Alteryx CEO: Why AI Agents Need Business Rules

Hi, everyone. Welcome back here to Techstrong TV. I'm really happy to have my next guest on.

It's Andy MacMillan, CEO of Alteryx. I was talking to Andy before we started today. The last time I think we interviewed Andy was in person at the Alteryx user conference.

That was in Orlando, I want to say it had to be a year ago. No, Andy? It's about right.

Yeah. Yeah. And that was a great show.

What I remember is the buzz, the energy at that show was sky high. Good stuff. It was quite a show.

But I hope you've been well over the past year and still doing what you're doing here. Obviously, you're still CEO at Alteryx, but it's been a heck of a year, hasn't it? It has.

It's been busy. I tell folks you work in tech because you like change and the dynamism of what's going on, and that's definitely the world we're living in now. And so yeah, having a good time doing what we do.

It is a story. I tell people this, and it's a shame, I need to write an article or a report on this. My belief is, Andy, is that AI, for all the talk about AI doing the work for us, I find myself working two X, three X times harder.

I'm doing more stuff with AI, right? Yeah. I'm much more productive.

There's a lot more output. But it doesn't just happen without me also working harder, right? So I don't know if it's that way for you or not, but I feel like this whole AI thing has us working longer, harder, but we're also two X, three X what we were before in many cases.

Well, I think it's rewarding. You get further, right? Right.

So you have this problem, and I think before maybe I would've gotten to a point where I'm like, "Oh, there's no more juice left to squeeze out of this. " And now you just go, "Oh, great. " And so you do, right?

I can keep doing my thing. Yeah. And it's like, well, now I'm just going to keep going because I can- Right ...

build something or build an analysis. And so I definitely find that's true, where I just get into something, and I get so much further, so much faster and so rewarding, I do more of it. So- I'm just able to do more.

But then I look back on it and say, "Holy mackerel. I'm working more hours than I was before even. " But anyway, we're going to talk about this in our discussion today.

But Andy, I wanted to give the folks at home, that sounds so old-fashioned, the folks at home a chance to understand a little bit about you and how you came to be CEO at Alteryx. Sure. Yeah.

So quickly, my background, I've been doing this a while now. This is the fourth company in software that I've been the CEO of. I actually started my career as a software developer a very long time ago back in the Java days, doing enterprise Java development.

com. That was my early career. I ended up going to a web content management company, so kind of continuing that path, but into a product area, into product management.

That company was bought by Oracle. com, and then I was the product group COO for Salesforce globally, and then thought running tech companies sounded like a fun job. So my last company prior to Alteryx was a company called UserTesting, which was a company people get feedback on the stuff they build.

We took that public, we sold that. I came over to Alteryx, in large part because I had started using AI to build things, and what I found was the things I was building needed a lot of business context with the data that we were building. And that's really what Alteryx has done for a long time.

So if any of your listeners aren't familiar with Alteryx or your viewers, it's really a set of capabilities designed to let a business professional bring to bear technologies on their data to solve a problem. So technologies like Python and SQL and machine learning and geospatial stuff, are really valuable if you're in the office of finance or supply chain or rev ops, but you don't want to be a Python developer. You want to take what you know about rev ops and build something.

And so it's a drag-and-drop, easy-to-use tool that lets people go get their data from spreadsheets or big data platforms and use their data to solve a problem, either one time or you can actually build automations. So you can say, "Oh, every week I have to compile the sales data and I need these three sources and I know a lot about it. " And in the olden days, that used to maybe show up in a Tableau dashboard.

Now that compiled data's showing up in ChatGPT or Claude or something like that. Excellent. Love it.

And thank you for that. You know what, while we're here, though, just let me get it out of the way. What's the website for people who maybe want to go dive in deeper on that?

com, is the easiest way to learn about Alteryx. And then I'm APMacMillan on Twitter and LinkedIn, all those things. So you can also find me there.

I'm always happy to chat with folks about this. APMacMillan. Excellent, man.

Thank you. So Andy, we mentioned AI, we talked about how it's made us all more invigorated to do more. It's been a heck of a year and more, but when we talk about AI now, of course, I think the last time I met with you, we were probably talking about generative AI.

Now, if it don't have agents, it don't mean a thing, right? Right. And so it's all about agentic AI.

But what we're finding is this isn't just set it and forget it, right? There's governance, there's guardrails, there's processes that need to be put in place. There's how do humans interact with the agents, how do agents interact with the agents, all of these things.

Talk to me about what you've learned and what Alteryx is doing to help people here. Well, I think a lot of what we're learning and spending time on is how do we take this amazing promise, right? This universe where an agent can help me close the books or price a deal, or figure out the best way to ship something and apply it to my business in a more autonomous way.

It doesn't mean it's fully autonomous, but let's just say for the sake of argument, a more autonomous way, right? Where it can help me make decisions and solve problems. And I think what we're seeing is that promise is running into some real challenges in enterprises and companies.

And I think the challenge is some of the difference between what it looks like to use those kind of capabilities personally versus what it means to use them at work. And so personally, if I use Claude or Gemini to help me plan a vacation or something, it makes all kinds of assumptions in the background, right? On what makes a good vacation and what kind of places I might want to stay at and what I might want to do based on what other people did.

