Francesco: [00:00:00] Good morning, good afternoon, good evening. Uh, actually I'm based in Singapore, so it's a different time zone from our two speakers today. Uh, hello Luca. Hello, Ben. 

Ben: Hey Francesco. 

Luca: Hello. 

Francesco: Um. We are, uh, talking on, uh, behalf of, uh, Tangoo and, uh, its listeners, uh, on a very exciting topic, uh, and saying it not just because I am facilitating this session, but because, uh, we wanted to give a different cut to a topic that has been on the lips of so many people nowadays, which is AI agents.

And we wanted to go through it with a very practical, uh, edge, uh, which is the user experience. Uh, so AI Agents is a new interface. [00:01:00] And that is because, uh, for 25 years, I feel like, uh, the web has been page based and keyword driven. Uh, to make a practical example, if you wanted to choose, you'd have the word and you clicked on a link.

And we feel like today that interface is dying, or at least it's changing. And we are moving towards an interaction based web where, uh, agents don't just find links, but they guide the discovery and they make the decision directly on, on the product page. And we have Ben Williams, founder of, Kiln, and Luca Mangiagalli, a tech lead in AI and data to help us, uh, unpack how we can turn AI agent in, uh, revenue, revenue opportunities for, uh, uh, publishers and, uh, and people around the web.

So, uh, welcome again, Ben and, and Luca and I would start right away, uh, with our panel. Uh, so Ben, uh. Many media and ad tech [00:02:00] business, uh, were built in a world without ai, you know? So from your experience, uh, what are the biggest challenges company face, uh, when transforming legacy tech and processes into AI led solutions?

Ben: Great question Francesco and thanks for having me. Uh, so I want to go back to something. Anne Coghlan, one of the co-founders of Scope three. She talked about when she launched on the day, uh, we launched a context protocol. She talks about AI isn't just about having sprinkles on a business, it's about doing the infrastructure work.

So in my perspective, any legacy ad tech business or media business, you know. 3, 5, 10, 15 plus years old that wants to transform to AI or to be AI native or to have an ai, uh, strategy really needs to start from the, the roots upwards. Um, one of the [00:03:00] challenges I think. A lot of businesses face, uh, is looking at the business processes that, um, underpin a business.

So if you are looking to automate parts of your business using ai, there's two big pieces of work that need to. Need to be focused on. One is not disrupting the business processes, which have got you to that point of success anyway. So introducing new processes will disrupt the sales cycles. It will disrupt account management, campaign management, so it has to be aligned to that.

And then the second piece is, again, at an infrastructure level, making sure that you have a, a strong data strategy so you can do. The automation, the machine learning, and everything on top of that. So I think the biggest challenge a lot of companies face is that they think it's just some sprinkles that can happen on top.

And in fact, it's infrastructure from, from roots up. 

Francesco: I, I love it because, uh, [00:04:00] once upon a time it was called digital transformation. Uh, and it always starts on understanding what the problem is. And nowadays, uh. You have users, you have interfaces, and you need to have data behind to really run, uh, the artwork of ml.

And yeah, in that case, I would, uh, turn to, to Luca because we, the background means data and I know you are an expert about it. So, uh, what kind of specific data, plumbing, uh, must the brand have in place, uh, before an AI agent can actually be useful for a customer? 

Luca: It is actually not much different from, in general, having a great data platform in place.

Uh, I mean whatever, uh, kind of, uh, integration you, you need to make with your products, uh, needs to have proper data platforms built at the bottom of it. And so whatever tool you [00:05:00] can now easily integrate with whatever, with a third party is always a good starting point, but. You, you don't really have to, to reinvent the, the wheel, uh, on the other side.

You cannot think that if you had a very difficult to use API not well documented without a good hair handling, then, then you magically have any agents that use it better than a person does. So, uh, it, it, I would use the metaphor of, of doing something that it's easy to use by people and can be used, uh, can be multiplied easily by a thousand agents.

So go, going, going, going that way? I think it, it's very easy to, to scale up if you're starting from, from a point where you have something usable. Uh, otherwise you have to restructure a bit, but it, it doesn't need the restructuring. Hmm. Forcefully, I dunno how to say 

Francesco: that is actually a relief because I been working [00:06:00] for nowadays, couple of years, uh, into, into programmatic and uh, I know that product seats at a mess.

