The Human Element of AI-Driven Transformation with Steve Lucas, CEO at Boomi

Steve Lucas 

The entire world is going to pivot from deterministic processes to non-deterministic agentic process. That's what's gonna happen. I'd say probably seven outta 10 processes that have ever been built will be augmented, rewritten, enhanced with AI agents to be more non-deterministic or agenda. If you're that business that believes that you can operate successfully.

Without Ai agentic or non-deterministic agentic processes, you are wrong. You will be the blockbuster video of the modern era. We humans have to take control of this situation, and what I mean is ai. It's wonderful that Sam and his Band of Mary, women and men as well as Anthropic and you know, a couple other companies are marching towards a GI.

Well, that doesn't mean that we have to live in a world defined by Sam Altman. In fact. I refuse to live in a world defined by a dude named Sam. 

Richie Cotton

Welcome to Data Framed. This is Richie. In the last few months, a lot of CEOs have given public statements about transforming their company to be AI first, as cutting edge as that sounds.

Today's guests started that journey years ago, and they've already written a book about it. Namely Digital Impact. I want to find out what he learned from previous data framed episodes on organizational transformation. A recurring theme has been that buying technology is the easy part. The tricky parts are changing your processes and retraining your staff.

So today we're gonna learn about the human element of AI transformation, as well as the practicalities of building enterprise ai. Our guest is Steve Lucas, the CEO at Boomi, and former CEO at Marketo. Boomi, he's building an integration platform as a service business, which is a fancy way of saying it's software connecting all your enterprise systems together.

Coincidentally, this is an essential step for building enterprise AI agents. With so many executives blowing hot air about ai, I'm really excited to speak to someone who's actually transforming their business with it. Let's hear Steve's story. 

Hi Steve. Welcome to the show. 

Steve Lucas

Thanks, Richard. It's good to see you and glad to be here.

Richie Cotton

Great to chat with you. Now I wanna pick up on something you said in your book that it doesn't matter whether you choose SAP or Oracle, Microsoft, Google for your Enterprise Stack, the choice of technology event doesn't matter. 

Talk me through, why do you say that?

Steve Lucas

Well, none of those companies are terrible companies

That was the the main point, maybe that got lost. They're all great technologies. But to a certain degree, all of those companies want to obligate you as a company to their stack. That if, if that's a shock, like, welcome to software, that's the way it's been forever. But it, it's that kind of stack obligation that I think starts to rotate all organizations that I've ever worked with in my 30 years into a bad place.

Right. We end up. Choosing this whole like best of breed argument, right? We choose that as versus, you know, kind of like best stack. And my view is, you know, choose what's best for your business. That's. That's the, but 

Richie Cotton

Okay. I guess that's kind of good that we're not just alienating like all the major tech players all at once.

But I like the idea that, yeah, you wanna choose best of breed stuff, like pick the right software for your business. 

Steve Lucas

Yeah, I'm not gonna come out guns of blade, so they're all bad. But having worked for SAP, uh, as a, as a case in point for the better part of a decade, I appreciate the stack obligation objective on the part of the vendor, but that's, that's not best outcome for the customer.

Richie Cotton

Okay. So, uh, talk me through then, how do you pick your tech stack?

Steve Lucas

 I think you have to start with out, uh, objectives. Outcome and, and the solution. What are you trying to deliver for your customer? And that's gonna vary wildly if you're Campbell's Soup versus say, Boomie. Um, you know, those are two wildly different organizations.

We may both need an ERP system, but the, the needs are radically different. So I think. Thinking about your end customer first and the outcome followed by what is the simplest solution that you can pursue. I'm a big believer in the the kind of least. Common denominator. The smallest unit required to deliver the solution gives you maximum flexibility.

For me, it's just 30 years of seeing the tech debt, the tech obligation, the stack debt that piles up on organizations over time. That I think is just, it's both backbreaking. In soul crushing at the same time? 

Richie Cotton

Absolutely. Uh, yeah, I think, uh, there's no idea of a fun time having to deal with lots of different pieces of technology, half of which are outta date.

I agree that a simple tech stack sounds amazing, but how do you get to that? I mean, certainly like even a data camera, a relatively small company, and we've got lots of different pieces of technology in place. How do you manage all that? 

Steve Lucas

Well, if you think about it, like the way I think about the world and I have lot of conversations with people, I still to this day. I love the movie The Matrix. I know it came out in 1999, but still, because it, it, you know, at the end of the movie when Neo, the protagonist sees the world for what it is. Ones and zeros everywhere, governing things. Um, he's able to manipulate it, and I think that that's requisite for any leader, not just technology leaders today and seeing the world as it is today.

Look, there's like five major things that go into any tech stack, right? And, and people are, I'm immediately gonna regret this because I'm gonna get like a thousand notes on LinkedIn, but, you know, you have your, your infrastructure layer, your security layer. Um, if we go through the seven layer OSI model for those that are nerds in the audience, uh, which we won't, but you've got like apps, data, APIs, code, those are the big elements that go into a, into a tech stack.

And obviously there are, are more than that, but really any business, take that Campbell soup example, right? Their job is. To ultimately produce the, the, the best soup as a case in point that I'm gonna love as a consumer and do it at a, at a high profit, but also high quality margin. And that's challenging.

So you're gonna need, you know, both front office, back office, technology, you're gonna need manufacturing technology, you're gonna eat all of those things. But the hard part. Is not, can I pick SAP to run my manufacturing? That sounds like the hard part. It's not. The hard part is how do I connect all of the systems that I choose to be that Campbell Soup is a case in point and orchestrate all of those systems with seamless business processes, ensuring the data.

