← Marketecture: Get Smart. Fast.
Loading summary
Transcript21 lines
- [00:00]
A
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 is Ari. We have a recording from a Market Live event back in March. This one was a crowd favorite, actually got applause from the crowd. The title is ad CP in action. Let the AIs rip. And this was a great conversation between Harry Tong, who's the Director of Solutions of Engineering Agencies and advertisers for PubMatic, and Georgiana Hague, who's MIQ Global Strategy and Partnership Director. And they have an actual demo of working with ADCP Live, so I hope you enjoy this.
- [01:07]
B
So as we were, as we were walking on stage there, Georgie, to that pretty excellent tune, which I think is apt given the title of the session, I was just thinking how nice it is to have an audience who's here at their own free will to hear about Ad CP and agentic advertising. The reason I say that is because my wife and certainly my one year old have heard A more than they like to and B more than they need to about this subject. So it's nice to have a receptive audience here today. As the title suggests, that's exactly what we're going to be talking about. But not just talking, we're going to show you this thing in action and what it can facilitate in the real world when plugged into the right foundational capabilities. Before we do that, a couple of quick intros. My name is Harry Tong. I work on the solutions engineering team at PubMatic, really facilitating agentic collaboration with our buy side partners and ensuring they unlock maximum value through our Agenticos. I'm very happy to be joined up on stage by one of our favorite partners, miq, and also fellow Brit Georgie Hague.
- [02:09]
C
Hello everyone, I'm Georgie Hague, Strategy and Partnerships Director at miq, Programmatic Media Partner. I'm running all of our global demand and supply partnerships as part of that role, including thinking about new buying models. So excited to get into AD CP today, but one other reason I just wanted to flag that we're happy that you're with us here today is to give confidence. So MIQ recently published an AI confidence report and we found that while 75% of marketers are leaning in and planning to ramp up AI usage, only 44% of them feel confident about it. So we'd like to give this room a little bit more confidence in how to get started with all of this.
- [02:49]
B
And the way we're going to do that over the next 18ish minutes is we're going to define ADP, define the vision for it. I'm going to attempt an acronym analogy, which sounds fun. We're going to show you the thing in action with a live demo and get both the buy and sell side perspective. So a lot to cover, starting with the definition. Georgie, you're the smartest one out of the two of us, so I think I'll pass over to you for this.
- [03:12]
C
We'll let the audience decide who's the real brains behind this. Okay, I'll attempt the definition. AD CP is an agent native open standard. So it provides a framework for communication and collaboration across the AD ecosystem. It's standardizing the workflows that we're all familiar with, so planning, discovery, optimization, reporting, and it builds on top of existing workflows. So it's augmenting, not replacing. And I want to just dwell on that for a second because I think both of us think it's really important. The RTB ecosystem is pretty incredible and I'm sure a lot of people in this room today would agree. Decisioning at impression level a trillion or so times a day, that's both highly performant but also very impressive. So we don't want to rip that up. However, we also recognize, and I think the industry is recognizing, that there are issues with it. There's fragmentation, there's a ton of human LED orchestration, a lot of time spent, and all of that comes at an opportunity cost. Not to mention that a lot of AD transactions actually take place outside of the RTP ecosystem entirely. So ADTP is really aiming to bring all of that together, to unify all of that, to drive operational efficiency, to reduce friction, to objectivity, and to ultimately enable outcomes, which is what we're all here for.
- [04:32]
B
I think that last point is really important as well, as we look to compete with walled gardens. Effectively bringing those three together is how the open Internet does that. Now, quickly moving on to an acronym analogy, which might be the highlight of the session, or it might be confusing. Either way, add CP, MCP A2A. We've spoken a lot and heard a lot about these different acronyms, and there's a key distinction between them that I just want to touch on very quickly. And I think of this as two cities being connected by a highway. MCP and A2A. They are the tarmac. They are actually the Connection protocol that joins point A to point B. ADCP is the abstraction on top of that. Right. It provides the rules of the road. It tells us which side of the road we drive on. Our supposed to indicate when we change lanes, the speed limit, how we get on and off that road. It's what enables bikes, cars, trucks, buses, everyone to go harmoniously from point A to point B. If you think what it would be like if we didn't have those rules of the road, you might just about get to where you're going. It's going to be pretty stressful. You're not quite sure on the outcome. Pretty much like driving in New Jersey. Everyone's pretty familiar with that. I live there. I know that that is not the experience we want in the future of agency advertising. That's why ADCP and other protocols are really important in scaling this ecosystem and doing it effectively. And with that analogy fresh in our minds. Georgie, I'm going to come back to you here. MIQ have obviously been very early in leaning in here, both with ADCP as a standard and agentic more broadly. How is it changing the way you're thinking about things?