I know that the way some e-commerce work behind the scene need a lot of manual work. So, um, most possibly, if this is. Or is something that you can consider designing it, uh, originally and, uh, can take care through that. It's already something useful. Uh, I don't know, Ben, if you want to add anything on, uh, on this from your own experience.

Ben: No, it's all very, uh, very wise words that Luca talks about here. And I think a lot what, what, what a lot of lay people don't understand that what underpins any AI strategy is a data strategy. And, um, you need to have the right people looking at that. And I think any big scaled ad tech proposition has a lot of data behind it.

So. That's a really core component of that. And Francesco, going back to your point as well about digital transformation, transformation, I think we're going [00:07:00] through one of the biggest platform shifts in technology right now. So anyone that's got experience of transformation is gonna be very valuable during this, during this transition.

Francesco: And it's also a matter of attitude. Uh, being comfortable with something that is not, uh, tested. It's basically launching nowadays, and that is a huge part of it from at least my, my point of view. And that goes through, uh, also experimentation with ai. Uh, there are a lot of technologies out there and not all of them turn into something useful.

So, uh, how are you shifting this focus from testing ideas to, to building real deployable solution for, uh, for media, or at least if we talk about ad tech? Uh, maybe Ben, if you want to go. 

Ben: Absolutely. And Francesco I think you, you hit the nail on the head when you, you look at problem solving, so. Uh, you know, you don't [00:08:00] want to have a solution looking for a problem.

You wanna have a solution to an active problem that you are looking at within your organization and containing your experimentation within that. And so experimentation is really important for any sort of, uh, uh, platform shift, uh, research and development, innovation, etc. Um, when we run. For my business and the projects we work on when we run experimentation, it's always very contained to the, uh, the business needs.

One of my mentors, a, a gentleman called Dion Price, always talks to me about not trying to boil the ocean. And I think you have to be very careful to just not experiment for the sake of it. So what we're seeing at the moment is we're seeing a lot of. Vibe Coding and Prototyping in Claude Code and Cursor and these platforms, which is fantastic, but again, it needs to be contained.

Otherwise, you are accumulating a lot of technical debt [00:09:00] if you spend all your time prototyping, so. Time boxing everything into specific use cases and objectives that are linked to, I guess, what the board of the company is trying to achieve. For example, we wanna unlock the small to medium business market across Italy, Germany, France.

That's a very simple business objective, and then containing your experimentation within that and then having, making sure that you're having a very good engineering plan when you're ready to switch from, from prototyping. And then taking that into what we call in product shipping. So actually building live products in a production environment.

So I think just be careful that just because you can prototype something doesn't mean to say we'll succeed in a live environment, uh, nevermind scale. So making sure that you're not prototyping for the sake of it, just because there are these cool tools and technologies available to you because you'll just accumulate [00:10:00] technical debt.

Really focus it on what those business outcomes you're looking to achieve. So then you can actually then switch into engineering and actually ship those products, uh, into your environment. 

Francesco: Luca, what do you think? Is there a secret source for, uh, prototyping and, uh, experimenting with ai? 

Luca: I, I don't think there is, uh, there are a lot of very good tools.

Uh, I've been experimenting with a few of them, and I think they're, they're very important to especially keep the pace and understanding what it's doable because of course when you take the decision of building something, you wait how hard it is to build it. So if this equation change or it's, it's important to be up to date on, on the other side.

I agree with what Ben was saying. Uh, bringing it to production is a completely different business. And in, in this case, I think it would be much more important to have, uh, um, a process and, and, and, and tools to bring things to [00:11:00] production fast. Then, you know, if they, if things are useful, if they work it, it doesn't make sense to have a hundred prototypes and do not deploy any of them, do not have feedback from, from the market and so on.

So in, in this sense, probably building this kind of, uh, process in inside the companies, uh, something that, that can accelerate everything and, and also allow you to, to, to bring. Further prototypes, production to production much, much faster. Especially if you're in the EU with older regulation and, and everything that else, that is much, much harder than in other countries.

Francesco: Or China, because I'm finding out that also data protection in China is more complicated than I was expecting at first. And uh, I don't want to go too much off topic, uh, regarding how AI is impacting on, uh, prototyping and experimentation. But if we follow something like design thinking, which [00:12:00] like, uh, diverging now with ai, you can diverge a lot.

