Latest / Signal & Noise: Bringing Clarity to Media, Data, Tech and AI / RampUp Requiem: Rio Longacre & Krish Raja on Identity, Data Collaboration, and the Future of AdTech
Transcript
- 0:03This is signal and noise bringing clarity to data, tech
- 0:07and AI. All right, welcome to Signal and
- 0:16noise at live ramps ramp up in San Fran.
- 0:20It's been an incredible event talking about data
- 0:23collaboration, data partnerships in the industry and obviously a
- 0:28lot of AII have here. Rio Longacre.
- 0:31Thank you so much Rio for for having the hosts at this event.
- 0:36We had an almost like a star-studded lineup of really
- 0:41fun and interesting and smart leaders all the way from the IAB
- 0:46all the way from the NAI to, to health data practitioners
- 0:50talking about, you know, what they feel is happening in the
- 0:53industry right now. And so, yeah, I want to hear
- 0:55about your take on on RAMPA and and what you've heard today.
- 1:00Well, first of all, thank you for hosting.
- 1:02They did an awesome job. We have some great interviews
- 1:05with really smart, interesting, diverse group of thought leaders
- 1:08with different opinions and like some really cool insights.
- 1:10Everything from policy to privacy laws to targeting to
- 1:17identity, you name it, it's really good discussions.
- 1:19Everyone should look forward to that.
- 1:20That'll follow this introduction here.
- 1:23So I guess a couple quick things on on ramp UPS.
- 1:26I've been going a number of years.
- 1:27Great event and I, I #1 is, I know it's not a considered an ad
- 1:32tech tent pole, but it's one of the main ad tech events.
- 1:36It's run by live ramps. I guess that makes it a little
- 1:38different, but it's, it's really well attended.
- 1:41So a lot of the same people I see at a lot of the ad tech
- 1:43events plus plus maybe some more.
- 1:46I've met a couple brand clients here.
- 1:48So I think they've been maybe more successful seeing getting
- 1:51some brands, a lot of I take people, a lot of agencies.
- 1:54I think it's a really good mix. It's around.
- 1:56I think they said it's around 3000 people.
- 1:58Wow. So it's a huge attendance.
- 2:00It's a good size and they're key takeaways.
- 2:05So I think in fact, I have some notes here.
- 2:06So like, I guess like, like #1 like it was the data
- 2:11collaboration. Not surprisingly, it's a really
- 2:14big topic. Yeah.
- 2:16And at the that I noticed a lot more brands here and talking
- 2:21about successful implementations this year than than Prime.
- 2:24Maybe that was just me seeing sort of yellow car syndrome, you
- 2:28know, see it because I talk about it.
- 2:29But I definitely felt like brands lent in more and we're
- 2:33talking about, you know, real solutions that are being
- 2:36implemented through through collaborate collaboration.
- 2:39That's true. I think data cleaners have gone
- 2:41from this thing that everyone thought was a great idea for a
- 2:44product that really first few years struggled to get a lot of
- 2:48adoption. And I even wrote that article
- 2:52clean room gloom about a year for COVID.
- 2:54They got a lot of traction about and maybe not meeting
- 2:57expectations terms of in terms of usage and like what are the
- 3:01actual use cases? But I think we're at a point now
- 3:03where it's more of it's become infrastructure and I heard that
- 3:06term a couple times. Yes.
- 3:08Cleaners of income infrastructure it's important
- 3:09part of infrastructure think for live ramp you know they made the
- 3:11hobble acquisition a couple years ago.
- 3:13Some people at the time like we're criticizing it, saying
- 3:16they may be overpaid, but look, it's part of how they go to
- 3:20market now. It's part of how they're,
- 3:22they're not just saying we have identity we have on board.
- 3:24You're saying where it's part of this infrastructure that we're
- 3:27offering so they can identify better to precision advertising,
- 3:32know who their customers. There's a lot of, there's a lot
- 3:34that comes out of that. And I agree with you.
- 3:36We're starting to see some actual implementations of, of
- 3:38data cleaning rooms. I'm working on three right now
- 3:40for example. Yeah.
