← Marketecture: Get Smart. Fast.
Loading summary
Transcript52 lines
- [00:00]
Ari
This podcast is brought to you by CloudX, the agentic platform for mobile advertising. Connect to the CloudX command line interface or MCP server and have Claude or Gemini pull reports, run experiments and automatically drive better outcomes for you. CloudX add infrastructure for the intelligence era. Head to CloudX AI Start to get started. That's CloudX AI Start. This podcast is brought to you by the Build, a new podcast from the guys behind Sincera, Michael Sullivan and Ian Myers. They built their company by figuring out clever solutions to a few important ad tech problems in our industry. And that's exactly what the show is about. Mike and Ian interview some of the smartest tech minds in the biz to hear about how they identified opportunities, solved their hardest challenges, and grew their businesses in the process. Listen to the Build with Mike o' Sullivan wherever you get your podcasts. This is Ari. I'm happy to bring you another recording from our Architecture Live event in March. This one is David Dwaran, who's the chief Product officer for Freewheel where I used to work. David is a great leader and has a lot of insights on product development and he's going to be talking about building with AI. The title is how building with AI sharply improves marketing performance. So I think you'll enjoy this one.
- [01:33]
Mike Trion
This is exciting. I'm excited. This is hard. Aries sitting in the front row here. Every time I talk about Beeswax, if Aries in the room, I'm really insecure about it. It's like you're like the pressure's on his child.
- [01:47]
Mike Sullivan
You're watching.
- [01:48]
Mike Trion
That's exactly what I said. It's like my father in law is coming over and judging everything because I've
- [01:53]
Mike Sullivan
got his daughter now he's got the shotgun.
- [01:56]
Mike Trion
He's got it. He's watching to see how I talk about his APIs.
- [02:02]
Ari
Cool.
- [02:03]
Mike Trion
So I'm excited to be here with Mike. We had a big announcement that we
- [02:07]
David Dwaran
did together
- [02:09]
Mike Trion
from us that's excited about our lean in. For those of you guys, this is, I think the first panel anyone has seen is going to mention this thing called artificial intelligence and agents. So. So we might have to do some introductory things. If you missed Jeremiah's great presentation earlier. Mike, why don't you tell us what this means for you, like for you at pmg. What did this release mean?
- [02:30]
David Dwaran
What are you most excited about in it?
- [02:32]
Mike Sullivan
Yeah, I mean for us, I mean ideally for me personally, this is super exciting time in the last, like I've been doing this for 20 years and I've been like waking up more excited to go to work because of like what we can do and how quickly we can do it. So I think over the last like six to 12 months, my ability, our ability and like this community ability within the agency to get our hands on and rapidly develop and then you know what's exciting about what we did with Freewheel is develop together kind of from both sides of the table. And we really value the Freewheel relationship and others because of like the directness to how you all work with publishers, the content that's there. But more importantly, you are a technology company at core with Comcast as well. And we now, you know, as an agency really pride ourselves as being a technology company now with media services and I think being able to work really closely around AI, how we iterate and really how we tool build on either side of that and the fact that you know, a lot of this was done incredibly quickly outside of, you know, development cycles is, is, is awesome for us.
- [03:44]
Mike Trion
Yeah, that's one of the things that I think about that's so exciting and, and I think about it from my personal perspective. Right. The speed that I can on things that I can prototype something, that I can take something from thought or an idea into reality has been incredible. I read something I think about all the time that I'm now able to take all of the ideas I've had in my backlog and just make them and realize how bad many of them actually are once they become real. But you've done some things that are actually really incredible with this and have added some value. So I'm curious, what are the things, the processes or the workflows that you've been able to transform with agents and agenc integrations into these tools that you never thought you could or you don't have to think about them again?
- [04:26]
Mike Sullivan
Yeah, I think the area we're iterating most is what I would call like our in depth CTV campaign management, the enterprise level. We're looking at brands like in Experian for example and how we operate their media in five to six platforms at a time and how we go deep and manage across those platforms and we with a like an agile technology layer like free will, buyer cloud, you know,
- [04:54]
Mike Trion
but that's what they call it now.