Is of high quality so we can make good decisions about running our business. That's hard. I mean, think about it, it's 2025. Have you walked into a company yet where they, they'll just say to you, Hey, you know what? We have all the data we need and it's perfect, and we run a frictionless business process. I mean, we've invented AI for crying out loud in the world that we live in.

We should have clean data, fluid business processes, 

and yet. We don't, 

Richie Cotton

I'm right there with you. It, it still seem like any time you want to achieve something, you need to interact with like multiple different pieces of technology. Quite often they don't talk to each other. Is there a solution to that? Like how do you work with multiple bits?

Steve Lucas

Yeah, there is. I mean, well that's partly what my, you know, and this is not meant to be kind of self-aggrandizing. That's partly why my company Boomie exists. We exist because this stuff is hard. It's hard to connect to all of the data in your company for one person. Now, multiply that times. You know, if you've got a thousand, 10,000, a hundred thousand or half a million people that work inside your organization, it's hard to just connect to the data.

Then how do you ensure that the data is accurate, timely. That's hard. How you then weave that data into business processes that have impact for your company and, and how do you make sure that they're agile, not brittle? So is. The world evolves that, that your processes evolve. You know, we live in a, an imperfect world, yet the business processes that are built in the SAP, Oracle Microsoft systems, they're very brittle, right?

I I have seen many, uh, a business and a process crater just because someone changed a field in a database and that can bring the whole system down. It doesn't. It flex, right? It just breaks and, and we need that bend not break mentality. Now, the exciting part, not to say AI too quickly, 'cause we say it, I think too much, but the reality is, is that AI in particularly agents in particular, will give or gift business processes, more of that flex where they can bend and not break and they can adapt.

Whatever, 20 years, every si or systems integrator out there has been talking about the agile business, right? I can show you how unile your business is. Let me go remove a data table, um, in terms of, you know, access to a process and everything is cratered, right? It's still very fragile and that's, that's what gets me excited about this AI world is just the, the flex that we can build into business processes.

Richie Cotton

Yeah, it's definitely, uh, a hot topic at the moment is, uh, um, so many companies just trying to automate business processes and the idea that you can use AI agents to do this, uh, it is, is just huge at the moment. But yeah, I, I can certainly see how there are a lot of very just brittle, fragile processes.

Like, okay, fine, uh, you. Customer intake form works great until the person doesn't have a surname or something, and then everything is broken. Okay. So can you talk me through, it sounds like maybe like have, uh, trying to engineer more robust processes is a good idea. Talk me through how you might want to change your processes, uh, in this way.

Steve Lucas

My number one enemy in the world of business today is expense reports. I hate everything about them. They're soul crushing. They're boring, they're tedious. How is it in 2025 that I have to do a bunch of work to just submit an expense report? And I'm using this obviously as a broad analog, the person that's submits it and it's like.

It goes off and you wait a week and then someone rejects it and you have to redo it like that. That the whole thing is just soul crushing. And if you're a company that, I mean, think about it. If you're a company that has 10,000 employees, and on average, let's be really generous and say that the average employee only spends five hours a year.

That may not seem like a lot doing expense reports, but that's 50,000 hours a year of wasted productivity, 50,000 hours for that one company. If I'm locked in the door and said, Hey, would you like 50,000 hours of productivity back? Most companies would say yes. Now multiply that times many orders of magnitude for, you know, something much more complex in manufacturing or, you know, process management, whatever.

It's. But this is exactly what AI can do, and that's what we've done here at boomie, is we've actually put in an AI agent that manages the entire expense report approval process. So there will be, there's no humans that are involved in approving an expense report anymore at my company A, because I just hate them.

But B, because that's what AI does and you say, well, how? How does AI benefit the expense report approval process? I submit. The name of one city is the reason why AI should handle expense reports, and that's Las Vegas. Las Vegas is the source for all expense report exceptions. It never fails, right? Every sales kickoff, there's some expense report you're looking at going.

How, what, how? Right? It's, and it doesn't matter. It could be the, the three person, $2,000 dinner and you're like, you gotta be kidding me. Right? But there's always exceptions right in, in this world and in a deterministic world that this is the if then else process world. If this happens, then do that. By the way, every process on the planet today is largely deterministic.

Um, you can't write enough if then elses to handle Las Vegas. It's impossible. It's not possible. But with ai you can have ai, which can reason. Now Apple might argue against that with their latest report, but, uh, I believe that AI has the ability to reason. But at a minimum, AI has this new gener, you know, generated pre-trained transformers.

It can improvise, it can say, well, you know, generally these things have happened and generally these expense reports have been approved. And generally these exceptions are okay. So I will generally approve it. The process doesn't break down or the expense report get rejected and you recycle and you waste hours.

So what we're seeing, Richie, and this is Long walk and short point, the entire world is going to pivot from. Deterministic processes to non-deterministic agentic process, that's what's gonna happen. I'd say probably seven outta 10 processes that have ever been built will be augmented rewritten, enhanced with AI agents to be more non-deterministic Orent.

And we can do that, by the way. That's something that we physically do, is we can take literally every business process you have and we can modernize it to be more age agentic. 

Richie Cotton

That's very cool. And certainly I'm with you that expenses are boring. No one wants to do them. Automated them seems a really good idea.

It does raise some mention questions though. Like, uh, you mentioned the, the idea of Las Vegas, everything's weird there. You're gonna get some funny reports. I presume you're gonna want a human to monitor those things. I mean, they probably don't. Necessarily wanna read what happened in Vegas, but uh. When do you want a human involved in these agent processes?

