Agent Swarms for Demand Orchestration

Julia Nimchinski:

Thanks, and we are transitioning to metadata, Lind. Welcome to the show, James Silvestri, Head of Demand Generation, and Lisa Sharapada, VP of AI and GTM Strategy. Long time no see, how are you doing?

Lisa Sharapata:

Good! It’s great to be here, Julia! Thanks for having us!

Julia Nimchinski:

Always excited to have you here! What’s the latest and greatest?

How’s your conference season?

Lisa Sharapata:

Yeah, I mean, it’s kind of starting to wind down a little bit, but good! Yeah, it’s been a whirlwind.

Julia Nimchinski:

How about yourself, James?

James Silvestri:

It’s been a busy year. Exciting year, everything is changing as fast as we know it, and so it’s, you know, keeping that momentum up is really exciting to see.

Julia Nimchinski:

Definitely, and before we get started, you know what I’m gonna ask. It’s been a year when we were experimenting, deploying all sorts of, you know, agents, AI integration, and just curious, you know, metadata is so ahead of the curve. I think, I do believe so, Lisa, correct me if I’m wrong, that you were the first to actually natively integrate within the LLMs, like ChatGPT, and curious, what are some early innings, ROIs, customer success stories, anything tangible that you can share?

Lisa Sharapata:

Yeah, so… We… you’re right, we are, and so we’ve got this integration now with Metadata and the MCP server. We actually just got ChatGPT working, so I’m excited for that, but we’ve been on Claude for over a month. over… well over 2 months now.

And we’ve got Zoom as our biggest use case, so they’ve been up on it, and… you know, I’ve… actually, I’ve got some slides here.

I’m gonna share my screen, so I’m just gonna pull that up, and show that, and… You can see some of the results, but yeah, they have… gone from, like, 3 weeks to 3… 30 minutes to set up campaigns, which is something we’re going to show today. And you can also see they’re dropping their cost per click, and then we have something called BidAgent, which is actually going in and optimizing LinkedIn bids.

They’ve seen a 24% savings there, but it’s also actually increased their influenced revenue by 177%. So, like, this autonomous bidding and going in and making adjustments and things like that.

So, those are just some of the results that they’re seeing, but, there are some that we’re going to show today.

So, excited to kind of show you how it’s done.

Julia Nimchinski:

Good start! Let’s do it!

Lisa Sharapata:

Alright, alright. So, first thing, I just want to kind of set it up a little bit, and kind of this whole idea of swarms, right?

And Julia knows, I was, like, reaching out to her, I’m like, I don’t like the word, because I think it comes off really, you know, not something you want to, you know, you don’t want to be swarmed.

But when I’m looking at it.

From the standpoint of what it actually can do in the sense of metadata, we’ve got… you know, the question is, like, what if agents could actually create a new paid ad campaign for you, and negotiate the best outcomes? So, when I think about it in that context, I’m like, oh yeah, okay, bring on some swarms, whatever that looks like, right?

And so, I just want to kind of tee up kind of how we got to this point, and also, like, why it’s so important, and I think it… why it’s valuable, and you would want this, so… I won’t go through the whole history of marketing, but you kind of get the gist here that, you know, 1999, there was no, digital channels that we were, like, worried about trying to market in, but as the evolution of digital channels came also this data, and with data became data-driven marketing.

And so then marketers started to be held accountable for results, and so we created things like MQLs so that we could show what we were doing, but we also then started to realize that we needed to tie our work to outcomes, such as pipeline and revenue, that were really meaningful to the business.

And… So, with that, we started connecting ourselves more to the sales motion, and… I mean, it became very complex very quickly, and you’re trying to orchestrate multiple campaigns across multiple channels, but also surrounding accounts, and not just driving leads.

And so… you know, to try to figure out how to address that, again, marketers have added all of this complexity over time, and so that’s why I’m so excited about this idea of the agentic Workflow, because really what’s happening now is agents are talking to each other to really help automate and get the outcomes you’re looking for, and I call it kind of doing the ing of marketing.

It’s all the stuff that marketers don’t really enjoy doing, and it allows the marketing team to be more strategic and really… be thinking about how, and what they should change, and the brand, and the market, and what’s happening in the market, and how do you make sure you’ve got your company positioned, so… With all that, I’m gonna dive into how is metadata helping to solve this problem. So, one of the things that happened when all of these digital channels came alive is that you’d start to put campaigns directly into those channels.