And those assumptions feel like magic, right? It just sort of comes up with a really interesting itinerary, and you're like, "Oh, that sounds great. Maybe change this," and away you go.

And I think those same kinds of assumptions become problems at work. " Well, it's going to make a bunch of assumptions. And those assumptions about how I run my business might be a reasonable way to do something like a revenue recognition calculation.

It's just not the actual way my company does a revenue recognition calculation. And the difference between that assumption and how I actually do it is me not passing an audit, me not giving the right answer. I'll give you another example.

" It needs to calculate our sales commission policy and give me the right answer. And so I know that feels a little obvious and also a little bit nuanced, but that's the difference, right? I can't ask an agent multiple times what my sales commission will be and get a different answer each time.

All of them reasonable assumptions, just none of them the right answer. That's when I can't trust the agent anymore. " That's not how businesses work.

No. The whole deterministic, non-deterministic thing as well. But I do feel like we've made a lot of progress in saying, "Okay, we know this is an inherent factor- Yeah ...

" And I think that's what people are looking for, right? They're looking for the rule sets. They're looking for the guardrails.

They're looking for the governance. And I don't mean security or cybersecurity. I mean- Yeah ...

rules of the road, how these things work. How do I make it productive for my business, right? And Andy, I think this is one of the big things we're seeing in real life out there too.

It's great for you, it's great for me to go use AI. We make things, we build anything we want. It's all good.

Maybe your productivity's gone up, my productivity's gone up. But when we want to look at an organization, an enterprise, and say, "Where's the return on investment? " It can't just be this willy-nilly that you're playing, I'm playing, and we're all having fun thinking we're more productive, and maybe we are more productive.

You need structure. This is business, right? Yeah.

I think it's a little of both. So maybe this is two opposing concepts that I actually believe are both true. One is, people are more productive with AI, and so I would argue if you look at something like email, I don't know how to tally up the value of email to my company, but I'm positive I don't want to turn it off.

And so I think there'll be some aspect of that that exists with AI, which is like you would be foolish to be a CEO running a company where you haven't given all of your folks AI tools to work with that are more productive. The flip side is, if I'm going to make the level of investment we're all expecting and the level of change to our company, you do expect it to show up on your balance sheet somewhere as real ROI. And I think that's where we get to this level of autonomy and speed across the company, not just individual productivity.

I can't just make the individual people more productive. I have to make the processes I'm running more productive. That either drives down risk, it drives down cost, it drives up revenue, but some way of running my business differently.

And I think we get back to that, how do I get to a place where I trust the AI to do that? We talk a lot in Alteryx about a framework we've developed, it's not specific to Alteryx, that we call VORA, and that's this idea of in order for me to trust an agentic process in any kind of autonomous way, the process has to be based on business rules that are visible, understandable, repeatable, and auditable. It has to be visible to me.

" And I have to be able to have people in the business be able to open it up and understand it, not just my software developers. " I have to be able to understand it. Right?

" So it has to be understandable, some kind of format they can read it. It has to be repeatable, got to be able to have it happen the same way every time back to our point on declarative. But it also has to be auditable.

" Can I go back and understand what happened? And I think if we do those things, we can start to trust the agents to do the work. And I would add, if we can do it in a way where the business can own that logic over time.

Because the other thing you're seeing a lot about is people are vibe coding something that solves part of a problem, right? They're taking that business logic and they're coding it into something. But I have to be able to change it.

Things change, right? " Well, I don't know, California, for example, just passed a new bit of legislation, we're going to tax software differently. So who's going to update the agent to make sure it does the taxes right?

And I don't think the answer is we're going to go find somebody to go in and change the Python who doesn't understand the tax law again, and every time something changes, I got to go find this guy. " My business changes all the time. Companies acquire companies, they sell off companies, they change pricing, they roll out new commission plans.

If every time the head of sales gets on stage and says, "We're announcing a new sales spiff," all your agents break, then I can't run an agentic company. I got to be able to have my sales commission guy go, "Oh man, the head of sales just announced a new spiff. " And so I think that's where you get to the real ROI.

If a company can say, hey, the people that run the business can be in charge of these agents. They can own the decisions those agents make. They trust the agent to make the right decision.

Again, it's visible, understandable, repeatable, auditable. They can make changes when needed. I think that's where we're going in the next year, is people figuring out how to do that.

" We're going to see a head of sales say, "I have an agent that helps me put deals together. It's not a black box. It's not something I don't understand.

It's not something some kid vibe coded for me. " And that's where we'll start to go, "Oh, wow, look how many more deals we can price. " And that's what I'm starting to see people leaning into.

Me too. I will tell you one word, and you mentioned it, you said the word trust. Right?

I think that's the biggest thing standing between us and this future that you foresee. I think a lot of organizations I speak to, they don't trust their agents. I don't know another plainer way of saying it.

Yeah. They don't trust the agents. They don't trust their agents, and they also don't trust maybe what, we talked about you and I both being more productive.