Steve Lucas

Well, I think Richie, 

you are giving to the heart of the matter here, which is, I don't think the discussion should, should center like societal level discussion on, is AI going to do more for us? Say. Yes. Like what it is that's going to happen. Is AI going to replace some jobs? Yes, it will. It will take time to do these things, but other jobs will emerge.

Here's what's gonna happen as we humans seed more work that. Honestly, we don't want to do anyway. Who wants to do the soul crushing, tedious work of watching every little exception to everybody. No one wants to do that, so we'll seed that work to ai. But you need then humans in the loop. You need a a what we, we call a a, like, well, you need a watch tower.

We actually have something that we've built called the control tower. It's literally. The agent control tower. So as you create agents to do things, manage and approve expense reports, watch a manufacturing process, ensure that a one of your most important customers never gets an angry, you know, letter or strongly worded from a system that it should never print it.

Whatever we seed to ai, we have a control tower that can literally watch thousands, tens of thousands, millions of agents at one time and bring humans in the loop when it matters. Is an agent hallucinating? Is it deviating off a well-worn path that you never intended? And so our agent control tower does exactly that and that there's a whole industry that's gonna get built around is being built right now around.

Humans seed more work to ai, but we need to be brought into the authorization loop. However, as you know, this is a cycle. What will happen is we'll get more comfortable with the AI handling the expense reports. We're never involved, and we'll move on to seeding even higher ground, how high that goes. I do not know if you're Sam Malman, you say, oh, everything, but it's his job to say everything because it sells 

more chat GPT. 

Richie Cotton

That's a very cool setup. I like the idea of having, uh, a control tower agent to watch the other agent. We'll maybe get into the architecture in a, uh, in more depth in a moment. But first, I also have a question rated this around responsibility. So our chief finance. Officer, no way you can like get anything past him, but it feels like an agent.

Maybe you could kind of, uh, game it a bit in order to get your expenses through. So, um, is there a level of responsibility there? Like, do you want an agent taking responsibility for, uh, expenses being passed? Uh, like if it's a sketchy, um, expense, or do you want a human to take responsibility? Like, uh, where does the buck stop?

Steve Lucas

It's interesting because if you look at the cycles of technology innovation, humans are, and ai, but humans have gone out and built things like applications, then we need to build systems to govern those things. There are literally, there's a whole industry around governance solutions, right? For data and decisions and expense management.

Here's what I'll say. There is no. CFO that will keep their job if they don't have an AI governance platform that, that you won't, you can't successfully adopt ai. Think about healthcare. What hospital would adopt ai Sanely adopt AI without a governance platform? I mean, you and I could sit here, Richie, and we could pontificate for hours.

And about all the amazing things which I truly believe AI will be able to do to help us live longer and provide better care. And we, I was at a hospital last week on Monday and they were showing us this AI system that detects a, a fall by any patient in any hospital immediately. It's a, it's fantastic technology, but you have to be able to govern it as you have ai that that at some point will produce a prescription recommendation.

But if you don't have the governance, the oversight, CFOs won't have jobs. This, this should be the number one. Priority for A CFO and A CEO frankly, is bringing in an AI governance platform before they bring in ai. 

Richie Cotton

Okay. So this seems like an important ordering of things. So, um, suppose like your CEO says, okay, we're gonna be all in on ai.

We need to run an AI transformation program. What's the, the kind of steps, you said you do AI governance before you start. Building stuff, but like, yeah. Uh, talk me through what are all the different steps in order then 

Steve Lucas

Here at Boomi, we actually did it the wrong way originally. So for my big giant software company, we started with ai.

We built the expense mainly because Steve's ranting in the background air going, I hate expensive work. Um, we've built, uh, customer success management agent, uh, we, we, we have agents that literally, I'm not exaggerating that. Produce marketing materials that like a perfectly formatted, marketing slick that is bespoke to a customer.

We no longer either, you know, through PDF, not, not that we would actually print it anymore. We don't hand out generic marketing material. You get your company's marketing materials Exactly. From Boomi with no human involved. That that's the kind of stuff that we're building. So we built all that. Very, very cool.

Then we quickly realized as we were building our own agent governance platform, wow, we're really gonna need this. Like this is a heightened, necessary thing. So for me, obviously the first thing that you wanna start with is how are we going to prepare our data and our applications to feed ai? The number one rule that has existed in software, specifically data analytics, you name it, is garbage in, garbage out.

So it really doesn't matter what AI does for you. If you can't feed it positive data and, and feed it high volumes of quality data, it won't learn to that 99.999% accuracy level that you want within your business. That's, that's number one. Data's gotta be ready. That rule doesn't go away just because we now have magic ai.

Then number two is. How will you govern it? That has to be ruled number or step number two. If it's not, you are 100% speaking from experience, doing it wrong, and this is in every industry without exception, but think about highly regulated industries, healthcare, banking, dot, dot. With ai, you would you Richie accept it if I said, Hey, I'm a retailer and I'm gonna have a customer support ai.

That generally is accurate 90% of the time in recommending a product. Sure. That, that's great. I might even get a promotion for 90% of the time. But you go to healthcare and you put in an ai That's right. 90% of the time, not acceptable. At all. Right? So that, that demarcation point is incredibly critical and varies wildly by industry.

So we have to be very cognizant of that. But so, so that's why governance kind of first right behind data, and then once you get governance, then it really is an outcome oriented conversation is where do we think the low hanging fruit? And I, I believe that AI has both a top down opportunity and let's just.

Let's use something like pharmaceutical as a case in point, right? Everybody immediately goes to something like protein folding, right? Which is like way above my head. Let's go get AI to simulate protein fold so we can a accelerate, you know, new drug discovery. Awesome. That's not what we do. What we do is bottom up ai, that that is look at every single business process in your organization.