But when you’re doing that.

You have a few options. You can manually go in to each one of those, and you can… cherry-pick from the drop-down menus the audience that you want to build, and this takes forever, and you can do it only in a one-to-one to one scenario. So, one audience at a time, with one offer in one channel, and you’d have to even A-B test that, you’d have to build another one.

And do it again, and then test it and see which came up better, but you’re manually setting each one up and looking at the results.

And you’d have to do that across each channel, and, like, another problem with that is, like, my Facebook profile’s very different than my Instagram profile and my LinkedIn profile, so even trying to figure out who is Lisa on Instagram versus, LinkedIn is a challenge, right, in this scenario.

So.

There’s all these complexities and problems with this direct-to-channel approach. So where metadata’s come in is this multivariant testing and optimization, where you’re able to Create an audience Match it in these different channels.

Create multiple offers, and run them simultaneously, and the machine is actually using AI and agents, talking to each other, learning, and figuring out how to optimize across these different systems, and you set the outcome. So you can say, I care more about pipeline and influenced revenue than cost per click.

Or maybe you care more about cost per click So you… you set the outcome, and the machines are, again, continuously talking to each other, they’re goal-oriented, they’re learning. Maybe a competitor comes in, starts outbidding you, they’re changing. They’re adjusting in real time and adapting and negotiating on your behalf, and you don’t have to lift a finger.

So… that’s the first thing, and I will show… we will show this in our platform, but before we do, I’m going to pass it to James to even go a level deeper, to really look at this orchestration and how it’s set up.

So, James, I’m gonna pass it to you, and talk about this.

James Silvestri:

Perfect. So, if we’re looking about industry-wide go-to-market motions, it’s very… siloed.

The paid digital has their own nurture, their own follow-up, their own sequences, email the same, retargeting the same, and hopefully they’re all being optimized for pipeline. And at first, it was manual, and then it was automated, and then we developed workflows, and now we can have single agents that do it, right?

You can have one agent that does one component of that. But when we’re looking at the future of where go-to-market motions are really headed, it’s more about coordination, these coordinated systems, where you have all these intertwined components behind the scenes, because you’re the buyer’s journey, and the buyer’s not just going through one channel.

Or one mechanism. And with those silos comes lack of performance in terms of the opportunity to perform better.

There’s solo feedback loops, and it’s not really taking the entire situation into mind, like Lisa was showing on the last slide, across channels, across profiles, and those different aspects of it. So, when we’re looking at this, it’s all optimized for the same goal, but they’re intertwined, and that is more of a realistic scenario of what we’re looking at in the market, especially today.

On the next slide, we’re really looking at how this plays out in the funnel within metadata itself. So at the very key and heart of it is that multivariate testing.

And behind the scenes, in all those different agents, and all those different components, the bid agent, the budget agent, the different channels, the auto-pause rules, all these things are going on at the same time.

So that you don’t have to be siloed into one of these areas of just adjusting bids, or just focusing on budgets, and then going even minute on one channel, and then on another channel.

And so, when we look at this, it’s like optimizing for a single day versus optimizing for a quarter.

It brings everything up to a much higher level. And so your performance increases because all these things are steel sharpening steel, and improving with it.

And this works throughout and going down the funnel, where, for example, like a dynamic audiences component, we all know That compared to a top of funnel, you know, someone who’s never heard of your company versus someone who has been very adamant on your website constantly, those are different styles of engagements, and so it factors all these different components in.

Lisa Sharapata:

Yeah, and, I mean, you also still have full control to say, I mean, now if someone’s coming to your website, compared to a year ago. they’re a super hot lead. Right now, we’re thinking of them as being bottom of funnel, whereas a year ago, it would have been… they probably would have been more top of funnel, and so you can make those adjustments and change how you want the rules set up within your agentic system.

But… You know, you still have the control, but then you’re, like, letting the agents go in and figure it out.

James Silvestri:

Exactly, and speaking firsthand, I have to monitor and check it vastly less, but it allows me to focus on optimizing for the year versus the day.

Lisa Sharapata:

Yeah, or the quarter.

James Silvestri:

For the corner.

Lisa Sharapata:

Alright, so I’m gonna jump in, and right now, we’re in our Claude instance, and so just really quick context. James, I think this is what was going on before. I think we might have a problem with the MCP servers not connected.