- [04:55]
Mike Sullivan
That's what it's called now.
- [04:57]
Mike Trion
Yeah, that's.
- [04:57]
Mike Sullivan
But within beeswax that we can take some of the processes we do as not as campaign managers but really analysts and extract them out and bring them forward. So a lot of the things we did, you know, with this MCP and Freewheel is look at the tasks that take two to three steps and oftentimes for me, even as an operator, some things, Step one sits in my downloads folder for a little bit. It's this extract of deal delivery yesterday by deal id, right? Step two might be looking at what I'm actually targeting and step three is actually doing the analysis to figure out where the missed opportunity is. And by you know, taking our hands on keyboard, you know, talking with you guys on the free will side and the engineering side and saying, hey, here's a three step process that you all might have a lot more insight into on your side because you can combine, you know, two or three APIs you all can look across like the demand side and the supply side and that becomes a, a prompt, right? A daily prompt, a check in or part of a larger workflow we can develop. As you know my, my analyst agent, they like to check deals first, then they like to check performance and how we can sequence a lot of those things. We're taking, you know, like our really, really good operators and hands on keyboard and you know, you know, tripling, quadrupling the amount of work that they can do on single campaigns and that just what it's leading to for us. It's becoming like micro segmentation. You can just, okay, rapidly iterate. I want to test this like weird new inventory source, this other source of data, this other tactic and our ability to iterate on that I think is going to be fueled by our ability to manage, optimize and extract from those things that we're actually doing.
- [06:47]
Mike Trion
What I love about that example and one, you made it really tangible. But I always, when I talk about AI, everybody has this idea of AI as cost savings, right? Like you take the agent and you've automated this workflow that in some ways is bespoke to you. Some people may not put it in their downloads folder the same way that you do. They may have some other process, but you've been able to take the mechanics of pulling that report together and doing the analysis. And to your point, now everyone's just able to do more, right? They can manage more campaigns because they're not waiting and clicking through screens and doing all of these sort of tasks that add friction to the real goal, which is analyze and understand and take action. And you've closed that route and you've opened it up to more people. One of the things that PMG has that I think is really cool is the Alli platform. That's my actual wife's name so just easy one for me to get. Can you tell me a little bit about how you're using Alli and how that's plugging in?
- [07:41]
Mike Sullivan
Yeah. So Ali at its core is developed along the the same timeline as the agency. It's 15 years old. Our CEO was an engineer. Our first two hires were engineers. And the idea of PMG really came from developing technology in parallel to media. And every step of the way, as you know, the media landscape has iterated. Ali's kind of been there to support that and accelerate that. I think one of the, you know, I came to PMG by way of Camelot, both Dallas based companies and like the number one synergy of like what we were doing is PMG just really had a handle on scaling with technology and setting everything up for an AI era. Because everything in ALI is really based in clean democratized and contextualized data. And that gives us the ability to say once you start connecting those sources and start looking at, you know, trends, themes, outliers, alerts, those are things you know, natively we were doing in Ally that are, you know, somewhat AI informed but really they're, they're tactical. Right. If this field that we're getting every day and reporting goes blank, Ally knows what that means, right? Oh, this deal ID has dropped off. Your, your pixel has dropped on conversion. Being able to accelerate, you know, some of those tasks have always been at our core now applying it. I think for the first time everything we've done in the last probably year and a half or at least in my world has been how do we turn that external, right? How do we better partner on massive amount of log file data that we can get as exhaust from freewheel, harness that and then give it to clients big, you know, again a big premise and a thing we hang our hat on as PMG and being independent is we are completely transparent non principle based media buying. We believe that, you know, our customers own that data but you know, many of them can't access it, can't control it. We look at it as our jobs as agents to be able to say hey, here's that data, here's access to it. And now that becomes a question to that data. What were my technology costs yesterday? How did my audience data CPMs change in the last week? Has that had an effect like we are, we want to build, you know, data and infrastructure to answer real marketer questions as well as in depth advanced campaign operator questions too.