So in that pharmaceutical organization, the expense reports is the very lame example, but nonetheless, it works as an analog. But every single business process, so billing, collections, you name it. That hospital that I was at last week, they said to me, and I quote Steve, one of our biggest challenges in our back office, the run, the business of the hospital, their number one challenges that all of these, uh, claims or handlers or the pro, the, the, uh, the payers.

They have these sophisticated ais that have become really, really good at rejecting claims. They have over a million rejected claims a year through AI alone. How does a hospital. Compete with these ridiculously well-funded insurance companies. When it comes to I, I did legitimate work. I'm submitting a legitimate claim and AI over at the payer finds some clever way to not pay for it.

You gotta fight fire with fire. So that's when. You must be in the AI world. You have to have an AI that can receive that just magically rejected claim, understand why, rework the claim and then resubmit. Humans cannot compete in this world. Not anymore. So that's a, those are high requirements, but just to get back to your point, I think it starts with data, then it gets into governance, and then my strong recommendation is look for the, the thousand flowers that can bloom across your business processes that create millions of hours of discovered productivity.

Richie Cotton

Okay. so that makes a lot of sense. It is kind of scary. The idea is gonna be like your company's bots versus another company's bots, though, kind of fighting it out. But yeah, maybe that's the, the world we're heading towards. 

Steve Lucas

It is the world, right? It's, uh, I mean it's, you know, things that you and I like, I, I would never, I, I've never even dabbled in things like day trading with stocks, right?

Because that world is largely dominated by AI now, and as that world is dominated by, so you just think it's just too sophisticated, too heady for me, but darn near every business process. Again, I think it's kind of a seven outta 10. We'll just go this way. And so if you're that business that believes that you can operate successfully.

Without or non-deterministic processes, you are. You are wrong, you will be the blockbuster video of the modern era. 

Richie Cotton

That's a, a, a pretty sort of  sharp warning there. Um, so it sounds like, uh, 'cause you mentioned that the requirements are very different from industry to industry or from use case to use case.

So is there a step before this then where you've gotta try and figure out, well, what are the suitable use cases for my business? 

Steve Lucas

Yeah, well that's the whole outcome, right? The outcomes are the use cases, the scenarios, and you know, I'm sure folks tuning in as well are gonna go, well, Steve, what about the model?

Absolutely. You wanna select the right model. I'm presuming that people will do their homework and choose the right model that's been grounded or, you know, either fine tuned or, you know, potentially grounded in your industry context or. You should do that work. Now my company does that work as well, so we can connect to the data, build the business processes, add AI to it, and also help you with grounding models for those use cases.

That's stuff that, it's fascinating because two years ago, if you and I were chatting just two, not, not even two years ago, a year ago. One year ago, when you and I were talking about AI agents in the enterprise. Probably most people listening would've kind of, you know, hiked up their pants a little bit and said, I don't know about that.

It, what is that coming? And here we are talking about AI agents, right? And it's totally normal. Like this is a very normal conversation, is where are we gonna deploy ai? But, but it would've, a year ago it felt like, like wizardry or witchcraft. How, wow. Where, where do I, how do I build an agent? We've literally created an agent building platform, not just the agent governance, but a, we've given our, our customers, 23,000 of them, the ability to create AI agents with just words, not code.

I can just describe the agent I want and our system builds it for you, and then you can. Drop it into a process. So now that we've gone from like it's witchcraft to, this is like not even slight of hand, just very straightforward. Now I think we're gonna get to the point where the vast majority of jobs, certainly in my business, we're actually doing this right now where accountants.

They won't do accounting. They will build agents that do accounting for them and so, or attorneys that will build agents that do legal work for them. In fact, in my company, we're taking a person or more, one or more people from every single department in my company, every single one, and we're gonna be retraining them on how to build AI agents with our own platform.

To, to do what they do, and then they will be the supervisor, the orchestrator thereof. I don't think it's that we, we, we don't have any intent to replace people. I just believe that AI can make a person like me superhuman, right? I can do five times the work I used to be able, or whatever it ends up being.

So I, I, uh, I'm not gonna be the guy that sits around and pulls the Sam Altman and prognosticate. The AI's gonna take everybody's jobs. I do think for the next five years we'll see this productivity renaissance that really is gonna make us super human. Beyond that, will there be some jobs Yes. That, you know, change or go away or replace, but I'm, I'm not a believer that AI just becomes this magic, super intelligence in the next 24 months.

Richie Cotton

Okay. So, yeah, very much, uh, grand in sort of practical, uh, improvement. It does sound like you've got a fairly sophisticated agent program going on at Boomie. Uh, I'm curious as to like what order you, you decided to build things out. Like, uh, you mentioned like, uh, a lot of the operations jobs, they've now got agents, like one of the first agent use cases you started with.

Steve Lucas

Well, that was the one I called out earlier, so it was like expense reports, customer success and support being two examples. Every company on the planet has done the same thing, I believe, which is. This is interesting. Let's try something right. And we've gone through and tried and learned, but then immediately you bounce back to, alright, how do I scale this across my whole company?

It's kind of like useful in a narrow sense, but exactly what you asked. How do I supervise it? How do I bring a human in the loop? How do I govern it? And so that's what we bounced back to and that's like two years ago we started building this. A agent control tower. Um, but I think the use case is, my, my recommendation for everybody with an earshot of this discussion is pick the low hanging fruit, identify the soul crushing, uninteresting work that tedium that no one wants to be involved with anyway.

And let's move that out of your business. You people will write songs about you if you can do that. That, and, and it was interesting 'cause, you know, Richie, we had this, um. When we launched the customer success management agent, now we have physical humans that we call customer success managers. And we from when we first said, Hey, we're gonna build a CSM agent, there was, and I mean universal weeping, wailing, gnashing your teeth.

