I'm also excited. The last conversation we had was one that I was particularly fond of, so, very happy to have you back here.

Matthew Kleiman: Sounds good. Let's do it. 

Tim Butara: So, okay, one of the main things that we talked about in our first conversation, which just got released, I think, in the first half of January of 2024, was the splashy technology syndrome. Can we start off by revisiting the Splashy Technology Syndrome a little bit and ask the question whether or not AI and especially generative AI falls into the pitfall of the Splashy Technology Syndrome?

Matthew Kleiman: Sure. And for the audience, we'll just, who may not have heard the last podcast, we'll just give a quick recap of what splashy technology syndrome is. And that's the term I use to call a company of any size adopts a new technology because it looks or sounds cool. It's almost more for marketing value than for actual business value, or it's a me too situation, oh, other people are talking about this technology. It must be good. I'm going to adopt it. 

But it's disconnected from actual business value. And what ends up happening is the technology deployment usually fails because people use it, and then once it gets out to the larger user base, they say, wait, this is making my life more difficult or isn't delivering value. And it gets canceled. And then people wonder why companies have such a hard time deploying technology. It's because they're often not using a rigorous analytical process, like we talked about, systems thinking, last time to make sure there's an actual business value. 

And in many ways, AI, which has been around for a long time, for decades,

and prior to late 2022, I think most people and certainly myself would have said, yes, that falls into the splashy technology syndrome. 

I remember when I was working for a large energy company in the mid 2010s, we had a number of startups and large companies that would bring their AI products and say, basically, give us all your data and we'll create some sort of value for you out of it.

And the problem was these systems were very cumbersome, difficult to use, and they required incredibly well structured and complete data sets, which in most companies, and certainly heavy industry and construction, just doesn't exist. The data is in folders and places or in people's heads or in spreadsheets that are incomplete.

And so they became impossible to use. But there was a real step change in late 2022 with the release of ChatGPT that turned AI from something that could be splashy technology syndrome into something that, oh, this could actually... I could see how it delivers value. And it graduated from, in my mind at least, from splashy technology syndrome into something that was actually really exciting.

Tim Butara: I'm really glad how we started off by revisiting the splashy technology syndrome. And also we'll make sure to link both of the episodes, I mean each of the episodes in the other's show notes, so that people who enjoy either of them will also be more inclined and more motivated to listen to both of them. 

And yeah, I agree. I mean... look, we recorded the last episode, I think it was fairly recently, like at the end of something like that. And things are changing so fast that maybe at that time we could have said, okay, maybe this could fall into this pleasure technology syndrome. But by now we can be certain that there's so many use cases and so many, so much innovation happening here that I definitely agree with your answer, and perfect opening here, Matt.

Matthew Kleiman: Yeah, absolutely. And I'm sure we'll get into this more. But the two big things that happened with ChatGPT is: one, the user interface became a chat interface that anybody could use and immediately see value of. It wasn't, upload data, then at some point in the future, value will be created. There was immediate value, and that's important for any technology.

And two, what's really exciting about LLMs, large language models, is LLMs are able to ingest and make sense out of messy data. And that's really important because that's what humans do. Humans, you get a spreadsheet from somebody, whatever your industry is. If you have some basic level of knowledge about what's supposed to be in that spreadsheet, you can go through it and see, hey, things don't look right, but I understand the point of this and I could adjust and I can make inferences of what's supposed to be here.

And that's what LLMs can do. And this is bringing that technology to the masses for the first time of, I don't need perfect data anymore to have a useful, whether it's a copilot or data interpreter or content creator, the LLM, much more like a human, can make sense out of messy data. And that was for us, certainly in the company that I'm the CEO of was the most exciting piece of it.

Tim Butara: Those are some really great points. And kind of moving on to our next question regarding, you know, the hype of AI tools, on the other hand, between AI tools that can bring actual value and drive transformation. In this context, how can companies as well as individuals, I guess, distinguish between these hyped up AI tools that fall into the splash technology syndrome and those artificial intelligence or LLM solutions that are able to bring actual business value and actually help drive transformation?