Do you want to check yours quick?

James Silvestri:

Yes, we’re also traveling, so bear with us one moment as I pull up my…

Lisa Sharapata:

We may, if I have to, while you’re doing it, I think, normally, this would be on, and… what I would be able to do is go in and prompt what I want to have happen, because all of these agents are set up in the background. So let me just look really quick, because, I mean, honestly, I think this is, like, the real life… situation of… What is, happening today for most organizations, so… Yeah, I don’t know, are you seeing the same thing, James?

James Silvestri:

Yeah, seeing the same thing on my end.

Lisa Sharapata:

Yeah, no, mine’s on. Okay. Well, here’s what I’m gonna do.

I’m just gonna pop over, because I’ve already practiced this yesterday, just to make sure it was working. I’m just gonna show you. So, yesterday, we were connected to the MCP server, and… it’s showing me, here’s the multiple tools or agents, that I have access to that are built.

So, we’ve got a campaign performance analysis, the budget group management, the campaign creation.

Ad and offer creation, audiences and targeting, and then all of these different channels that we just talked about, so integration into all of those channels, and creative generation. And so what the prompt here is, what we’re asking in this case, to be done, is to actually build a new campaign.

And this is, you know, when we’re talking about Zoom, like, how do they save this time?

We’re building a new campaign, we’re saying, research. our company and the competitors and compile a list of job titles to target, and I want you to break it into these three tiers with the primary decision makers, influencers, and specialists. And then, I want you to create these separate tiers.

Into audiences, and make some recommended usage for the offers. So… Metadata, you know, we’re actually going in first, we have a competitive agent, it’s going in, it’s looking at the competitors.

Then it’s starting to look at the different buyer personas within, you know, our group. Then it’s connecting to our CRM and all our database, everything that we have, and it is saying, okay. Here’s your tiers.

Now, within these tiers, I’m gonna start creating real new audiences for you that meet the descriptions of what have been defined.

So… We’ve got our 3 new audiences.

And the recommendations of what we should offer them.

All getting set up here. And James will show how this was actually built into the metadata instance, and… Again, just for the sake of, like.

why this is such a big deal. James, can you just talk about, like, the process that it would take today to build an audience, even if you’re not going direct-to-channel, even if you were using a platform like Metadata, before this MCP server could do it for you?

James Silvestri:

I’m not gonna lie, I was about to jump in. This is, like… sometimes a 2-3 day exercise to, you know, it’s first finding your ICP, doing a closed one analysis, a closed loss analysis, to find out who you should be targeting, the companies, the job titles. And then cross-referencing what job titles are available, what filters are available on all of these channels.

And by the way, none of them are the same.

Which makes it even more fun. And then documenting it, building out the process, and actually building out these audiences in each one takes a long, long time.

And I’m not exaggerating when I say it can take multiple days initially, and then even reviewing them and coming back to them a week later can take another 4 hours.

So it is… Even with, like.

Lisa Sharapata:

integrated spreadsheets and stuff, right? I’m like, you’re still starting in a spreadsheet today, so this is like pulling everything from your CRM and your database and building that all for you. So then, like, the next step to create these multivariant, testing experiments and campaigns is you’ve got to have your creative pulled in, you’ve got to have a landing page to drive it to, you need to set up your budget group.

All of that, so… In this case, we… we asked, Claude to create the landing page with the value props. Use these creatives that we already had built. And use this budget that we wanted, to pull from, and then schedule it to start Go run for 30 days, but not start for… until the 30th.

So… Oh, again, Going in, it’s starting to… write all this stuff back into the metadata instance, creating the campaign.

It’s now going in and saying, okay, I’ve created these campaigns successfully, I’ve built these landing pages, here’s your links, this was the audiences that were built. You know, et cetera, et cetera, et cetera. You know, go check for it in your, platform, it’s there.

I could actually tell it to deploy, but especially since I built new audiences and all those things, like, I… would prefer to go back human in the loop and check it before we just hit deploy. So then, with that, I’m gonna bounce it over to James, and he’s gonna show you, like, what that looks like inside metadata and how it works.

James Silvestri:

Perfect. So this is that campaign drafted, ready for review.

And so we can quickly see here that it has it ready.

LinkedIn, Facebook, Instagram were the three that we wanted it to run on. And when I drop down into each one, we can see the three audiences here.

