Insight ON How Stagwell Stopped Agent Sprawl — and Turned Agent Building Into a Revenue Stream

Audio transcript:

How Stagwell Stopped Agent Sprawl — and Turned Agent Building Into a Revenue Stream

Merrill Raman (00:02):

The agency that builds these agents should get some type of benefit from publishing these agents for other agencies to use. So there's almost a licensing component. If you really like my agent, hey, maybe you should kick a few dollars my way.

Jillian Viner (00:18):

A licensing marketplace for AI agents built inside a company. I know I had the same reaction. Welcome back to Insight on I'm Jillian Viner and we're continuing our series on the agent economy. Today's guest is Meryl Ramon, global CTO of Stagwell, a marketing and advertising holding company with more than 70 agencies under its roof. Merrill has one of the most genuinely innovative agent stories I've come across because he didn't just green light agent development. He built the infrastructure that lets every agency inside Stagwell build, publish and share agents with each other, with real financial incentives for the agencies that create them. If you listened to our last episode and came away thinking, okay, I understand what agents are, this episode is going to show you what it looks like when someone actually built at that scale. I think you're going to find this one hard to stop listening to.

Let's get on with it. Well, Meryl, thanks for joining us. It's lovely to have you on the podcast.

Merrill (01:17):

Amazing. Thank you for having me.

Jillian (01:18):

I'm excited to talk to you because we are in the year of agents, right? Yes. That seems to be what everybody's talking about and everyone is at very different stages of their agent development story. For some organizations, it might be just individuals developing agents for unique tasks that doesn't really involve the larger org. In other situations, it does get a lot more complex. You have a great story because you have figured out something that I think a lot of companies are hungry to figure out, which is not only are you in a very complicated environment, but you have teams who are building agents and actually being rewarded for building agents. I want to know this story so badly. So take me from the very beginning. Where did you guys start building agents and what were the first agents that you built?

Merrill (02:09):

Right. So I think it's important to understand the environment that we're operating in. Stagworld's a holding company. We have 70 plus agencies. These agencies have some level of autonomy in the sense we did not want to stifle their creative processes. We wanted them to be unique in their own way. That's how they sort of service their clients. And so as it comes to adopting the agentic enterprise, we do not want to stymie their progress. And so our goal's been to provide them a foundation to build agents on. And there's a ton of experimentation going. We do not try to block or stop these experiments. What we do say is as these agents are alive and being used, let's find a way to publish these agents in a central location so everyone can learn about what these agents can do and reuse these agents. With 70 different agencies, one of the challenges we're trying to overcome is not to have duplicative agents in the enterprising.

That represents wastage and we want to sort of minimize that as much as possible. And so when we look at agencies in the creative realms, if they are building agents to support content creation, we're telling them, "Hey, let's collaborate. Let's make sure everyone's aware of the type of agents you all are building." And once the agents are ready, publish them in Stagwell's agent cloud so everyone sees what's out there and can use them.

Jillian (04:03):

The duplicative efforts and the agent sprawl is a very big concern. I don't know that a lot of organizations have really solved for that and it sounds like you've done that.

Merrill (04:13):

Explain- I wouldn't say we've fully done that. You're working for that. Yeah, we're working towards it. We believe we have a pretty good framework. It's all about making these agents discoverable. And once these agents are discovered, chances are people will build on them as opposed to rebuilding them.

Jillian (04:31):

Great approach. Give me a better sense of the world that you're operating in, because again, complexity is a really big piece of it. So what does Stagwell as an organization look like?

Merrill (04:44):

Yeah. We operate in every realm of advertising and marketing. So if you think about advertising and marketing, you have market research, you have creative services, you have media operations, you have digital transformation. All these realms are areas where Stagwell has agencies that are very prominent in-

Jillian (05:14):

And you're the CTO-

Merrill (05:15):

All

Jillian (05:15):

Of these organizations.

Merrill (05:17):

That's right.

Jillian (05:18):

What does that look like from a technology stack? Are they all using Google Cloud, Gemini? Are they on different platforms? What does that

Merrill (05:25):

Look like? There is a little bit of fragmentation. We follow the 80 / 20 rule. We want 80% of the estate to be largely standardized. We do have Google Stack can use quite extensively across our organization. We have the Microsoft Stack and use at our organization as well. The last 20% is where we let our agencies really develop their special capabilities. And we do not want to limit them in any way. They should be free to select the technology that best suits them, that best suits their clients or the need at hand. Yeah.

Jillian (06:02):

Best needs fit. So tell me again then, now that we understand the round that you're working in, the different organizations, different business purposes, in some cases, different platforms, what does that agent cloud look like? If I'm someone in marketing and I'm building an agent, what do I have to do?