- [10:01]
Mike Trion
So you hit on something earlier and it's been one of the things I thought was so cool about the partnership we've done is how rapidly we could iterate. And I think many people experience this. I know I talked to all my friends at Tech experience this. You have the way that we've traditionally built products and built capabilities, which is we have to work through the sort of engineering process in a software development lifecycle. And it's got to be able to serve a lot of different users with sort of the common thing, this other end of it, which what you're talking about, where you'll be able to build really, really quickly based on this specific workflow that you've identified that you want to automate or agentify or whatever we
- [10:40]
David Dwaran
want to call it.
- [10:40]
Mike Trion
Now, how do you balance that within pmg? I have thoughts about how we do it, but like, you still want some things that you're doing within the ALI platform that are going to serve everybody that have a deployment cycle. And we continue to use agentic coding and better processes to speed that up. But there's going to be other things off to the side that's like, this is the Mike Trion, skunk worksy thing. How do you guys as an organization manage that? And you don't have to say anything that's going to get you in trouble,
- [11:05]
Mike Sullivan
but no, yeah, I think I'm proud of how we handle that. I think a lot of what drove this is, you know, the story for me is I had an application that I wanted to develop that I knew was core to what we do was really based around deal IDs and operationalizing deal IDs, again as a background. And programmatic, it is one of the most, like, exhausting pieces of programmatic tech that just has so many falloff points. There's so many, you know, room for error, operational inefficiencies and wanted to really develop something around deal IDs and, you know, it takes a long time to communicate a business problem to engineering, to scope that work to derive value. And, you know, with the rise of, you know, again, you said agentic coding or agentic engineering versus, you know, I started vibe coding.
- [11:55]
Mike Trion
I think we've matured. I used to call it vibe coding and now the quality has gotten a lot better. That's at least why I, why I started calling it.
- [12:01]
Mike Sullivan
Yeah, and I mean, I just, I picked that up, you know, eight months ago and did it myself and just sat down and learned it. I, you know, my background 20 years ago was an engineer, but I spent most of my time on the business side.
- [12:15]
Mike Trion
And that's the thing for me I feel like I didn't have to learn anymore, right? I used to have to go, like
- [12:20]
Mike Sullivan
I said, looking up it's ways of thinking, you know, And I like, I've always thought that, you know, balancing your ability to talk to an engineer and a marketer at the same time was like, that's where my focus has been in my career, to be able to bridge that gap. And for the first time I was able to do that in the same working session at my, at my desk. And then what happened is, as this iterated and I started sharing it with my team and I was fortunate to have like a chief technology officer and a chief product officer that didn't like, push me out of the room that said, you know, oh shit, this is where we're going to go. And, you know, our chief product officer said, you know, like, listen, I don't believe that some of the code, a lot of the code we're going to write isn't going to exist in six months. Like, we have to support this. So what they built, and the first thing that they built was like, okay, you know, we need some, some rails around this because everybody in the company obviously started picking up with like AI Studio, Cursor, Anti Gravity and like doing all these things. And our team said, first thing we're going to do is Ali Labs. Like, we need a repository, we need a framework where people that develop this, that these ideas can be productionalized and in a way that takes into account our credentialing. Of course, everybody's first application, all the credentials are stored in clear text somewhere. And everybody's like, all right, shared it in the GitHub, that kind of learning and just like, yeah, GitHub. Hey, Mike, here's the process of pull requests and commits and how we think about that. And we solved that in like six months where we can now deploy this to a lab, we get a staging infrastructure in production. You start thinking about like our single sign on credentialing, how we do service accounts to, you know, 15 platforms that are wrapped up in there. And I credit my team for looking at that and saying this is something that we know we have to do. The agency itself is, you know, 35% tech and product, right? And we, they see technology in a way that enables business. And the fact that we could see how much this could empower not just our engineers, but everybody that wants to solve a problem and has the ability to do it can now do that, graduate it, and if it becomes Mike Trion's application, that's Great. Mike Trion just became a little bit more, you know, a little bit more productive. Or we did this and if it, you know, decomposes in like six months or whatever, like that's great. That's happened. We haven't, we haven't dug a huge investment into it.