- [06:08]
C
So ADCP gets real for us as the rules of the road in a couple of places. One of those is standardizing how MiQ. Sigma, which is our AI powered technology, talks to other platforms across the industry. So Sigma acts as well. One of the things Sigma does as an orchestration layer across say 15 or so different buying platforms and traditionally we would need to integrate each of those directly. So all of the things that come with that. So scoping engineering, the time taken, not to mention building to fairly rigid APIs and the limitations around that. But with ADCP as the rules of the road, to take your analogy, we can quickly evolve our existing workflows really, really quickly. We're finding. So with Primatic, we're building MCP server support and we're also testing in parallel with Claude to move fast and to quickly learn things. And then the second area that we're leaning in is around automating PG style, direct IO handshakes with publishers. So that's an area where traditionally we've been limited by our platform's support. For that it's had to be bought outside of the programmatic ecosystem. But again, with ADCP as the rules of the road, we can broaden our horiz. So with some of the wider testing we're doing with other partners, we're leaning into that and looking at things like cost improvements, efficiency improvements, scale. So some interesting results there.
- [07:31]
B
I Think super interesting and early on, but already clearly lots of learnings there. And I think to double down on that point quickly, are there any key learnings or metrics that you're happy to share with the group without revealing too much?
- [07:44]
C
Yeah. So we had one of our traders, Russell, he's a brilliant trader at MiQ. He tested out the setup workflow with PubMatic earlier this week. This is not a mathematical. We haven't tested this exactly, but he said it was a 98% time saving to set up a complex campaign versus doing it in the UI. So, I mean, take that as you will, but that is ultimately hours of trading time per week saved, which is pretty astonishing. The other thing I would really call out is just the transparency and approval at every stage. We want Russell in the loop, we want a human in the loop. So that was really interesting to see. And then in our wider testing we have seen CPMs reduce as well. So worth calling that out.
- [08:27]
B
Yeah, I think that that 98% number, I'm pretty bullish on this kind of thing. That is. That's a wild statement, I think, and incredibly impressive. And think if you can scale that out as we are doing, that becomes a huge operational saving across the board, leading to the other outcomes we've talked about.
- [08:44]
C
Quite rare to have two Brits be so optimistic.
- [08:49]
B
That's true, very true. Okay, so we said we're going to go into the demo, which we're about to do. I think it's worthwhile setting some context before we jump in as to what we're about to show. We're going to be using, as Georgie mentioned, Claude as our orchestrator or buyer agent here, really to demonstrate how simple this can be. That is plugged into pubmatics agentic OS underneath, which is really powering the workflow with the recommendations that are going to come up all through the ADCP framework and leveraging that schema. So specifically what's going to be happening is as we go through the workflow, we're going to be calling a couple of agents within Permatic, the Inventory Marketplace agent, the Audience Discovery agent, and the Media Activation agent as well. The way I would think of it is all of these agents, that application layer make the foundational capabilities of Permatic available natively through this agentic workflow. I think it's worth touching on the foundational capabilities aspect there. The. The capabilities of a partner really dictate the ceiling of value that you can unlock through that partner. That is true today. That will continue to be true in an agentic Future, I think for Pubmatic, what that means very quickly is ubiquitous direct access to the open Internet alongside an expansive data Marketplace with over 250 partners, a direct to supply activation platform giving buyers flexible solutions and 20 years of decisioning intelligence fueling performance algorithms. Last point here before we dive in. There's lots of buildup, I know, but the last point is that agentic workflows we believe and I think generally the consensus is very much predicated on meeting people where they work and not the other way around. So what that means to us is yes, we're connecting in through Claude, but via our MCP server, all these capabilities can be accessed through a partner's proprietary technology layer or a third party buyer agent or an enterprise LLM like ChatGPT or Gemini or Claude. So without further ado, because I've already built this up, now we're going to jump into the demo. So we are in Claude. You can see that we have previously connected to Permatic's Agenticos there that requires an account with Permatic and full API accreditation and authentication. I've shared some basic details that I have a new campaign for the fictional company Acme Corp. Looking to drive online sales of their new sports clothing line. And very simply how can I set this up? So keeping these prompts relatively basic, but you could paste in a whole campaign brief here. What happened there while I was talking is it sent that to Permatic. Permatic sent back the different tools and capabilities that we have that