Merrill (06:22):

Right. So first and foremost, you want to make the agent cloud available to all. And so the agent cloud is prominently featured on our intranet. Everyone at all Stagwell's agencies have access to our intranet and that's called the Hive. And so folks log in to the Hive. They have the ability to go into the agent cloud. Within the Agent Cloud, they're going to see a variety of agents that are categorized based on the type of services that they perform. And we have an entire team dedicated to curating agents and ensuring that the agents that are most prominent and successful are front and center on the agent cloud.

Jillian (07:11):

And the agent cloud is something that you guys built.

Merrill (07:13):

It's not something that you

Jillian (07:13):

Bought.

Merrill (07:14):

No, no. Something that we built internally.

Jillian (07:16):

Tell me a little bit more about the functionality and how are you assessing what's a good successful agent that should give more prominence?

Merrill (07:23):

Yeah. So it's very easy to understand success of an agent, right? That you have a ton of information from the agent that tells you how many times it was fired, number of users that have accessed it. And then the agent cloud does a really good job of providing us that type of information.

Jillian (07:43):

So you have this agent cloud, which is like a library of agents.

Merrill (07:46):

Yes. But

Jillian (07:46):

I also understand that you're using it almost like a marketplace.

Merrill (07:51):

We're using it as a marketplace because we want to incentivize our people that build these agents. And one of the ways to incentivize them is the agency that builds these agents should get some type of benefit from publishing these agents for other agencies to use. So there's almost a licensing component. If you really like my agent, hey, maybe you should kick a few dollars my way to help me further subsidize development or turn it into a revenue stream of sorts.

Jillian (08:23):

It's really brilliant. I haven't heard of any other organization doing something like that. Again, you've got what, 60 plus agents?

Merrill (08:32):

70 plus agencies. 70 plus agencies.

Jillian (08:34):

All of them are developing agents.

Merrill (08:36):

All of them develop- So

Jillian (08:37):

Agency A, creates something that agency be

Merrill (08:40):

Really less.

Jillian (08:40):

They're basically-

Merrill (08:41):

Let's work on a licensing relationship. Let's work on ... Yeah. And in some cases, I mean, some of these agents get baked into client solutions, in which case there's revenue share opportunities. And so it's almost a new world of opportunity. Yeah.

Jillian (08:59):

How has that changed your business?

Merrill (09:01):

Look, it's still early days. And so you have agents that in some cases are fairly simple. They perform simple tasks, but the goal is to set a foundation for agents to be daisy chained to do real meaningful and transformative work.

Merrill (09:25):

It. And we have a business unit called the Stagwell Arketing Cloud and the Stagwell Marketing Cloud's whole goal is to develop SaaS revenue streams. And these agents, they correspond to SaaS revenue streams in many ways here.

Jillian (09:47):

Now I'm a very competitive person.

Merrill (09:49):

Yes.

Jillian (09:50):

So if I'm in one of your agencies, I of course want to be looking for the most effective agent to go build. What do I have to do to go do that? I'm assuming I first go to the Agent Cloud to make sure that my idea isn't already out there. And if it is, maybe it's something that I can riff off of, but do I have to go through some sort of, I don't know, InfoSec process to get approval or what does that look like? How enabled are people to develop agents?

Merrill (10:15):

Yeah. So when you have an idea for an agent, you have one of two options. One, you either have an internal dev team that can sort of help you bring that idea to life or you can say, "Hey, agent cloud team, we would like your help in building this and bringing this to life."

Jillian (10:37):

So I

Merrill (10:37):

Just

Jillian (10:37):

Bring them

Merrill (10:37):

Like a proof of concept. Irrespective of the approach you take, SDLC principles are going to be followed and part of SDLC software development lifecycle principles is to pay attention to security, pay attention to the type of information the agent has access to. So that doesn't necessarily change.

Jillian (10:58):

I see. So the ideas can originate from anywhere, but you still have a dedicated team that's doing the build and doing the CRE.

Merrill (11:05):

Yes, yes. Look, are there people experimenting with building their own agents, low-code, no-code agents? Yes. Anytime they do that before they publish the agent, there is a security review. There is a-

Jillian (11:23):

I see. Tell me a couple of agent examples that you're aware of that you're particularly proud of.

Merrill (11:30):

Oh my gosh. Let's pick any domain. I can tell you numerous agents we've built. Market research, for example, we've built an agent that takes on the role of a moderator and so you can put a potential respondent in front of this agent and the agent will have a conversation with the respondent to understand how it can pull key information about the research that needs to be done.