Matthew Kleiman: Yeah, I think there's a few things companies can do for that. One is, when you're looking at a new technology, really make sure you're distinguishing what is actually available versus what's in the demo or the video. I think we still see, as powerful as these tools are, we're still seeing companies, both big and small, release demos of their products, having to later walk back and say, oh, that was edited or that was... you know, we were showing, in some future state what this is going to be, not it's actually available today. 

So that's the most important thing is to really understand whenever you're looking at a technology, what is it actually doing today? What it will do in the future is important, that's exciting, but that's a huge amount of risk and uncertainty into your adoption.

So what does it do today? And then once you have a good understanding there, look at, depending on whatever your use case is, where is it going to get data from? Look at my business. Where are the data sources that the AI will need? Is it just relying on publicly available data? To make it really valuable to a business, it probably has to use some level of private data.

Well, where is this coming from? Is this being uploaded each time? Is this accessing some sort of file repository? What condition is that data in? Really understand that, because the AI results are only going to be as good as the data that you feed into it. Even though AI today, for what we talked about before, can do a lot more with many more different types of data than in the past.

And then the output, you know, what is this actually doing... once I put the data in, is this actually generating value? And value can be, it's creating insights that I don't know before. It's it's saving time. It's doing something better than a human could. Just because, you know, it could be you have an AI tool, but the output, the result is no better than a person or sometimes even worse than a person.

So that's what you have to really understand. One, what is it actually doing? Two, where are the data sources and what condition is the data in? And three, is the output that it's generating actually creating value? Is it saving you time or telling you something you wouldn't know before? And if you can answer those three things, there's a really good chance that what you're looking at actually is valuable and worth spending time on.

Tim Butara: So not being able to answer these things would equate to some of the main risks of falling for the hype around AI. 

Matthew Kleiman: Exactly. And what that will just lead to is disappointment because it'll be, it could be exciting for a few days. And then once people realize, oh, this isn't actually helping me do whatever it was advertised to do, then people will revert to whatever the status quo was, and it'll be another failed technology deployment. And that would be really unfortunate because in this case, there are really exciting use cases that could add value to a lot of companies. 

Tim Butara: Can we talk about some of those? Do you have any great examples?

Matthew Kleiman: Sure. At the risk of being slightly self promotional, I'll talk about what my company is doing with AI. So, my company is called Cumulus. It's a quality management system for work in the field, maintenance and construction, as you mentioned in the beginning. 

And, we... our company is five years old. And for all of the five years of our existence, companies would work with us to digitalize their procedures.

So there were... wherever the workers are supposed to do the fields. We had a workflow builder. They would manually create these workflows in our system. Well, we've now integrated AI into our system, so a customer can either just upload their existing procedures, they might be on PDFs or some other form of documents.

And the LLM can understand what the procedures are supposed to be and the LLM can create in a couple of seconds what would have taken a human an hour or two to do in our system of creating these digital workflows, because the LLM can understand the procedures and understands how our system works and is able to make those workflows. 

And we're also able to generate procedures from scratch, so if a customer doesn't have a procedure, they could ask our AI, you know, please create a procedure to do a weld inspection or how to pour concrete or whatever it might be, and our system will be able to automatically create a procedure for them.

So that's something that the data is clear what's coming in, just to use the framework I talked about before, the data is the customer's own procedures, or if they don't have a procedure, it's publicly available information about work types and work procedures.

Two, it is saving time. It's taking something that would have taken a couple of hours before and doing it in seconds, and it's integrated into our existing product. So it's just making our product easier to use for customers.

AI is helping. And I've seen lots of examples like that across industries. 

Tim Butara: So it improves both the user experience as well as the performance and efficiency. 

Matthew Kleiman: Exactly. You get that double on both sides. It's, it's making life better for the user, but also making the system more valuable to the user. 