- 3:41Wow. And I think the word
- 3:43infrastructure that you said there is definitely a really
- 3:47resonant topic where data and identity moves from being just a
- 3:53signal that is found somewhere and you know, copied and moved
- 3:57on in the industry. And we try and match it here and
- 3:59there and, and use it where we can to, you know, data and
- 4:02identity becoming like you say, infrastructure where it's a,
- 4:06it's a really foundational layer of any enterprises business and,
- 4:11and that is what's going to really make collaboration
- 4:14possible and secure and scalable and monetizable, I think.
- 4:19Identity, you just brought it up.
- 4:20I think that was the other big trend.
- 4:21You know, it's not maybe been the sexiest topic the last few
- 4:24years, but identities made a comeback.
- 4:27This realization that identity whether it's ramp ID or UID 2 or
- 4:32or any of these identity spines really are the way that we are
- 4:37going to not only measure but the target, especially as media
- 4:42gets more fragmented, we've got more platforms and so many
- 4:45places you can buy, especially with the rising retail media and
- 4:47CTV in recent years. So I think the role of identity
- 4:50getting kind of making comeback coming, I think it may be even
- 4:55more important, right. But then also a trend I'm seeing
- 4:59is like the center of gravity shifting back a little bit on on
- 5:03on 1st party got an identity back from marketing to ITA
- 5:06little bit. Yeah, I'm seeing, you know,
- 5:09realizing OK, if we are shifting this kind of composable 0 copy
- 5:11world, like there's going to be cloud data warehouse, where's ID
- 5:14going to sit probably closer to that.
- 5:15Yeah, where where's all the decisioning are a models getting
- 5:18built on top of probably there and who's a steward of that?
- 5:20It's IT Yeah, right. So I think there's like a center
- 5:22of gravity shifting back and and I expect identity to continue to
- 5:25be more prominent. Yeah.
- 5:27Yeah. And that, that sort of sits in
- 5:29line with identity as infrastructure, which yeah, is
- 5:31always managed by engineering IT in that way.
- 5:34But yeah, I think that was, that was a topic that Shaylee brought
- 5:36up in the, the, the IB tech lab. The CEO talked about their their
- 5:41initiatives to to propagate those signals inside the CDN
- 5:45because that is your layer of control as a as a business.
- 5:49Yeah, I think we're related to that too.
- 5:51Like if that's your layer of control, if that's where you
- 5:53have identity, if, if that's, I mean, that's probably where
- 5:57you're going to be building out your AI, your AI models from.
- 6:01And in order to do that, you need good data.
- 6:04Yeah, in order to have good data, you need identity, whether
- 6:06it's adjustability for targeting, for measurement, for
- 6:08all these things. I mean, it's really, it's really
- 6:10the foundation of it. So live ramp kept hitting at
- 6:12home in a lot of older sessions and I thought that was
- 6:15effective. Yeah, I think it was it one of
- 6:17the, you know, throughout the the conference itself, I think
- 6:20one of the themes that kept hitting home, obviously AI is
- 6:23just there in every single sentence now, every
- 6:25conversation. But it was, it felt like
- 6:29everyone's really starting to understand how data identity and
- 6:34AI collaborate with one another or come together and what they
- 6:38are to one another. Because obviously we know the,
- 6:42the catch phrase, you know, AI is only good as the data you put
- 6:45in. But it's, you know, now we're,
- 6:47we're moving past that. Everyone knows that.
- 6:50And I hope everyone knows that. And we're now moving towards
- 6:53what actually is data to AI in, in this world.
- 6:57And I think you know, Scott from live ramp said it really well,
- 7:00which was, you know, if you, if you leave your car unlocked,
- 7:03your, your possessions are going to get nicked.
- 7:05And so it in in the same way, if you consider data to be a
- 7:09valuable asset to your business, lock it up and for many.