Jillian (12:07):

So if I'm a customer being surveyed for something, I'm having a conversation with an agent.

Merrill (12:12):

Rather than doing a traditional survey, you're having a conversation with a bot and that bot is well versed in the domain that we're trying to do research in. And so we'll have a very educated conversation, we'll probe on certain areas. If you try to game the system, it'll tell you, "Hey, no, let's stay on topic. This is what we really need to talk about. "

Jillian (12:38):

Interesting. Yeah. Have you found that people are more honest when they talk to an agent than if they were talking to a person in those situations?

Merrill (12:47):

So look, I think we've done experiments where we've divulged that it's an agent, it's a bot and where they're unaware and we honestly can't tell the difference. Yeah.

Jillian (13:03):

Okay. All right. That's a great example. Do you have other workflows where agents are in there and how do you decide where to keep the human in the loop?

Merrill (13:13):

So look, human in the loop is a critical part of the process because agents can hallucinate and so it's important to have that human verification layer. And remember, we are human led machine augmented. And so irrespective of the process, are humans going to be involved. What we see is now humans are playing different roles and the need for more humans is also very clear just because the scope and scale of the work that can be done now is so much improved.

Jillian (13:55):

Yeah. Let me probe you on a content creation workflow because I feel like content creation is one of those easy go- tos that people think of with AI generated content. How has AI impacted your content generation workflows or how-

Merrill (14:09):

Yeah. So personalization is the name of the game today. Yeah. We are now able to take content and personalize it to a very, very high degree and the cost associated with building thousands of variations of content has gone way down. So in a typical campaign, you may create maybe 50, 60 pieces of content and you would make some decisions. So I need this content activated in a certain region, so let's make some modifications to the content based on that. Now that is increased by a pretty huge factor. You can personalize the content to the nth degree and it's all generated by AI.

Jillian (15:09):

What's been your biggest lesson or takeaway from the journey that Stagwell has been on with agent development? What's something that leaders often ask you about?

Merrill (15:20):

Don't be afraid of failure and you hear all these efficiency numbers being spouted and you will get there, but it's going to take a little longer than you think. These solutions aren't turnkey solutions, right? The adage, people, process, technology really applies in simply providing and technology alone is not enough. And another key insight I'd love to provide is do not lift a bad process into this environment. Review the process, re-engineer the process with the tools and technologies you have available. All too often, I see teams very excited about the technology, want to simply apply the technology to a part of the process. It's better to take a step back and reimagine the entire workflow, start to finish, and then bring the technology in to support it.

Jillian (16:37):

I bet that reveals some surprising insights about the way that people work.

Merrill (16:42):

Oh, very much so. Yeah.

Jillian (16:44):

Now you're kind of touching on a little bit of a human element. So before I let you go, what would be your top three pieces of advice for leaders who are trying to get their orgs to adopt AI, trying to do agents, but they're feeling that friction of that cultural resistance?

Merrill (17:02):

Yeah. This is going to be an emotional journey. I think folks are going to go through a variety of phases on this journey. The first phase is going to be this sense of amazement when you discover what the technology can do. The second phase is anxiety. How do I use the technology to do what I want it to do? Third phase is going to be despair because you're going to have some failures and so you will second guess yourself. You will question whether the technology is really ready. You have to sort of persevere through this. You'll have the breakthroughs and then you will go through elation and momentum and you'll be able to achieve the transformation you're looking for. And so being with your people when they go through these emotions is of critical importance. You almost want to let them know, "Guys, this is normal.

We are going to go through these challenges as we try to transform."

Jillian (18:21):

Just live it. Just live it with them.

Merrill (18:22):

Live with them, exactly.

Jillian (18:23):

Experiences. Yeah. Awesome. Meryl, thank you so much. It was so lovely to have you. I've learned so much. I'm fascinated by your internal competition for building agents and being able to share them across teams with the benefit. So I hope our listeners gained something from you and thank you again for your time.

Merrill (18:41):

Thank you very much. Thank you for having me.

Speaker 3 (18:44):

Thanks for listening to this episode of Insight On. If today's conversation sparked an idea or raised a challenge you're facing, head to insight.com. You'll find the resources, case studies, and real world solutions to help you lead with clarity. If you found this episode to be helpful, be sure to follow InsightOn, leave a review and share it with a colleague. It's how we grow the conversation and help more leaders make better tech decisions. Discover more at insight.com. The views and opinions expressed in this podcast are of those of the host and the guests and do not necessarily reflect on the official policy or position of Insight or its affiliates. This content is for informational purposes only, should not be considered as professional or legal advice.