And I was recently reading about how much vibe coding, actually you start building a feature and it brings you on 10 different features. So you end up building something that is totally different from what you expected at the beginning because you can, like, this is the potential that you have right now.

Uh, going back to, to, to the main topic of today is agents. Uh, so, uh, look at. Do brands need, uh, to rebuild their systems, to support AI agents? Or can they connect AI into their existing platforms and databases? Uh, what does it usually look like in practice? So that is the most important part. Like how do you do it?

Luca: I think it depends on the situation. There are a lot of tools that can be used to, to connect also to legacy systems. Uh, MCP servers had a lot of success on this, and they can be used for, for this type of use cases as well. Uh, so I don't think you have to re rebuild [00:13:00] everything. Uh, but of course, uh, if it's a multiplier, it can amplify if you already had some issues somewhere.

So, uh, in this case, maybe you have to polish something that wasn't so evident before. And of course, all the issues related to hallucinations and so on need to be managed at a process level. Because if, if we were in a physical shop with a person that is explaining products to a customer, of course if we, we want to organize that, that kind of selling, uh, we are taking in account that the person can, can make mistakes, can forget prices, can, uh, say something wrong about, about the product.

At the same time, uh. We need to build processes that take into account that sometimes things can be, uh, some small errors might happen, but if we have a product catalog that, uh, is fixed and we aren't, we're not giving full, full permission to, to everything that is running. Uh, I think it's quite [00:14:00] manageable.

Francesco: Just for the sake of a part of a million listeners to this conversation, uh, can you, uh, just explain to us what MCP stands for? 

Luca: It, it, it's, I, I'm sorry, I, I started it. The model context protocol is, uh, a protocol. I mean, it's quite technical, so I'll not go into detail, but in general, it's meant to be used to provide models, uh, the information in their context that is needed to perform specific actions.

So instead of explaining, uh, I want to, to query my database, I'm exposing from. Service all the information needed to make that query. And also some, some functionalities that do not need to be reinvented every time, but it, there are, there is a lot of documentation explained much better than what I'm doing.

Uh, I think it's a very interesting read for, for everyone. It was, uh, first proposed by Anthropic, but it's being widely adopted now, so I, I [00:15:00] think it's a, it's a very interesting, uh, deep dive.

Francesco: I think it's super important because it's one of those words and acronyms that you, uh, that you read around and you need to start understanding if you really want to, to implement agents in this case, or integration layers so that your own models can, can really apply and make a difference at no bank.

If you wanted to, uh, have a take on this. 

Ben: Yeah, model context protocol was a big step forward, Luca, wasn't it, for the industry. And that was the underlying protocol that add context protocol, which we'll probably touch on in this discussion was built on. So it was a huge step forward for innovation to allow for agent to agent communication.

And there's some really great experimentations across different industry verticals happening right now and in media. It's meaning a step forward in [00:16:00] how media can be traded, which is more efficient than how some of the real time bidding protocols in programmatic were, were set up over the last 10 to 15 years.

Francesco: This is a topic that has been ongoing for quite some time. You, you mentioned, uh, uh, the context protocol, uh, which is. Very, very recent, uh, at least for, uh, IB talks. Uh, IB Tech Club have been talking about in a recent webinar they had, which was very interested. Uh, so when we hear the term gentech, which is something that has been going on, uh, a lot from media buying and business perspective, uh, what does agen actually mean?

How should media leaders think about building roadmaps around it? Ben, if you were to take it 

Ben: Yeah, I'll jump in on that. Great question. I think, uh, definitions are really important when [00:17:00] things are, when things are shifting. So AgTech software that communicates with other agents to make decisions or take actions, the technology equivalent of a booking agent.

And I'll give you an example of where we're starting to see innovations like that happening. I, I'll give two examples actually. One, a consumer example. So if you go into chat GBT and you ask it to go into research mode, that is AgTech. It's looking at other information sources and bringing together these third party sources into like a cohesive research report.

Then I'll bring it closer to home. For, for media, add context protocol is a great example of, of what Agen is. So, uh, a brand agent on one side will say, Hey, I'm looking for. Football fans in Argentina as part of my World Cup 26 marketing strategy, that the buyer agent will communicate that it's looking for those [00:18:00] audiences.