Tim Butara: Win win.

I mean, that's what you want from an efficient system, right? 

Matthew Kleiman: Exactly. Exactly. And I think we're going to see more and more use cases in all different companies. And that's actually what I'm most excited about is not so much the blue sky, brand new AI capable AI applications, but how are existing companies integrating AI into their existing products to make them much, much better.

And I think that's where the initial models fall. Major value is going to be created in the SaaS software space, not so much new applications, but the companies that can successfully make their current applications much much better.

Tim Butara: Man, you are such an expert at perfectly leading the conversation to the next question that I was going to ask. So I was going to ask you Matt, what have you seen to be some of the most important benefits and use cases of AI, LLM tools, especially in the industry that we've been discussing today, so construction, maintenance and the like?

Matthew Kleiman: Yeah, there's two sides of it and indirectly touched on a little bit in the course of the last few minutes, but one is just making sense of messy data. So the construction industry is perhaps the worst offender of any that I've been involved with and having data that's either messy or non existent and the fact that LLMs can look at these data sets, can integrate different messy data sets together and be trained to make sense of it and to understand what's happening is hugely valuable, both compared to other older dive analytics tools, but just compared to what humans can do themselves, there's a statistics out there, I forgot who put this out, but basically someone said more than 90% of not, they said this, they found this more than 90% of data generated in a construction site goes completely unused by all the different players on the site because the data is just not ingestible into most systems today.

And if AI can ingest that data, make it usable. That's going to go a long way to making construction safer and more productive. The second piece is using this data to generate procedures like we are, which is our space, but specifications, designs, architectural drawings, work plans, a lot of these are largely standardized, but all have slightly different twists to them depending on the specific project and an LLM with generative AI tools can understand what the standards are and also understand what's different about a particular project.

It could help accelerate the whole process of, of planning and managing projects much better than current systems today. 

Tim Butara: So there's definitely still a lot of potential in AI and LLM innovation and getting the best possible use out of your data. 

Matthew Kleiman: Huge. We're really just scratching the surface because we're just seeing, the first use cases of AI that you saw last year were largely chatbots.

So I'm going to chat with my data repository and that's fine, but that was really. The easiest thing a company could do is put their own layer over something like ChatGPT and make it slightly customized for their use case with their branding. That was fine, but that's not transformational change for anybody.

That's, you know, that's a chat bot and the chat bot's fine, but you're just now seeing, I think in my company is one of them, but there are many others that are taking, actually taking these. LLMs and other generative AI tools and turning them into real products, real capabilities that are going to be transformational and we are just scratching the surface.

So I am incredibly bullish on the space. I think AI is an area where it has graduated from being a splashy technology to something that's very real and will permeate the ecosystem. 

Tim Butara: Matt, as I said in the beginning, I already immensely enjoyed our first conversation, but this one might have been even more enjoyable.

Thank you so much. All right. Just before we wrap it up, if for some, to me, unexplainable reason, our listeners don't decide to go check out the previous episode and want to contact you, where can they do that? 

Matthew Kleiman: That's their best places on LinkedIn. I'm I'm active there. So Matthew Kleinman at Cumulus on LinkedIn, send me, send me a direct message.

I, I respond to everybody that's not a spam bot. So, so please do that. But also feel free to contact me directly by email. Which is Matt, M A T T, at CumulusDS, Cumulus, as in Cumulus Digital Systems, so CumulusDS. com. And, happy to respond there as well. 

Tim Butara: Thanks again so much, Matt, and it's been a pleasure.

Matthew Kleiman: Thanks, Tim. Great speaking with you again. 

Tim Butara: Well, to our listeners, that's all for this episode. Have a great day, everyone. And stay safe. 

Thanks for tuning in. If you'd like to check out our other episodes, you can find all of them at, as well as on all the most popular podcasting platforms. Make sure to subscribe so you don't miss any new episodes and don't forget to share the podcast with your friends and colleagues.

Listen offline