can facilitate this workflow. And the vernacular here is in line with the ADCP schema. So you'll see Get Products which is supply and product discovery. Talking to our inventory marketplace agent. We have GET Signals which is signal discovery and audience discovery which talks to our audience discovery agent. You then have a couple of creative tools here that enable both trafficking and checking compatibility of creatives. Then you can obviously create the media buyer, you can get delivery, you can update the campaign as we go too. That's kind of as long as I'm going to dwell on for this side. But with ADP being an open standard, full documentation is available on the GitHub page page. So let's kick this off now. So I'm going to say great, I'm happy with this workflow. Can you please share some supply and product recommendations? Claude has understood the intent there what I'm looking for. It has enacted the get products call which again packages up that specific ask sends it to PubMatic, our inventory marketplace agent in real time. There is looking through existing RTB infrastructure So this isn't dependent on new infrastructure on the sales side, but leveraging what we already have and then sending back three recommendations there we can see this happened in under 10 seconds. That is a workflow that probably takes days. Currently multiple email chains very much compressed there. And these three different packages aesthetically look quite similar. That is by design, that is what ADCP does, it's what it standardizes. That's a key part. If we dig in just a little bit though, it's pretty clear they're very distinct in what they're offering. So the first one is an online video sports package targeting premium supply again across existing RTP infrastructure. This was curated in real time by our agent. The second one is a high propensity custom deal made in partnership with Chalice. This already existed within the buyer's account and it is using the IAB's ARTF framework on the backend to power it. So a great example of rather than these protocols competing, you can bring them together and achieve a much better result. And then lastly, we have a publisher specific performance package in this case with Raptive. Again this can either use existing RTB connections or connect directly into a publisher seller agent. So really flexible in how publishers can surface their supply. So I'm happy with those three recommendations. We know the last two there have decisioning built in already. But I think for that online sports package we can benefit from some audience recommendations. So I've asked for that again, Claude has understood the intent. It has triggered the get signals call within ADCP, sent that to PubMatic and PubMatic's audience discovery agent then looks through close to 100,000 segments that are readily available within this buyer's seat. Again, that took less than 10 seconds. It also incorporates predefined buyer preferences so you can choose which data providers you want surfaced first and then shares all of those segments back in a very standardized way, again thanks to adcp. So we have the name of the segment, the provider, the CPM and the estimated reach. So for the purpose of this demo, I'm happy to go with those recommendations. You can drill down a bit deeper if you'd like to though. So I'm going to say let's go with that recommendation and quickly before I hand over to Georgie to actually set up this buy, I'm going to share that I have a 15 second video asset. The reason I'm doing that now is because it ensures that I'm checking compatibility in my creator before the campaign goes live, rather than waiting for three days in and are still Q&A why we haven't started delivering. So now we've got to this stage, we just need to share a few more details and I'll pass it to Georgie to take us home here.
- [15:01]
C
All right, so I'm now going to be cosplaying as an M. IQ trader. So I've got a clear plan. I've got three packages to test, one of which has got an audience overlaid on top. So let's go ahead and we'll enter the details that it's asking for. So let's give it a name, start and end dates and set pacing, add a frequency cap, and Claude will go ahead and get this set up. As you can see here, it's calling out that I've actually forgotten to give it a budget, which is pretty important. So that's just one example of the guardrails that are built into the flow. So what we'll do now is add a budget, let's say 300k, and we'll ask Claude to split that evenly across each tactic and even just set a naming convention for the line items as well. So what I'll get now is just a final summary to review so I can say that everything looks good. Okay, let's get this started. And here's my summary. This is all looking exactly how I want it to. So let's say Create Media Buy. So this is now engaging the Create Media Buy function of adcp and the campaign is being created as we speak and pushed to PubMatic. So I should get confirmation any second that the campaign is ready to go. There we go, it's live. And here's a confirmation summary. So the final step here is really important and really valuable to us as traders because we can see that this has happened. We've had confirmation, but you can now switch into the PubMatic platform. And you can see we just refreshed and our campaign appeared. So here's the campaign. You can click in, see the line items, the budget and the naming conventions. Which brings us to the end of the demo.