And what’s really nice is I can actually preview them, too, before I actually spend any money on it.

So it goes a step further and adds a level of reassurance for me.

I can click and view any of the ads and the offers. Now, what I love about this is with this campaign structure, it’s going to combine each one of these variables where we get the multivariate testing from.

If we recall back on that slide with the connecting circles.

this is one of those key elements, where it’s gonna find what is performing best, what ad is resonating most with each audience, and what offer across channels, too. So, for example.

I’m not a big Instagram scroller, so I’m not gonna be performing very well to any of these ads on Instagram, but I do scroll through LinkedIn very frequently. And so, it will show that engagement, and it’ll actually shift budget to the campaigns that are performing the best based on the metric I pick.

So, being pipeline and revenue-oriented, I have this budget currently optimizing for pipeline itself.

And now I’m gonna go over one tab and showcase what that actually looks like.

And so this is a behind-the-scenes view of what you’d see in the platform of that Agentix form.

And that goes back to that slide I was talking about earlier as well, where we can see it’s adjusting budget based on pacing.

How are we doing with our budget throughout the month? It’s changing bids, so we’re not overpaying, or we’re not underpaying, and not getting any placements whatsoever. So it’s shifting budgets, and side note.

I have never shifted bids. to this minute detail, I’ve never changed anything to.43. It is doing all these calculations to ensure it’s just the exact amount needed to get the optimal placements without overspending.

And as I scroll down here, we can see there’s more pacing ones, but also, it’s changing budgets based on pipeline, or what we call triggered opportunities.

And so, with this, we can see it’s pushing more budget to this one, and to several of these, and if I go through the other tabs here, you’ll see where it can also pull budget from, and the reasons behind.

So it’s show… showing you the reasoning, so you never have to… Feel like you don’t know why it’s making those decisions.

And, I actually just realized this.

I have it filtered down for the last 30 days. And there’s about 25 rows of data here, and there’s 40 pages, so that alone is looking at around 1,000 optimizations that were made that I didn’t have to lift a finger for.

Lisa Sharapata:

Yeah, so let me just chime in a couple things. Like, first of all, I love that it shows its work, right? Because it’s not just this agentic system that’s happening, and you have no idea what it’s doing or why.

You can see it, and you’re like, oh, why did it go up? Oh, well, here’s why, and here’s what it’s… the outcomes and the data that’s happening with it. So, and then… James, you touched on this, but… Do you want to show, too, how you… what you’re adjusting to?

So you said, like, you get to select the outcomes, right?

And…

James Silvestri:

You could be…

Lisa Sharapata:

Running a whole, like, a hundred different campaigns if you wanted, each with different outcomes, but each time it’s doing its… negotiating, and it’s learning within that system. So, it’s not… you know, going to impact your other campaign.

You’ve got this… this system that’s built, and it’s doing the, you know, multivariant testing with… inside of it, and it’s learning as it goes, and it’s adapting and negotiating on your behalf for that goal, but you could have a whole other one set up for a different goal.

James Silvestri:

That’s exactly it. So, if I’m looking at these different structures, I have this brand awareness one dedicated to optimize for cost per click.

I want to generate as many qualified website visitors as possible. And that engagement.

And because it’s ungated, cost per click is a great metric for this.

And so, it’s… every campaign I choose to be within this group is going to be optimizing across all of those variables for cost per click.

And then, I call it the guardrails on the highway of these auto-pause rules.

I get to determine the criteria of when the experiment will just pause itself. So I don’t have to constantly babysit it, I can focus elsewhere in what future campaigns I want to launch.

And with this, it’ll automatically turn off if I decide to turn any, By channel, essentially, what that cost per click would look like.

Compared that to one that we’re driving, you know, actual meeting requests with. I want to set this up for triggered opportunities, which is pipeline in our system. And the nice thing is, regardless of the custom metric, I can choose it, and there’s enough logic built into it where if there’s not, you know, pipeline’s a lower funnel metric.

it’ll recognize that there’s not enough data for that, and it’ll move one step back. So it’ll say, there’s not enough data to make an accurate decision for pipeline.

The agent then is going to say, what is the next best criteria, which in our case would be influenced pipeline, and then one step back, SQLs, or MQLs, or your three-letter acronym, because everyone has different three-letter acronyms. And so, it will automatically adjust for that, too, based on the settings.