- [14:41]
David Dwaran
Yeah. What I. So our philosophy, the strategy we're leaning into on this, we call it like built in.
- [14:46]
Mike Trion
Built on.
- [14:47]
David Dwaran
Right. We have certain things that are going to be core to the system that we're going to build in there. You get scale from your data, you want access to data, things that only we can do based on where the technology sits. And then how do we empower someone like you to build on? And I think we're in this phase right now where the way that people build on is through something like the MCP server. Right. We can train these agents to interact with the system or we can enable you as a human to interact into the system in plain language. Your agent may be able to interact with our system. Other systems go across them, pull data from a linear campaign and a digital one side by side, whatever you're training them to do. A year ago, I think if I had said that people would be like, you live in sci fi land where you're talking to the Star Trek computer and you're just asking it to do things. But here we are. It's so interesting to me because I'll spend time in my head sometimes as an AI skeptic where I'm like, I'm not seeing the value. I hear from the guys at the labs. And then a year later I'm just late to the party on it.
- [15:51]
Mike Trion
If you think about the future, today we've got all these things. You're building code and applications and tools rapidly. You're interacting with our system and other systems using plain language. You're interacting with code that interacts with our system by using plain language. Where do you think we are a year out from now? What's sort of next in your head that's percolating and what are the new things that are scary or you don't believe in but you're kind of ready to be wrong about?
- [16:19]
Mike Sullivan
Yeah, I think there's a couple areas that I see this impacting and hoping it impacts. I think one for us has always been, again, I think we've done ourselves a disservice in programmatic in general and training hands on keyboard. In the last probably 10 years, like everyone's been trained on a UI, right? And it's been a UI not controlled by anybody but a platform company. And that's, you know, reduced our ability to switch, to test, to integrate APIs, data, you know, everything that you need to be able to connect that. And so I think this one goal we have is I really want to accelerate that, that ability to switch, use and add on and connect and create optionality, right? Create that system where those dollars can go to, those systems can go in smaller ways, tested and that feedback be, you know, be captured and iterated upon more so. But I think it's a huge enablement tool for teams. I also think, you know, the next step behind us and like, again, the reason this is going fast right now is because you have like, you know, buy side and sell side or technology in this going together is like, I think once we put marketers in here as well, and you start turning these marketers onto it. So a part of what we're doing with Labs is like, okay, not just internally, but like, we have to let our clients be able to do this too. If we're going to let all their data sit here contextualized and accessible. If they want to build an application on top of it that works just for them, we need to support that and, you know, be able to build on it. But I think I want to be sure that we're really adding a lot of value and really capturing the things that we do well. I think we have a team of like, incredibly awesome seasoned campaign managers, analysts that have been doing this for 20 years. I want to make sure that every workflow, every playbook that's in their head is somehow captured, that it can be repeatable by six other people and just like accelerating on so many paths. But I'm really excited for the prospect. I think, like, once you start getting marketers into this and it becomes, hey, you know, some of these questions that we've just never been able to answer or access to this data that you needed to unlock. I mean, it was a year ago that I'm hammering like data engineers for like, like, please, can I get a snowflake license to go in and see what data is in there. And now we're, you know, we're using like snowflakes cortex code to be able to like read into that, develop, see what we can do and pull out of that. I think acceleration on the brand side and the marketer side, when they connect that to like, how they think about consumers and how they answer questions and build applications is going to be really, really cool.
- [18:57]
Mike Trion
So sort of like diffusion through the agency up into the marketer.
- [19:00]
Mike Sullivan
Yeah. I mean, listen, again, we're very public about George. Our CEO has been on a path to transition from a media services company with technology to being a technology company with media services. So if you want to pull our media services out of a piece, like, we still think we can accelerate your business a ton as being a media technology partner as much as we can be hands on keyboard.