- 7:13Friends, they'd argue it's one of their more valuable, if not
- 7:15the most valuable asset, right. And I, I think one of the
- 7:18airlines, because United were saying a couple of years ago
- 7:20they that their frequent flyer filed like the, the data they
- 7:25have on the frequent Flyers for the one care program is
- 7:27potentially more valuable than the airline, right, which is
- 7:31airlines aren't super profitable business, but they have the
- 7:33business. That's an absolutely great
- 7:37example of this where, yeah, like you said, this is actually
- 7:40translating into the money, the valuation of that business
- 7:44division. Like you say, sometimes that the
- 7:47the frequent flyer division is propping up the entire valuation
- 7:51of the airline. Well, you're not only can they
- 7:53use that to understand who the better customers are and offer
- 7:55them, offer them concessions to get them to spend more and more
- 7:58loyal, but you know, they didn't begin to monetize that in
- 8:00different ways as well, whether it's through advertising or or
- 8:04yeah, data syndication. There are lots of other ways to
- 8:07which I think is interesting. And I guess the last point
- 8:11probably worth bringing up was while we've seen clean rooms go
- 8:15for more like, OK, an idea to actually gain you some tractions
- 8:18as more of an infrastructure than a standalone product.
- 8:20I think for AI, we're starting to see, at least I saw first
- 8:23really good examples of, and everyone's been talking about
- 8:27agentic trading. How's AI going to actually
- 8:28impact the weather's planning, measurement, activation and
- 8:32management. You know, all the things that
- 8:34the media so all seem kind of speculative up until now.
- 8:37I saw the first really good examples of, OK, how would this
- 8:39actually look in action weather? Some of the work IB tech labs
- 8:42working on I Newton research demoed me some really, really
- 8:45interesting things where they're looking at starting with
- 8:48measurement, then going back feeding results into, into
- 8:50better planning and taking the planning and pushing it actually
- 8:53into into activation. So I thought first time starting
- 8:56to see these things become real. I think that's very exciting.
- 8:59Connect together. Yeah.
- 9:00Yeah, absolutely. I think that you know that.
- 9:02All of it. For me anyway.
- 9:04I don't know. I just sort of look at how the
- 9:06trends of all the topics we've heard about today and yesterday
- 9:11come together. And for me, it helps to paint a
- 9:14picture where advertising and subscriptions and future forms
- 9:21of monetization paper crawled. They're all just symbiotically
- 9:25part of 1 monetization strategy for, for a business as opposed
- 9:30to what we've had typically in the last 20 years, 30 years or
- 9:34whatever, which is, you know, your ads business is there, your
- 9:36subscription business is there. Your loyalty program might be
- 9:39there. Way over there, absolutely not
- 9:42connected in the slightest. And I think that that that's
- 9:45kind of exciting for me, someone who's been working in, you know,
- 9:48the data space for a long time is that it becomes the literal
- 9:53connective tissue between for a business to monetize as opposed
- 9:58to just have an ads business. And I think that's meeting
- 10:00that's going to meet the consumers where they want to be
- 10:02as well, because hybrid monetization structures are
- 10:05going to be really, really important.
- 10:07And knowing your customer maybe is anonymous or logged in or
- 10:11could be a subscriber one day, but isn't yet.
- 10:14I think all those things are going to be really important if
- 10:17publishers with declining traffic want to treat their
- 10:20users like they know them and actually get them to sign up or
- 10:24actually get them to subscribe. You know, all of that needs a
- 10:28data flywheel that that sees across it all, which I think all
- 10:32of this stuff points towards. Now points points towards being
- 10:35able to actually use that data. We talked about full funnel and
- 10:38customer journey and we talked about this stuff for years,
- 10:40right? That was becoming becoming maybe
- 10:41more realistic. Now I'd actually do that.
- 10:43I think you're right, publishers, they play a right to
- 10:45stand to benefit. But I mean it's at what's what's
- 10:48their law of the last 25 years. Publishers always get.
- 10:50Hosed. Yeah.
- 10:51So hopefully the changes this time.
- 10:52But yeah, I hope so. I think so.
- 10:54I'm optimistic on that front for sure.
- 10:56The publishers. All right, good.
- 10:58Any other any other thoughts from I mean you're going to ramp
- 11:00up before, is this your first time?
- 11:02No, this was my first time actually.
- 11:03Yeah, I'd, I'd done a lot of, you know, I'm new to New York.
- 11:06I've been here for a year and a half.
- 11:09But yeah, I'd certainly say it was, it was just full of ideas
- 11:12and, and really smart people. So I'm, yeah, I'm super glad I
- 11:15came. Very good.
- 11:16Yeah, Well, thanks again for hosting this and for everyone.
- 11:19And we'll have a bunch of other great content to me on Signal
- 11:23Noise, including every week we've got the pod and we've got
- 11:26some really great feature length articles that we'll be we'll be
- 11:29publishing over the next month or so.