There was a freak out across the board because universally, I, I don't wanna, no, I don't wanna engage with that thing. And when we said, whoa, whoa, whoa. Hold on a second. This is to augment. The work that you do. So think about I, I said to a group of CSNs, how much time do you spend trying to figure out which customer has access to which products and which licenses they bought for those products?

How much time? Oh God, Steve, like, you know, ungodly amounts of hours per month. Okay, well we just took that to zero. Oh. O okay, I understand the outcome here. So, and, and the message was this would, this agent will make you super human. It'll make you the best version of you, you will. Be that 10 x person that you've always dreamed of being.

And when we did that, plus we brought them into the process of creating, defining, et cetera, the adoption went through the roof. But the other thing I found out was, and, and just reinforced is if you wanna lose engagement with humans really quickly beyond the, like, you know, evolve them is, make it wrong.

Give it bad data and that, that's just, you know, that's like, think about it, right? It when, when, uh, back in the two thousands when business intelligence was the thing, and we said dashboards a lot, um, the easiest way to reduce Dash dashboard or report adoption to zero is make the data wrong. And that's exactly where we are with AI, is just make its input highly variable, inaccurate, and so you gotta spend time dialing it in.

But I think the use cases are nearly infinite and, and in a minute I'm, I'll, I'll pause here and take a breath, but I'll tell you about this automotive agent that, for this automotive manufacturer that we're working on, you wanna hear about it? 

Richie Cotton

Yeah. Yeah. I'd love to hear about the automotive agent .

Steve Lucas

Tariffs seem to be a thing right now.

I don't know if you've. Heard anything, right? 

Richie Cotton

Just a little bit. 

Steve Lucas

So, you know, this automotive company sat down with us and said, well, you know, we have thousands and thousands of processes back, office processes, manufacturing problem, so processes in the plant. We have no sense in real time for when a part, a particular part is at risk for tariff impact.

And that could be anything from perhaps not tier one suppliers, but tier two and tier three suppliers. So if we see a tariff go up 60%, are we going to see not just a disruption in terms of cost, but a disruption in terms of supply, right? So costs. You can kind of manage through it because that's a cogs thing.

But what about supplier? Are we gonna see the supplier just completely shut down? How can we create agents at every single supply point in our process so that we can start to get real time alerts? So this agent could be aware of any tariff that is specific to its part that it looks after from its particular supplier in its particular country, and it shoots a flare up.

To the business leaders if there's a problem coming. Now the challenge, right? You said, well, well, can I just write code to do that? Yeah. But what if the tariffs are changing every week and how do you know which, which website do you go to for what? Tariff information and what if it's changing daily, right?

This is a highly fluid situation. It's a perfect use case for AI risk identification, highly fluid environment, and that's exactly what we're building right now. So those are the kinds of things that you can do with AI at an industry level, you know, way, way beyond expense report approvals. 

Richie Cotton

I see. that's pretty cool.

It does seem incredibly challenging then, 'cause you've got, uh, changing policy, uh, stuff changing. Well, I mean, it was. On a almost daily basis at one point, but yeah. Uh, and you've gotta change your strategy based upon that. It requires some high level thinking, tons of different data from different sources.

So this is, uh, yeah, an incredibly challenging thing to solve. It sounds like, um, there's a huge scope then in terms of the agency can build. So you can go from something that's very narrow. Mostly deterministic, uh, and especially like a software workflow with a little bit of AI in there, right through to something that is, uh, incredibly complex and almost sort of employee level capabilities.

Is it, do you have a sense of like where the sweet spot is for where you should be targeting most of your efforts? 

Steve Lucas

Well, I'm gonna do shameless plug for book here is, that's why I wrote this. I mean, so, so digital impact, right? It's called the, the human element of AI driven transformation, if you want to go get a copy, but that's exactly what I talk about in the book is the scope, the where to start.

So I interview, for example, there's industry leaders that I talk to in the book, Mark Fields, who's the former CEO of Ford. I spent a ton of time interviewing Mark Paul Cormier, the former CEO and Chairman of Red Hat, spent time with him, Vishal Zika, the former CEO of Infosys, and just talking to these industry leaders about where they see AI and how it's going to impact those specific industries.

And then I, uh, I have, there's about a dozen, um, success stories with AI that I cover in the book. Anything that ranges from. Uh, uh, companies like, uh, Tony's Chaco only is a case in point to, um, the Australian Red Cross is a great case in point. If you remember a couple years back, the, the, the wildfires that blew through Australia, they're just ravaged massive portions of the country.

The Australian Red Cross prior to those wildfires. Were, they were handling, I, I think like 30 cases a day, right? Well, when there's no emergency, there's not that, you know, as much to do. They had to scale up to 300,000 cases in 24 hours. And how they tackled that, uh, what an incredible story. American Cancer Society.

I tell a story about how they're using data insight and uh, and, and automation to just literally arrange rides for elderly patients to go get their care. As it turns out, when you think about getting chemotherapy for somebody that's over the age of 80. Their biggest challenge is actually not insurance.

Their biggest challenge is how do I get the care? How do I physically get a ride to get there? And, uh, and so what a brilliant, you know, and by the way, lifesaving life impacting story. So the, the book covers literally dozens and dozens of both real world scenarios as well as, you know, I'm talking to industry leaders that predict where this is all going.

Richie Cotton

I do love the, the stories where it's like, oh wow, weather, like a real world impact. That's just. Ben Fit Society, like helping old people get to their, get to the hospital, uh, helping deal with wildfire. This is why we do this, right? It's not just like, oh, we're, we're trying to make, I mean, make more money is very nice, but, uh, having a good impact on the world is also, uh, wonderful.