And on the publisher's side, a sales agent can be set up across a publisher's ad tech stack, um, and can actually send signals out that it has the audiences and the campaign. Available to that buyer agent, and that's the fundamentals of ad context protocols. That's a brand buyer agent communicating with a publisher sales agent, and again, agen is the shorthand to explain what those two agents do to together.

Then there's a B2B internal application of it as well, which is if you're gonna be automating processes. You can create multiple agents within your technology stack that can all communicate with each other internally as well. So there's a, a consumer facing version, there's a business to business facing version, and there's internal versions of that as well.

Francesco: I recently, I, I love it because, uh, agen actually is one of those buzzwords, [00:19:00] so sometimes it gets, uh, misused. And recently, uh, I think it was on, uh, YouTube, uh, on a Jeff Sue channel, and he was explaining the difference between LLM AI workflow and AI agent. What, like how much the knowledge base and the ability to actually carry out actions make a difference.

As, as Ben was explaining right now from the buyer side and from the, from the seller side. And this is actually going to change a lot both on programmatic ward and uh, actually inside agencies. How, how they perform actions and. When you talk about ai, uh, you always have another buzzword, which is hallucination, hallucinations, and in a real retail or product environment, uh, look at how do you prevent an AI agent.

From going ballistic, like giving correct answers, suggesting something that doesn't actually exist. 

Luca: There is actually stronger search suggesting that, [00:20:00] uh, ha solution are much more likely when the models don't have enough information to answer. So they're trying to, to fill the gaps a bit. Uh, so of course having very well curated, uh, context of information and information available for our, for instance, products available.

Helps to re reduce the opportunity for, for this, uh, on the other side, as we were touching briefly before, uh, we need to, to keep track that there is a small risk here, but it's the same risk that, that we would have with with a person explaining capabilities of products. Uh, so, so we try to, to keep this into account.

We have some, some gate when we prepare an offer or a, a, a shopping chart for, for the, for the client. And at the end, we, we can de-risk this a bit, but the opportunity is much bigger to, than, than this small risk. Also, I, I think that most of [00:21:00] you're using, uh, uh, state of art tools like GPT, cloud, Gemini and so on, and, and this.

Type of area is strongly reducing over time. So it's not something very, very relevant. Uh, at this point. I, I don't think that this type of change can be really stopped just before just because of this risk.

Francesco: The risk that brand, uh, you want to get away from, which is the same reason why then that is so heavy. And, uh, brand safety is the reputation. So, uh, if you give me, if. To any of my, my clients give, uh, a wrong answer. So I have a feeling that the brand reputation is being, uh, torn to pieces, but I totally love your point is like, uh, how much are you, uh, aware of your own product?

How much can you sell? How much, uh, [00:22:00] can you describe. So the same risk that you are, uh, having when a sales person is going out there talking about your brand or somebody, an employee talking about, uh, your products is kind of close to what a model can, can do. So I love the, the association. It, it makes a lot of sense and, uh, I.

I dunno, 

Ben: Francesco I'll just come in there, uh, momentarily actually, we're talking about hallucinations and a lot of the smart people, they're making sure that there's a human in the loop, uh, for these automations. And I dunno if you used to watch the TV show Silicon Valley, but where one of the engineers built his AI system, son of Anton, and it would just make all these decisions and completely rewrite the backend code and.

Create these hilarious disaster moments. The smart people have humans in the loop, so especially when it comes around to reporting, [00:23:00] analytics, customer service. Having a, having a human to check before it goes out to the client or goes to the board is, is where the smart is, where the smart people are. And I think where autonomy works is when you are going out and looking for new markets.

So I, I, I wouldn't feel comfortable working on a project where AI is replacing people. I purposefully would not choose to work on a project like that. I'm interested to see where AI can multiply the existing talent that they have, but. Yeah, every project that we're on has a human in the loop somewhere.

So if there's a hallucination because there's an error somewhere, the human should be able to check that and, uh, and fix 

Francesco: love the point. Uh, actually it also, uh, touches a very important part, which is how you design an agent and what kind of actions you really want to refer to technology. What you want to, uh, [00:24:00] keep a human in the loop, uh, for, so that is super important also because sometimes those actions have repercussions.

So you need to be legally aware of what, uh, you're doing and what the, uh, responsibilities of, for what, for that action. Uh, go for, uh. I think we touched before on protocols, Ben. Uh, so a tech protocols, uh, that really help programmatic going on and that always shaped how media is traded. So as AI becomes more central, how are industry protocols evolving and what impact is that, uh, having on digital publishers specifically and how brands, uh, are trying to find new ways to trade?