- 11:30So stay tuned and thanks again. Stay tuned.
- 11:33Thank you so much, Rio. So we're at Ramp Up San
- 11:37Francisco 2026 and I'm here with Daniel Block.
- 11:41Who's head of corporate? Business development at Fetch
- 11:45Now. I'm an Australian who's just
- 11:47moved here and I'd not heard of Fetch.
- 11:50Do you get that a lot? At in Australia, yes, in.
- 11:54The US as well. We're we're starting to get a
- 11:57lot more traction in the United States.
- 11:59Company's been around for a long time and has amazing brand
- 12:02recognition in the consumer space.
- 12:05Millions of people come to our app every single day to live
- 12:08rewarded and earn our currency, which is called Fetch Points.
- 12:11And really excited to talk with you about not only what we're
- 12:15doing in the on app environment, but how we're leveraging and
- 12:18monetizing that asset off off platform as well.
- 12:21Yeah, I mean how much in terms of consumer data give us some
- 12:26numbers as to the the scale of fetch.
- 12:28Yeah. So Fetch is in a really unique
- 12:31position from a data perspective.
- 12:33We're seeing 13 million receipts a day, which is amassing to $215
- 12:38billion worth of GMV or sales data on an annual basis at the
- 12:43SKU level. So we're talking product,
- 12:46receipt, brand, store, etcetera. And what's really unique about
- 12:51all of that GMV is that actually comes out to amassing the scope
- 12:56of data to be the third largest retailer in the United States
- 12:59besides Walmart and Amazon. Wow.
- 13:02That's an incredible claim. And so you must be, you're
- 13:05obviously sitting on this goldmine of of data and in your
- 13:09role you're commercializing it. You're, you're turning it into,
- 13:13I guess the next era of your business with an asset that
- 13:16compounds and sort of helps you to, to do better customer
- 13:21experiences as well. So tell me a little bit about
- 13:23that, because obviously been you're very famous on the
- 13:25consumer side, you're now at ramp up.
- 13:27So you're, you're talking about data collaboration.
- 13:29You were on stage just yesterday.
- 13:31So give us a little recap of why you're in at this conference and
- 13:35what your, your, I guess your narrative is here.
- 13:38Yeah, I mean, I think you nailed it right.
- 13:40So the one of the key focuses for us and then initial growth
- 13:44lever of our business is really highly focused on this exhaust
- 13:48data that we have access to. We've sit on this treasure trove
- 13:52that we feel is completely unique and differentiated.
- 13:56It's unmodeled and the scale, breadth and depth of the asset
- 13:59is something that we feel like is a huge asset to brand
- 14:03marketers. And for us, what data
- 14:05monetization means is actually using the data for more
- 14:09effective targeting, but also closely measurement and
- 14:11attribution. The interesting thing about
- 14:14Fetch is because our relationship is with the
- 14:16consumer, we've got unfettered access to this receipt data that
- 14:20is online physical scans of receipts.
- 14:22But that's also, sorry, that's offline physical scans of
- 14:26receipts, but also online connections with the likes of
- 14:29Gmail and Amazon and and a bunch of other providers giving us a
- 14:33complete picture of consumer spending.
- 14:36Right, OK. And, and your session yesterday
- 14:39was with an audio network, yeah, if I'm not mistaken.
- 14:43So if I so, you know, in my experience with, with data
- 14:46collaboration and create clean rooms, usually the easiest way
- 14:49to do all that is cookies and you know, just, you know,
- 14:52standard collaborations on segments and things like that.
- 14:57I wasn't actually out the session yesterday, but if I'm
- 14:59thinking about who you're talking with, you've got a
- 15:03unique data set that is online and offline and an image
- 15:06captured and lots of different formats of information and
- 15:10you're presumably discussing matching into an audio
- 15:15environment. Yep, this feels complicated.
- 15:18Talk to me a little bit about the layers above and beyond that
- 15:22you guys have gone through to to stand up such AI guess complex
- 15:27and sophisticated collaboration. Yeah.
- 15:30So for us, we've got 100% coverage over a number of
- 15:35different identifiers, which really helps us from an identity
- 15:38perspective understand who that consumer is.