Steve Lucas

Richie, you said it perfectly 'cause like the thing that we all have to think about. I've, I've been in software a long time, 30 plus years now. For 30 years, the message has been largely the same, which is, hey, if you just use this thing, all of your problems will go away. Your productivity will go through the roof.

It'll be amazing. Well, here we are with AI now, but the real question is, does do any of us, when we go to work, do we really feel. More productive or do we feel burdened by technology? The, oh my God, I gotta log in and go, you know, click on 17 things that I didn't want to anyway. I, I feel more burdened by technology today than I do enhanced, and I hope, and I believe that AI this time around, and that's partly what's in the book as well as.

Let's get it right. Let's build software for humans. Let's make the, let's take away the soul crushing work and give back more time to humans so that we can do the things that we are extraordinarily good at, that we can imagine that we can see past what the, the corpus of knowledge allows for. And to maybe put the viewer or listeners' mind at ease here.

The thing right now that AI is good at is working within the corpus of knowledge. It's extraordinarily good at that. But what it is not yet good at now, I'm not gonna say never, is, is, is looking beyond that corpus of knowledge. And that's where I think humanity needs to go is beyond the, not to get like star trekky, but to the next frontier.

And let's imagine, and that's where I hope we are, but with boy, we've gotta get it right this time. And it comes back to those fundamentals we talked about. 

Richie Cotton

Yeah, definitely. I agree. We use so much technology on a day-to-day basis. Some of it's joyful, but there's so much bad software that it's just a, an absolute grind to use.

I know we've got a few builders in the audience. Please do, uh, builders some nice software, but, um, you mentioned that, uh, you. Things should be built for humans. And like the subtype of your book is about the human element of, uh, AI driven transformation. So let's talk about the humans then. So, uh, suppose you do want to make use of all this AI stuff.

What do the humans have to do? So who's involved in this? Who runs these AI transformation programs? 

Steve Lucas

Well, I, I, you know, I will say, and I'll, I will answer that question directly. I think it's worth me saying as well. We humans have to take. Control of this situation. And what I mean is ai, it's wonderful that Sam and his band of marry women and men as well as philanthropic and you know, a couple other companies are marching towards a GI.

Well, that doesn't mean that we have to live in a world defined by Sam Altman. In fact. I refuse to live in a world defined by a dude named Sam, and I think all of us have got to put our foots down and say, no, it's not gonna work that way. So that's why we built this agent control tower to start, and we're thinking at a much more grand scale around AI orchestration.

We humans, this is our world. We should not seed it to an intelligence larger than us. And we end up what, you know, weaving baskets. And that's pretty much it. I think there has to be a better vision for the world than Sam Malton and makes a lot of money. And then we all lose our jobs. First and foremost, we all have to put our foot down with our legislators as well as within our companies that we are going to control this.

We're not going to seed control to some self-driving enterprise entity. So sorry for the ramp, but that's number one. Then number two is we all have to sit down and decide within our organizations what we want AI to do. What is that waterline and what are we prepared to seed and why? And how will we measure the ROI and what are we not prepared to seed to AI and why?

There's just not enough of those conversations going on right now. So let's have those conversations. So if you're a CEO and you're listening and you're not. Driving those conversations that you haven't even, you don't even have a piece of paper or you've written down or, right. These are the things that are probably good for AI to do and these are the things that are not good for AI to do in my business.

Then shame on you. Get your act together and pull that stuff, you know, onto a piece of paper. So, I, I just think that right now there are some very important conversations, both at a societal. Kind of macro level that we must have and all of us have to participate in. 'cause I, for one, will not sit idly by and watch Sam decide what the world's gonna look like.

But at the same time, we have to do that. It's kinda like, think globally, act locally. We have to do that within our company level as well, which comes back to if your company doesn't have the governance platform you're already making, uh, you're already down the wrong path. 

Richie Cotton

I love that, yeah, you've gotta have some limits on what you.

What do you not want to do with ai? I'm curious, do you have limits like that? Uh, Boomi, uh, are there any things you said, okay, we're not gonna use AI for this, uh, Boomi. 

Steve Lucas

Yeah. Um, there's a lot of things that we do and do not have, you know, our intent, uh, to do at boomie. So first of all, when we think about the easy stuff is like, are we going to have AI run our HR team?

No, no, as far as I'm concerned. Now, are there productivity enhancing elements? Sure. But, but people manage and lead with people. That's the way this works. Um, and until I see an AI that demonstrates true empathy and understanding, no, and the, the easy way to think about it is go ask an employee. Anyone, just, just go, go to work and or one of your colleagues and say, would you be comfortable with AI conducting your annual compensation review?

Would you. 

Richie Cotton

That's pretty brutal. 

Yeah. Maybe, maybe not. 

Steve Lucas

Now. I'm sure there's somebody that's gonna find the exception and go, Hey Steve, I think AI could be more impartial than a human. Maybe. Maybe. But I'm still not gonna have AI do my comp review. That is not happening. So there's that. Right? So I think the, the, so for me.

People leading people is an area where I do not believe AI should go right now. I think, you know, we talk about AI writing code. I absolutely think that AI should build code, but the real question is. Where does the innovation come from? Ai, again, working within the corpus of knowledge that it has, right?

It, it, it has data and synthetic data to go model. It's, its recommendations on, but I can tell you even the most sophisticated AI models that we've laid our hands on, they are not coming up with the innovative, beyond the frontier ideas that my engineers do today. They will continue to do that for a long, long time.