- [16:36]
B
We're probably going to go over by 20 seconds for that round of applause now, so thanks. Thanks for that. So just to summarize quickly there, in about five minutes, maybe six minutes, we have shrunk workflows that used to take days into literally seconds. We've discovered and applied new tactics, things using new protocols like the IAB's RTF framework and referencing Chalice in this example, but also leveraging the existing infrastructure we have and making that available through agentic pathways. And we've gone on to set up the buy and actually get it live as well. So I think that's pretty cool. Sounds like everyone else in this room agrees with that. But as we know, the speed of innovation and new releases within Agentic only accelerates. So I wanted to look ahead to a couple of new features that will be available in the coming months to augment this workflow. And the first one is partners often come to us with a preformed media plan. So jumping into the pulmatic production environment here I've dropped in that media plan, our assistant is extracting things like budgets, line items, audiences and inventory allocation and in one step converts that into a campaign that's ready to go. But not just that. Based on the troves of sell side data and historical data that we have access to, we can then proactively surface recommendations on that media plan for the buyer to choose whether to accept or whether to stick with their original. I think a key point here is that rather than waiting for the campaign to finish to understand the fees associated with that buy, we will be servicing that front and center and forecasting that. I think the key point to land here is that we're not saying all fees are bad. Value exchange is very much a real thing. But what we are saying is buyers should have full visibility into those fee structures ahead of time and they should be able to make informed decisions before they spend their pdr. And then finally we know campaign setup is just one part of this process. So skipping ahead a few days here, once we have statsig data we will then continue to proactively surface optimizations and this is the real value of agentic. And then so a couple of examples here. Redistribution of budgets across line items, adjusting frequency caps and more. And the level of autonomy will be at the buyer's discretion here so they can choose whether this happens autonomously or whether there is a human in the loop approval process. And that does bring us to the end of the demos there. So hopefully a bit of a teaser there for stuff to come. Georgie, now that we've seen all of that, what was the thing that surprised you the most about building it and starting to work this way?
- [19:05]
C
So I think I'll go back to our testing earlier this week where Russell, who is a brilliant trader but by no means a developer, was able to get this set up without a single hour of engineering resource. That's pretty huge. So we can do that now via Claude, but the plan is then to build that into our platform, which should be very, very easy. And then the Second thing is that it's really additive. It's not a replacement for the current ecosystem. We're not ripping anything out and replacing it. It's more of just a gen AI layer that goes across everything.
- [19:39]
B
Yep, that makes sense to me. And I think we've seen a similar thing from the sell side. The level of engagement we've had from our partners and the ease to actually turn these ideas into action has been incredibly refreshing. And we're kind of coming. We've got about a minute left, so I want to leave. I want a couple of more thoughts from you in terms of a year from now. Do we think this is mainstream niche, the new default? And what is one piece of advice you would give everyone in the room to do this week based on these
- [20:06]
C
learnings a year from now? Mainstream for certain functions, so for integrations, yes. For certain campaign types, yes. But we firmly believe that Programmatic Rails are sticking around for the next three to five years and this will work in parallel with those. And then one thing to leave you with is the barrier to getting started is lower than you might think. You don't even need a Claude Pro account to get this set up and connect with, although we would recommend that.
- [20:33]
B
Definitely recommend it.
- [20:35]
C
But yeah, find a partner who has MCP support, get started with them, work with Pelmatic and get testing and learn things quickly.
- [20:41]
B
That sounds like good advice to me. But in closing, Georgie, thank you very much for joining us up on stage today. Your perspective is valuable as it always is and I think three maybe takeaway thoughts from me. We've talked about this a lot already, but ADCP and Agentic doesn't replace what we already have. It augments it. It allows us to get more out of the existing ecosystem. Second, as we've just seen, this is real and happening today. This isn't some future vision. Which kind of brings me to the closing thought here, which is if you do not currently have an agentic framework and testing strategy, I would recommend you rectify that quite soon. The reason being is we're not suggesting, I don't think either of us are suggesting agents are going to do everything autonomously in three months. I think that autonomy will happen quicker than maybe we think though over time. What we are saying and what we've talked about here is that the earlier you start testing with the right partners and the right framework, the earlier you start learning and you create a very clear advantage for yourself. So thank you everyone for joining us and I think we'll leave it there.
- [21:46]
C
Thanks you.