Lisa Sharapata:

Alright, so Julia, we could go on forever, but I think we’re close to time. Do you have any questions or thoughts you want to share?

Julia Nimchinski:

This is so cool. Thank you so much, Lisa and James, for the amazing presentation, and Yeah, it’s… I think it’s the, you know, ultimate dream of every demand gen leader, marketer, But… I would imagine that there is a lot of friction in terms of transitioning from, you know, the old dashboard type of experience into something like this self-driving car, oh my god. So, how are you handling that?

All of the objections, rejections, and I think, yeah.

Lisa Sharapata:

It’s a great question. What I’m realizing, right, I’ve been here about 6 months, but I’ve always been in kind of a B2B tech world, and the art of the possible is always, like, really intrigued to me… intriguing to me, but when I talk to people, they’re like.

Still stuck in this mode of what they’re used to having controls over, and how they’ve run things in the past. Which is even why some of these diagrams, right? Like, we’re trying to explain, like, this is… The possibilities, and… You know, just really trying to open people’s mind up to see, like, you don’t want to have to sit in here and bid on these campaigns, and neither could you ever optimize to this level of granularity.

This allows you to be strategic again. This allows you to come up with your next cool, creative idea for your next campaign, or try two different variations of it, right? And… at scale and have fun, instead of being in there, like, manually adjusting things.

And so, I think when people kind of Have some of those aha moments and start to see, like, how this opens them up to be able to be more creative again, and kind of have more fun, and be more strategic.

they’ll, you know, let go a little bit, but yeah, it is a… it’s like… I just wrote a post on this, it’s like. asking what kind of an engine is in an electric car? Like, you’re just… you’re so programmed to think that a car has an engine that you’re not really realizing that it completely works differently.

It’s like the wrong question to ask, so… That’s been a lot of education on just this whole change of mentality. James, I don’t know, you’re, you’re even more…

James Silvestri:

I mean…

Lisa Sharapata:

do it.

James Silvestri:

On my end, it’s a change of philosophy and getting used to the new norm. You know, I still… I’ll be having dinner, and I’ll be sitting down, and it’ll hit me. I think I actually called Lisa when this happened a couple weeks ago, too.

I called Lisa, and I was like, do you just realize we’re one of the first marketers… if… the first marketer to launch a campaign from Claude, from ChatGPT?

Lisa Sharapata:

Like, end-to-end, and I was like, oh yeah. We are!

James Silvestri:

And that was…

Lisa Sharapata:

I knew exactly what you’re.

James Silvestri:

Yeah, and it was, like, two weeks after it. And it settles in, and it’s a change in philosophy, and we’re seeing a lot of, you know, early entrants and people that are really driving that that curve and driving the… what the new norm looks like are really picking it up as fast as they can.

So there… there is a lot of resonance with that.

Julia Nimchinski:

And building on this innovation, James and Lisa, what’s your prediction for 2026 and all things Agentic Marketing?

James Silvestri:

I do think there is going to be more acceptance And space created for it? Where it is, I mentioned a little… a second ago, the new norm. That it’s not going to be this scary new thing, it’s going to be a fast adoption and a race to who can master it the fastest.

Lisa Sharapata:

I’m starting to see… The aha moments, and also the advancements in… The technology, for example, like. skills with Claude, and the ability to build agents so much easier now, and workflows easier, and I think that Once people start to see the power of it. and how it will actually solve for problems and make things better or easier, it’s gonna start to be dialed in on real use cases that move the needle, and right now, I feel like that’s very few and far between, but I think that the light bulbs are starting to go off.

So I think next year’s gonna be really pivotal in The art of the possible, again, and people actually seeing results.

Julia Nimchinski:

Super excited for this, and yeah, Lisa and James, thank you so much again. What’s the best way to just, I don’t know, start experimenting with metadata?

Just go to the site and hit demo, or how does.

Lisa Sharapata:

We actually have, we have a free trial for BidAgent, which is… Really a 5-minute setup, and you… toggle it on. But, you know, it’s also, I think.

come to LinkedIn, or come to our website, get a demo, set up time, talk to us, because back to that art of the possible, and, like, really understanding, you know, even the kinds of shifts that you might want to make in your current campaign strategy. We’ve got people like James who can help with that, and services to help you get it set up, and I think that’s really the best way to do it.

Julia Nimchinski:

Things again.