- 15:42So consumers when they sign up for Fetch, things like first
- 15:45name, last name, e-mail address and phone number, we're here at
- 15:48Ramp Up. One of the unique parts about
- 15:50our relationship with Live Ramp is all of that data and
- 15:53information comes into a waterfall and gets resolved to a
- 15:57unique ID, like a Ramp ID. We also have other identity
- 16:00providers such as UID too. So for us being able to match
- 16:04any of our unique identifiers to exposure logs to be able to
- 16:09prove outcomes for brands is something really unique and and
- 16:13differentiated. On the audio side, you know
- 16:16those exposures are to your point a little bit complex and
- 16:19complicated. SiriusXM also happens.
- 16:22So in Pandora, Pandora does have a logged in user.
- 16:24And so those digital formats allow for more effective
- 16:28targeting and measurement. I think early, early stages of
- 16:31how Sirius is thinking about data monetization and their
- 16:34their commerce media play. I'm not going to allude to what
- 16:36they're doing on their side, but for us, you know, we want to be
- 16:39able to meet buyers where they are.
- 16:41So we want to take any impression and match it to our
- 16:45verified transaction data to prove sales return on ad spend,
- 16:49conversion rate and ultimately incrementality.
- 16:52So leveraging the power of the Fetch data, we feel very bullish
- 16:56on our opportunity to do that at the SKU level.
- 16:58And you know that that's the the product that we want to give to
- 17:02brand marketers. Yeah, absolutely.
- 17:03And it's like I said, you've, you've obviously been sad On
- 17:06this date for a while and I'm sure you've been shouting from
- 17:09the rooftops to to, you know, use it better use it more
- 17:12collaboratively in in markets. I guess one of the, you know,
- 17:16we've covered a lot of the goods and all of the the good things
- 17:20about data collaboration, what you want to achieve.
- 17:21Let's talk about the bad and the ugly, not from a company
- 17:25perspective, but more just you're obviously a very tenured
- 17:28executive in the industry. Just observationally, what do
- 17:31you think about data collaboration?
- 17:33When are they, are they topics or are there outcomes that you
- 17:36just sit there and you're like, this should be better by now or
- 17:39this, you know, just just almost a frustration that you've held
- 17:43in and around the topic of data collaboration that you would
- 17:47like to see resolved with a magic wand.
- 17:49So I think that there are two things.
- 17:51One, I'll talk a little bit about live ramp because we're
- 17:54we're here and we're also partnering with them on this.
- 17:56So cross-platform measurement, so the ability to actually
- 18:00measure at scale I think is very unique.
- 18:03And I think the next wave of digital outcomes being proven in
- 18:09offline transaction data being able to help kind of facilitate
- 18:12that. So for a long time, measurement
- 18:14has been done in silo. This platform delivered this
- 18:17return on ad spend. This one did this, but there's
- 18:20no real all of my publishers in one kind of clean room box to
- 18:25help me understand here was my social investment.
- 18:27Here was the conversion rate return on ad spend and Lyft.
- 18:30Here was my connected television.
- 18:32What that allows brands to do is understand things like channel
- 18:35mix, reinvestment into different mediums, pull back, push
- 18:40forward, heavy up. And I think for for starters
- 18:44like the ability to do cross-platform measurement is
- 18:46absolutely something that is on the precipice of really
- 18:50interesting net new opportunities.
- 18:52The inverse of that and one thing that going back to your
- 18:56question that in my opinion needs to really stop is the
- 18:59pendulum has swung so far to folks that have data assets
- 19:05modeling their data without deterministic quality.
- 19:09What I mean by that is if you look at things like the live
- 19:12ramp data marketplace, you might see a transaction data provider
- 19:16have YARN buyers, for instance, they might have 200 million YARN
- 19:22buyers. Is that real?
- 19:25Likely not. So at Fetch, what we're focused
- 19:27on is completely deterministic data, being able to take
- 19:31deterministic seeds that are highly relevant and provide an
- 19:35actual outcome all back to identity.
- 19:38So, you know, as we think about the future of data and data
- 19:42collaborations, building audiences with deterministic
- 19:45data is really where I want to see the pendulum swing back, as
- 19:49opposed to folks just looking for the biggest audience sizes
- 19:52to scale campaigns. Absolutely.