You have an intake photo. 

Ben: Yeah, so past six months or past five months has been, has been lots of innovation happening there. So in October we had the launch of ad context protocol, which was the first AI media protocol to be [00:25:00] launched ever. And think of these protocols following on from things like open RTB pre-bid and just, you know, standardization.

So. The industry can, can grow. Um, and then the IAB launched their roadmap a couple of weeks ago, and again, they don't compete. They're very complimentary to each other. Aconex protocol is a really interesting one because the way I see it is programmatic was essentially, I'm, I'm showing my age a bit here, but it was initially set up really building on from ad networks and publishers would say, Hey.

I haven't sold all my inventory. You'd give it to an ad network, dead, sell it and give you a revenue share. Programmatic followed up from that and went into really technologically advanced states, you know, real time bidding, you know, very high queries per second, you know, very technologically advanced field Add [00:26:00] context protocol.

Whereas programmatic can sometimes be seen as a race to the bottom on pricing. Um, especially when you have bid shading, bid throttling happening in the ecosystem. Add context protocol is at the other end of the spectrum, which is a publisher saying. Hey, I've got this great audience. I've got this great first party data.

I've got this great content. I'm making those signals known to the ecosystem. So a brand agent, um, in addition to their, you know, usual brand, uh, publisher partners can look for additional in additional incremental reach. Um, and the nice thing there is that the media trading is done on a human to human interaction.

So in, in the request from the buyer agent will be a CPM. The publisher will negotiate and then accepts a final price, and this is where I really like it because it's humans. Then trafficking that on the existing ad tech stack, so it feels like an evolution building [00:27:00] on everything we've all been building for the past 10 plus years.

With Ad tech and programmatic, it's a premium way to, for, for brands and publishers to transact and then traffic all of that on their existing ad tech stack. So. I'll finish on this. Terry Kja in the us, uh, one of the big m and a, uh, commentators, he believes that the industry will grow from $1.2 trillion to over $2 trillion by the end of the year.

And AI protocols will be a, a big, a big component of that. 

Francesco: We are studying and following it closely, uh, with tech team, uh, in, in Tangoo because, uh, to be honest, there are several layers that you can apply. The inside of a, of a protocol, you, you mentioned there are two sides and one side focuses on, uh, simplifying and really automating the way that, um, brief, uh, can be turned into, into revenues, uh, on the other side for the [00:28:00] publisher.

Uh, that is also linked to the topic for today. Like the user experience and how agents are installed or, uh, can have an impact on, uh. The experience of users fund, uh, searching for or query for information. So I will just go to look for a second, uh, to understand what kind of, uh, different experience can ai, uh, have on, for example.

So how do you embed AI into the front end, into the user experience of a web or an app? Do you have any, any examples, uh, coming from your own experience? 

Luca: I, I think we are at the very beginning of this, so we've seen around chatbots and similar things, uh, with, uh, a bubble like charts or something that doesn't look like bubbles.

But, uh, I think there is a lot of room of improvement and creativity in this. [00:29:00] Of course, this type of interface helped a lot because it helps people, uh, provide all the, the information they have, even if they're, they don't spend the, the energy in formalizing a, a, a full request. So if you have client that is interacting with a shopping website, uh, is not preparing the document, explaining the requirements and everything that he, he or she would like to buy.

But it's saying, I would like, uh, a pair of shoes. Okay, which color? What is the purpose? You want to run? You want something elegant? And that type of interaction works very well on the other side, I, I think it's a bit, uh, reductive for a lot of, um, of use cases and also for the potential that models currently have.

In other fields, for instance, in coding where we have the, the biggest impact of AI agents so far, uh, we're starting to see a lot of success of solutions like, [00:30:00] um, software to create code that already integrates agent. So the agent is not, uh, relying on what. Text him, but, uh, he's watching what you're doing.

So you open this file, you copy something from here to there, you have this files in your working folder. And, and then what you're asking, it's being enhanced by, by the context the agent is working in. And this is just, in my opinion, the first place where we are seeing this and going on, uh, as long as people will.

Continue experimenting and have a spark of creativity, we'll find something interesting, I think.