Okay. " I don't think I go, "Oh, okay, great. " I think I go, "Well, how did you do that analysis?

" And we would talk about it. But a lot of our personal productivity is very sort of single player. You're doing your thing and I'm doing my thing.

" And you're going to go, "I don't know. " So we get to a universe where we're in an environment, this again, where a lot of our customers will use something like a Claude or Gemini instead of building Python, they'll go build an Alteryx workflow, which is this visual workflow. " And then we trust it because we can see it because it's visible and understandable.

I just wrote an article on this last week about we got to be able to see inside the black box. Right. Right?

So there was an old "Twilight Zone" I saw a couple of months ago on TV. It was one of those apocalyptic futures where, I guess, some nuclear war has wiped out most of humanity. There's this one village that survived because there's a guy in the village who goes to the old man in the cave to ask questions.

Is this food canned good, good? Should we plant crops here? Should we do this?

And they ask the old man in the cave. The one guy, he's the only guy who talks to the old man in the cave. " Then somebody- I haven't seen this one, and I like the show, but yeah, keep going.

So then some vigilantes come by, they claim they're in the US Army and they're reconstituting the US or whatever. And they're not. They're just looking to rip off any pockets of survivors to get whatever they can.

" And he won't. Finally, he takes them to the old man in the cave. And the old man in the cave's a computer, right?

" It's amazing. Once the people realized that it was a computer, of course, this computer doesn't know what's right, destroy it. And they destroy the computer, and the next day everybody's dead.

They ate botulism- Yeah ... canned goods or whatever. Yes.

That's kind of what we're doing here, right? Right. It's like there's an old man in the cave who told me this works, and it's good enough for me.

Right. And you're in this long process of interrogating it, right? We work with our agents, it's a back and forth, and then you get to the end with a conclusion, but nobody else has come along on the journey with you.

So it is a black box to them. They haven't had all these interactions. Then they ask you to explain it I'm just, I know what the old man in the cave told me.

That's all I ever know. Right. I can't explain it.

That's right. So how do I go back? And so we have to find this way to have a language that we're all operating in, and then I want to be able to share it.

The other thing we're seeing a lot of people spend an enormous amount of money on is paying tokens to recreate the same logic, but slightly differently. Right? " And I go, "Okay, that's really interesting.

" So I go over into Gemini, which is what I'm using, and I give it a similar prompt, maybe the same prompt. But I run through my interactions, we end up 5% off. Not only do we have different answers, I just spent all those tokens to regenerate the thing you already did.

" Again, we do a lot of this in Alteryx. It creates an Alteryx workflow that has this business logic. You and I can look at the logic and agree, "Yeah, that's right.

That's how we're calculating sales commissions today, and it takes into account this thing I'm worried about. " And now my agent calls that logic. I don't pay to recreate it.

Right? Right now, we are spending, as an industry, so many tokens on multiple people creating the same but slightly different logic, and then trying to figure out why we don't trust these agents. Like, we're just- But this is also the difference.

Yeah, I agree. This is also the difference between GPU versus inference. Yes.

Once I have that model, I should be able to reuse that model using just inference, which should be a hell of a lot less tokens. That's right. Right?

And- And then it will change over time ... charge me for that. Right.

Right? And then we change pricing, and again, somebody has to go and spend the tokens to update it, and then we should use that again. Yeah, but that's a one time.

It's not like doing it every time. Right. Reinventing the wheel.

That model both feels obvious to those of us who've been in the industry for a while, and sort of, but it's not what companies are doing. They're handing everybody an LLM and telling everybody, "Go use this to calculate everything," and they connect it straight to their big cloud data warehouse, and the AI is having to go spend a mountain of tokens- And don't trust what the other guy gave you. Right.

Right. I will spend a mountain of tokens to sort through the same data everybody else is sorting through and come up with slightly different conclusions. It's madness.

But we're working on getting out of that phase. " And I think when we get there, we'll get to your point, which is now I can build something I trust, and I think then the ROI is tremendous, right? The ability to have anybody build that kind of logic and trust it.

We got to get there. We got to get there because I think- We will. We're getting there ...

people's fuses are getting short here, right? They're spending money and they want to see returns. Anyway, hey, Andy, this was supposed to be 15 minutes.

We're over 20. But for people who want to get up on Alteryx, we gave them the website, but you guys on the road anywhere people can come catch you? Yep.

We do a regular series called Alteryx on Tour. We just wrapped up a whole bunch of them over the summer. They'll kick back up again as we get into our conference season starting next year.

We're a big global company as well. There's probably lots of Alteryx people nearby wherever folks are at. So you can always reach out to us through the website.

Again, I'm always happy to have people reach out to me directly and I'll connect you with somebody if you need some help. Our LinkedIn's been pretty active with a lot of good content on VORA and how we're seeing people do this. Again, not just as a Alteryx capability pitch, but as a framework of how to think about this problem.

So always happy to engage with people online as well. Fantastic, man. Appreciate it.

Andy MacMillan, CEO of Alteryx here on Techstrong TV. We're going to take a break. We'll be right back.