- [19:22]
Mike Trion
Yeah, I love that. And I think it's this vision of, right. You're democratizing access to things. You're allowing people to shape the world that they're in and sort of create it for themselves almost in a bespoke way. You hit on something interesting too. For the AI to be successful, you sort of have to document the steps and the workflow. I think one of the value adds that people have. Right. I would fight people on workflow documentation or clear writing. And the thing that's changed is now I'm like, guys, the machine doesn't know what you're thinking unless you can clearly communicate it. It used to be important for people. Now it's really important to be a clear thinker in communicating to an agent and being able to do that.
- [19:58]
Mike Sullivan
Yeah. Document on three levels for engineering, for users who've never seen it. Just like the amount of documentation. I think the cool thing about this is like I look at everything I'm doing and I was like, man, if they destroy this in a week, the thing that will still be captured is all the knowledge that's in there.
- [20:15]
Mike Trion
Right. It gets fed into tomorrow's agent.
- [20:17]
Mike Sullivan
It's right there. And if it doesn't look like this ui, I'm putting on top of it, like, who cares? There's a lot that we're capturing and that becomes, you know, rag knowledge base and just like skills md. Right. Like these are all things that we are capturing there and those are going to become. That's what, that's what's going to make a difference in how things work in the future. I think when you're operationalizing things, so do we.
- [20:38]
Mike Trion
I know we had practiced doing questions. Do we have one minute for questions or now is the time that we take questions. We have one minute for questions. Does anyone have, like everyone's got a question.
- [20:48]
Mike Sullivan
No questions.
- [20:50]
Mike Trion
We answered everything anyone could think of. Oh, we got it.
- [20:52]
David Dwaran
Yeah.
- [20:53]
Mike Sullivan
I don't. Still don't see any questions.
- [20:55]
David Dwaran
Hardware question about my baby?
- [20:56]
Mike Sullivan
No.
- [20:57]
Ari
Just a quick clarification. So when you think about the Freewell Buyer Cloud, formerly Beeswax, is mcp the layer that's the most important for the customers, or are they using the traditional rest APIs on their own with AI on their side? And what's the mix there?
- [21:12]
Mike Trion
Yeah, that's actually really cool. What's happening? So some people, we built our MCP server, we've put the rest APIs in it, we've built a bunch of tools in IT based on things like we've worked on with Mike or other people. Some people had already been building their own MCP servers on top of the rest APIs. And that's fine, it's not super hard. One of the things we've realized that's interesting though is the people building the MCP server on top of the rest APIs aren't. That's not like their goal with it. Their goal is to do some task. And so they're not spending time maintaining it and continuously improving it and putting in all the security features that they're going to want to be in it later. There's a lot of value in it to me, if you want to build on, we've got the REST APIs and Buyer Cloud or Beeswax was architected from
- [22:01]
David Dwaran
the beginning to allow you to do that. That's what I think is so great about it as a platform, is it allows you to build your own ui, it allows you to customize every aspect of it, whether it's the bidding agents
- [22:13]
Mike Trion
or your bid modifiers or the UI or using the data in real time.
- [22:17]
David Dwaran
And because of that, it's sort of like built for AI, right? You can do all of these integrations and let the agent integrate with every part of the system without having to force someone into a UI with those high switching costs, if they want to do that. To me, that MCP server gives you like a head start in those integrations so that you could just take off the shelf whatever agentic tool you want to use and plug it in and get started really quickly, assuming you're giant corporate IT organization allows you to do that.
- [22:47]
Mike Trion
But if you want to go build something more sophisticated and have a coding agent, hit the APIs directly, that works too. And we've seen people play there. So to me, it's more about creating a lot of paths to entrance where we focus on that underlying enabling technology that makes things work really well and let everybody's creativity bloom on top of it. Because Mike's got great ideas and other people have other great ideas. We want to create the infrastructure for that to happen and for that abundance to go forward. But also the MCP server is a great example where we can scale it faster than everybody because we do it across all of them.