Again, no matter what Sam says. Um, so there is that. So that's the second thing. So I think people, leadership, obviously the, the innovation concepts, all of that, and there's a long list of things that we've drawn that, that we believe AI should, should currently not do. So, you know, we'll see where this game goes, but I do think that, you know, for the next.

Three to five years. I think what I'm arguing will largely be realistic and kind of manageable within the world of software. And I, I, I realize again, if you're like the ai, you know, dude or do that or whatever, that, you're gonna argue against this, but businesses right now need to focus on the most practical rate of return AI scenarios out there.

If we don't and we try a bunch of science projects and it fails, it sets it back years. That's all. 

Richie Cotton

Okay. Yeah, that seems very practical advice and certainly, um, it may be fun to these sort of science projects and you, if you're a research organization, probably go for it. But in general, yeah, I would, uh, stick to, uh, what's sort of gonna give you a high chance of getting you a good return on investment.

Cool. So, uh, one thing, uh, that's kind of the tricky part is you've gotta change all your processes to incorporate AI in some way. How does that, uh, act of process re-engineering work? How do you go about it? 

Steve Lucas

Well, most organizations today have some fairly sophisticated software process. Their, you know, process discovery technology.

They've got a lot of their processes mapped. And now if you don't, then you probably should go through a process, discovery process, mapping process, uh, you know, kind of process. Once you've done that. Then conveniently, and here's where AI can come in. We have technology that can optimize those processes for you.

We, out of our 23,000 plus customers that we have, Boomi as a company, we have almost half a billion, whether it be unique business processes that we have, ma. Modeled over time. So we have a intergalactically large database that we've put the processes into a vectorized model and now we can interrogate and understand and AI can talk to it.

That's awesome. But that process optimization, what you're looking for, I'll give you a perfect example. So if you have a process that has, uh, is a deterministic, we'll go back to what we talked about at the beginning, a deterministic process that has 2000 steps in it. Our model can look at that and say, eh.

I could probably skinny that down from 2000 steps to 20. The reason we, and how we could do that is if you create agents for these five or six things, that kind of the bulk of the process, now you have maybe 15 deterministic steps and five agents, right? So that kind of process transformation is, is the kind of discussion and exploration that we can have with any, any customer, because again.

What doesn't change is you're doing integration, you're doing automation of your business. You're just doing it with more intelligent elements in the process. So that's, that's kind of how I think about it. But I think process discovery, and then there's process remapping. There's one company we're working with right now where they have a, an antiquated billing system, and it's probably 25 years old, is my, you know, implemented in 2000.

And they're looking to modernize that billing process without ripping out the guts of it. So it's how do we bring agents that can be invoked at certain pieces and parts of the process? So there's, we've, we've integrated Boomi and there's trigger points in the process that we shoot a flare up, an agent does some work, and then we.

The output from that agent back into the billing process seems fairly straightforward. But when you're dealing with code that was written 25 years ago, that's the hard part. That's what Boomi excels at, is we can work in a hybrid environment where you've got an on-premise old school billing system, you've got a modern cloud interface for your customers, but you want to have, uh, you want to identify, to use that buzzword, the process and make it significantly more modern.

That's. That's what we do. 

Richie Cotton

That sounds fantastic. There was one part that kind of scared me though. Did you say you have half a billion processes? Like the idea of writing down like one process, that seems like a lot of effort to me getting all your processes documented at scale. How does that work? Well, 

Steve Lucas

Well, ironically, we wrote an agent for that.

So, and now to be clear, we've been storing metadata about, so with 23,000 customers, it's really not that hard to get to half a billion processes. We've been storing metadata about those processes. We don't actually store the customer, you know, the data within the process you, that, that's not something that our customers would allow us to do, and rightfully so, but the metadata about the processes those go into have gone in for over a decade from, from our company into a database that we've.

Had and managed and curated, and it's helped us deliver better solutions, which is a good idea. And then along comes, you know, generate pre-train transformers, large language models, and we then started to move that data into more of a vectorized database. And then that gives you the ability to then start to have the AI refer to it as well as.

Fine tune the AI that that we use. And then we said, all right, now the next thing we're gonna do is we're actually gonna build an agent that documents all of these processes. So we literally have an agent where you can say, Hey, I just built a business process, document the whole thing, everything about it for me.

And over time we've refined this agent, it, it's one of our most popular ones. So here, here's a case in point. So when we launched it, we actually launched a number of AI agents about a year. Almost a year and a half ago, not quite, but almost. And those AI agents, they help our customers do things like document processes, even build processes.

You can literally say to one of our agents, Hey, I need you to build an on an employee onboarding process for me. And it just. Does it because it, it knows the systems that you have and the data and blah, blah, blah. So we have that. We have another process that monitors all of your data for PII, personal or private information infraction.

So if you're moving data from France to the US when you shouldn't, it kind of, you know, lights up a little dashboard and says not good. And, uh, and so all of those existed, uh, or, or so we launched a year and a half ago. We have about 40,000. Of those AI agents that are in daily active use in our customer base now.

So our customers are actively using these agents to document processes, build processes, monitor data. And that has exceeded my expectations. So these are not cutesy little, you know, co-pilots that go, Hey, you might want to, you got a spelling error. These are real hardcore heavy lift agents and, and we have a massive, uh, uh, deployed base of those agents.

Now it's, um, it, it adds up very quickly to this flywheel where you're starting to get more insight than the agents get better and they're, there's reinforcement learning with the agents. It's, um. It's quite the flywheel. 

Richie Cotton

Absolutely. That does sounds fantastic. But 40, so 40,000 agents. This is in production just with your customers.

Can you talk me through like what do you think the ratio of like work being done by humans to work being done by agents is at the moment? 