Francesco: Thank you. Thank you so very much. I love the fact that we are, uh, uh, going on different fields and, uh, uh, getting from our own experiences and field of expertise. And, uh, if we want to go that way, look at, uh, many brands don't have expert. [00:31:00] Tech team? No. So they, they have somebody that takes care of the VIT side of things, but they're not necessarily AI experts.

So what is the single biggest pro and con on the other side? They should wait before committing to a custom AI agent in your opinion? 

Luca: I think it's fundamental to have a clear goal in mind because, uh, when you're getting an AI agent, it's like, uh, getting, uh, an employee. With quotes, let's say, with an employee that does exactly what you are telling him to do.

So if, if you don't know exactly what's needed to make the things work, uh, it does not work very well. And probably there are many contexts where people who manage the, the team think to have well-defined tasks. But actually things keep working because. People actually doing things are solving a lot of small, uh, small issues every day.

Uh, and then when you're missing one of this person, you realize that, that that person [00:32:00] was solving a lot of issues you, you never thought about. So being in that situation in an unstructured place, uh, is more complicated. On the other side, uh, I would focus more on, on the data platform side because, uh, the technical part of building agent is something.

The, that is been partially solved by off the shelf, uh, components or services that help you to, to, to build your agent without writing custom components, maintaining them and updating them, fixing security issues. Uh, so if you already have the data platform and you exactly what the agent must do, I don't, I don't think you need to, to hire a, a team of experts to do that.

But there is quite a lot of work to do to get there, I think.

Francesco: Ben, what do you think about it? Uh, what is the biggest pro and biggest con, uh, when treating with, with ai, [00:33:00] with a non-technical team? 

Ben: Well, I think, um, I look at it sort of from two angles and whenever I'm kind of looking at my business or. The projects that I'm working on with, with my clients, I'm always wary of some of these new solutions that come into the market and say, Hey guys, you know, we can, we can improve your floor pricing and we can, you know, we can do all of this magic, uh, improvements to your business.

And I think. A lot of the time, unfortunately, these businesses just want to be integrated with your business to take your data essentially. You know, ad tech has a lot of issues around revenue and data leakage, so I think just being very cautious about some of these magic bullets that exist in the marketplace.

Number one. So, and that's usually is a shortcut. Like, oh, instead of building it ourselves, we can just license this third party. Just be very careful who you are letting into [00:34:00] your, um, into your technological environment. And then the other piece, uh, is. I like to see improvements happen with low hanging fruit, so getting the humans in in your team to do what they're great at, which is human to human interactions.

A lot of the low hanging fruit that can be automated in your business is things like reporting. For example, you know, reporting is always manual and clunky and takes longer and has to be checked. Agents that can run those reports and generate those reports, and then the human can. Check those reports and maybe adapt them for the particular client so they can actually spend their time communicating that report with the client and building that human relationship.

And again, there's some great technologies out there. Tableau was the big one from last decade, which has now been acquired by Salesforce. But once you start to build out your data [00:35:00] strategy, there's great use cases for about how agents can, can make a quite a, quite a big step change to your business without huge in huge investments.

Uh. 

Francesco: Data is the key in the end that, that's a mantra. Uh, I, I, I think I remember from 2019 and I, I'm thinking about Italy 2019. So we are not at the edge of trends as we were speaking about data driven strategy versus data strategy. Whereas one is actually, uh. Something that is meant to create the data to be exploited.

The other one was simply the, the decision making based on, on data. So I believe we are in, in the first stage and in, in, where data is so important that you need to have investments in place to make that available, to really build on top of it. Uh, any technology, any model that you need to go for, uh, needs to rely on, uh, compliance and reliable data.[00:36:00] 

The compliancy is not necessarily the smallest part. Um, okay. Uh, regarding, let's say, uh, let's go back to user behaviors and, uh, how users behave online, uh, we said is changing quickly and especially from, uh, traditional search like, uh, Google, Google search, uh, any kind of browser to interfaces like ChatGPT, where you're prompting and again.

You are, uh, making much more of an effort when, uh, you are putting your own intent into. And that's what a lot of the advertising world is looking for. 'cause many of the signals has been, uh, overused. So this intent is fresh, fresh air for, uh, for the system. So Ben, how do you see, uh, this shift from a traditional search to prompting impacting content discovery and the way media business create value?