Steve Lucas

Well, I mean, just even a year ago, I would've said that it was more pre, pre the launch of these agents, let alone our agent builder platform. I would've said 100% of the work in Boothies being done by, by humans a year ago.

And I'd say today. It's probably 80% of the work being done by humans, but 20% by ai. Now, interestingly, there's work that was never done by humans. Let's take documentation. Look, we have the ability for people to document processes, but again, that's soul crushing work. Who writes stuff down? I'm, I'm kind of joking a little bit, but not really.

Right? It's like. And if you do document a process, it's like seven words. It's like, yes, it monitors, you know, uh, you know, beer production done. It's just, you know, the, the documentation was never thorough in detail. The documentation that is done by our agents today, first of all, it's done. Second of all, it's comprehensive.

And third, it's compliant. Right? You know, our organizations and our customers have, they work in, in high compliant require compliance requirement industries. And so even just giving them that capability is, is amazing. But I'd say probably a. In a year, we've gone from zero to 20% of the work done by ai.

Richie Cotton

That's very impressive. And I agree on documentation. I hate writing things down. It's like, I don't think many humans do enjoy it. Uh, so yeah, certainly a great candidate for outsourcing stuff. Uh, what's the end game for this then? Like, how much of your business do you think you can have running, uh, using agents?

Steve Lucas

80% of the work should be done by ai. 80%. Now, that takes time. The the thing, the one thing I will caveat that with is. We, we wanna make, you know, anyone that uses Boomi. Ridiculously productive in terms of integrating and automating their company. That's what we want to do, but we also want to gift them the ability to create their own agents to do whatever they want.

Right. And that's what we launched probably like three weeks ago, is our, IS Agent Studio that create agents without code and then govern them with the platform. But I kind of think of the analogy like, um, you know, self-driving cars and I, you know, I know how people can kind of get. You know, uh, a little fussy about how they feel about Tesla.

So let's not use Tesla as the analog, but when you observe self-driving cars, it's not like we got into a Tesla three years ago and it just drove all around the city, right? It, it kind of, originally we, if you remember it, have like adaptive cruise control. You could get on the highway and it kind of slow down, speed up, and then it kept you in the lane automatically.

And you know, eventually now there's self-driving cars that will take the dramatic. Turn left across oncoming traffick or Right, if you're in the uk. But we do that. And so we've had these incremental steps where ai, the ai and the, the, the cars, it's gotten smarter. You can go to San Francisco now and get your Waymo app and click a button and there's nobody in the car and it comes and picks you up and drives you wherever.

And so we're, we're, I think with Boomi, we have that similar vision, right? Which is, you will always use this. Service like, like Waymo for transportation, you'll use Boonie to do integration and automation. But we would love for the, you know, the Boonie cars, so to speak, to come pick you up and take you on an amazing drive and make it awesome.

Richie Cotton

Nice. So it does seem like, uh, you can really push this, uh, very long way. And, but the truckers do it in stages rather than like, don't try and jump to your, your own business has gone onto the self-driving car straight away. 

Steve Lucas

That's the thing. So I think maybe if I rewrote the book, I'd probably title self-driving Enterprise, because we will get there, but it will be many moons.

Richie Cotton

Nice. Something to look forward to. Um, alright, so, uh, I'm curious, do you have any final advice on just how you can go about like building agents into your enterprise workflows then? 

Steve Lucas

Beyond the get your data right, drive the governance, um, which we've talked about many times is you gotta, you just gotta start now, right?

It's like, I mean, if you're not, the safest thing I could possibly say in this conversation, Richie, is if you are not proactively building agents, transforming your processes into a gentech. Processes, you will be the next blockbuster video. You will be that and, and whereas we all have that blockbuster cautionary tale.

You know, Netflix got the internet right, blah, blah, blah. There will be a million blockbuster videos. There will just be a million companies that just cannot compete with largely AI driven orent processes that the company that can go from. Hey, I'm automated. Yay for me, which there's still a lot of companies that are not doing well with automation, but the companies that can go from I am, you know, I'm semi or automated to 50% of my processes are driven through a gentech, you know, process.

They will win. But you've gotta do it quickly. And if you don't do it now, I can promise you, you will be Blockbuster video. 

Richie Cotton

Yeah, there's a big gap there between my business has gone completely folded and I'm making a. A large amount of money. Yeah. Everyone wants to be Netflix rather than Blockbuster.

So just finally, I always want recommendations for people to follow. Uh, so whose work are you most excited about at the moment? Who should I be looking into? 

Steve Lucas

Well, shoot, you know, I'm gonna contradict myself and say like, like if you're. If you are not daily staying in tune with, you know, across primarily three entities, right?

Philanthropic and you know, claw in the engine, that they have open AI with chat, GPT as well as operator and many of the other things that they have. But Google with Gemini, I think Google kind of went from. Last technologically, uh, two years ago to really a, a leader in, in this space in many fronts. Like you gotta stay on top of what they're doing.

So I would really kind of like, you've gotta be dialed into news feeds, but then I will say, you gotta stay dialed in there, but take what they say with a grain of salt. There's a, you know, their, their job is to kind of overhype this, 'cause it, it gets headlines, but it, it is one of those things that I, I have spent more time in the past two years.

Relearning everything I know about software than I did in the previous 25. 

Richie Cotton

Absolutely. I mean, there's so much exciting stuff coming out all the time. So yeah, certainly following open ai, andro, Google, they, they just journey out like lots of interesting things. Cool. Alright, wonderful. Uh, thank you so much for your time, Steve.

Steve Lucas

That was fantastic. It was a pleasure. It was great to talk to you.