Ben: Good question, and it's a big one for the industry because, you know, if we kind of look [00:37:00] at how content creators are, uh, are categorized, you've got the open web, you've got walled gardens such as meta, TikTok, et cetera, and then you have. Hedged Gardens, which are a hybrid of the two and open web. So, you know, anyone ad funded editorial content has been under threat over the past couple of years because approximately 25% search traffickers has gone, is gone into Gemini, it's gone into chatGPT, it's gone into Claude. And so what that means is any business which was set up for page impressions has had a, a really choppy time over the past couple of years. You know, one of the things is for the traffic that you do have coming to your site, make sure you're monetizing it to the maximum yield possible.

You know, have a look at how you monetize your content, where you are leaking data, where you are leaking revenue. Things like add context, protocol [00:38:00] designed exactly to give more yield back to publishers actually. So AI is doing good in, in that regard. And then for the large language models, you know, open AI are now experimenting with what an ads business can look like for them.

So we're already starting to see ads come into the large language models. Now, Google has clearly transformed its business into an AI business, and that's reflected in in their recent valuations. Um. I'll finish on this though. The, the opportunity for legacy publishers is to make sure that you are protected, that you are not having crawlers index your site for the benefit of the large language models.

So make sure that you are protected. Um, CloudFlare offer a suite of products and services to publishers in that regard, but there's also an opportunity through Cloud Fair's Suite. You can actually. Um, go into an auction with a large language models where they can actually [00:39:00] bid to train their model on your content.

So there's a monetization aspect there. And then also look at indexing your content so it is compliant with the, with the LLMs. You know, wouldn't it be great if you are a big publisher, whether you are a news publisher or a lifestyle publisher? If a user is prompting in chatGPT or or Gemini, your content was to be discovered there.

So I think there's a big opportunity for how search optimization looks in the future for publishers. And again, I'm seeing a few projects where publishers are looking to see what the future state of, of of search looks like for their business. 

Francesco: It is, it totally is. Uh, I've been talking to, to people here in Singapore, and one of the questions that I got is, how do I optimize my website or my, my e-commerce appear more on, uh, LLMs results.

And that is a service per se. It [00:40:00] used to be, say, O now is, I believe, called Geo. Uh, but that is how you try to adapt from a publisher side. You, you really want to monetize. And people are starting to see on, uh, Google, uh, uh, on, on Google platforms that part of the traffic is coming from, uh, LLMs. So that is also something that we need to, to understand and, uh, it has an impact on a programmatic side.

Both on the strategy and how you can use that information to, to really understand the, the results of your own campaign. And I think we are, we are close to the end. Uh, as we wrap it up. Uh, I want to thank, uh, both Luke and and Ben for, uh, their patience, uh, their contribution and their knowledge, and, uh.

It's clear that the transition that we're leaving now from a click to agent is the biggest infrastructure shift, uh, since, since mobile maybe. And, uh, at Tangoo with, uh, our own team and [00:41:00] our own understanding on, on, on the industry of programmatic, we are starting this technology to, to, to make sure that we can work with partners and test these agent interfaces.

And we want to test it in production, as we said before, the difference between something that works and has an impact, uh, against something that has been prototyped for, uh, for the sake of prototyping. So, uh, we feel we are well positioned, uh, both with our, uh, own connection with IB Tech Lab and, uh, both the publisher and, uh, and, and the brands were.

So we have a full perspective on what the programmatic. Environment looks like today and might be look, lie, look like to tomorrow. So we want to really contribute at both a brand and, uh, institutional level so that LLM as a new programmatic channel, uh, might arise and, uh, and be driven by agents. Uh, we don't really, uh, believe right now that we have, uh, [00:42:00] or anybody has an off shelf solution, uh, for a revolution based big.

And we are looking for more inputs and, and partners to co-create these standards. So, uh, what I would really, really invite any listeners to do is to get in, talk in contact with us. Uh, we are searching for, uh, someone who wants to deep dive on this edge technologies and might have an interest in monetize, uh, their own traffic through agent.

And of course, if you have any questions for our wonderful. Speakers, uh, Ben and Luca, feel free to, to reach out to, to our channels and, uh, we'll try to, to answer to, to the best of our capabilities. I think it's time for goodbyes. So thanks again and have a wonderful morning, afternoon, and, and evening,.

Luca: Thank you. [00:43:00] Bye.