- 19:54Yeah. I think emerging technologies, I
- 19:55totally agree that emerging technology is going to make it
- 19:58easier for people that have data to use it to monetize it, which
- 20:03I think hopefully will start to reduce the need for these
- 20:08slightly questionable, you know, segments of 2 million pet
- 20:13intenders which are like, really?
- 20:16Like, Are you sure? So yeah, I think I think that's
- 20:19definitely going to change the the paradigm to, you know, in
- 20:22favour of data owners like yourself.
- 20:25Now let's final topic, which wouldn't be podcast bingo if we
- 20:29didn't put AI in there a little bit.
- 20:32AI obviously sort of intersects with data collaboration in some
- 20:35ways, but every business is using it in a very different
- 20:38way. So how do you think about it
- 20:39personally amplifying? Maybe it's your part in your
- 20:45company or maybe it's the products that you're monetizing.
- 20:47Yeah. So we made a strategic
- 20:50investment to hire a Chief AI Officer here at Fetch.
- 20:54And his team is running what we call Fetch Labs.
- 20:58They're responsible for all agentic projects that happened
- 21:01at the company. We've tripled down in that area
- 21:04and are highly focused on on that in really 2 distinct
- 21:08fashions. One is delivering better in app
- 21:12experiences. So how do we continue to model
- 21:15and refine what we're doing for ranking and relevance,
- 21:18leveraging all the purchase data to make sure that we're
- 21:21delivering the right message at the right time to the right
- 21:24consumer. So every single app experience
- 21:27looks completely different. That's on the in app side now
- 21:30off platform. What we believe going back to
- 21:32this conversation around data, seeds and and intelligence,
- 21:35there is agentic buying and agentic extrapolation of data.
- 21:41So showing future propensity to purchase a product based on a
- 21:44really strong signal. It's kind of the future of of
- 21:47buying. And we're talking to partners
- 21:49right now to better understand, OK, if someone bought at
- 21:52McDonald's, how do you start to leverage that seed and signal to
- 21:57deliver propensity and future buying behavior and pattern
- 22:01recognition. And that's what we're really
- 22:02excited about in AI. And does that.
- 22:05That's a really good point actually.
- 22:06And does that to, to further, I guess elaborate on that.
- 22:10Does it help you to also illuminate the blind spots where
- 22:14say for example, data collaborations might not reach
- 22:17because they're e-mail matched or they're one to one or they
- 22:20may just not have as much reach in certain channels?
- 22:23Is that something that you look at as well?
- 22:24Like to use AI in that way? Yeah, I think so for sure.
- 22:27I think anywhere that Fetch can bring in additional signal to
- 22:32help better understand what this consumer is doing is certainly
- 22:36helpful. The we feel the greatest
- 22:39indicator of a future purchase is a past purchase.
- 22:42Having that deterministic signal tied to an identifier and then
- 22:45leveraging all the other inputs that might be available to
- 22:49better target and ultimately measure the impact of
- 22:52advertising is really what it's all about.
- 22:54And for us, leveraging this conversion data at scale is
- 22:57completely underutilized. And we want to make sure that
- 23:00instead of a brand marketer going back to their CFO and
- 23:04saying, hey, I delivered 3% incremental reach on my linear
- 23:08television investment, they can go back to the CFO and say my
- 23:12investment here in media this past quarter drove FS in sales.
- 23:18And that is how you can continue to defend your investment.
- 23:21And in a world of outcomes, there's no better signal than
- 23:24than Fetch to help brands do that.
- 23:26Oh, absolutely. You're going to make a lot of CF
- 23:28OS very happy, for sure. Yeah.
- 23:31So to just finish off, is there anything that anything that
- 23:35either you didn't want to say on stage, a hot tape from yourself
- 23:39that I thought I, I will save that for, for another time that
- 23:42you want to leave us with or anything else that you've
- 23:45observed throughout ramp up that really resonated with you and
- 23:48maybe a sound bite that that you found meaningful?
- 23:51Yeah. I mean nothing that we didn't
- 23:53say yesterday on stage. I think the areas of interest
- 23:58for Fetch that we continue to be bullish on and and definitely
- 24:02continued investment in Connected Television.
- 24:04And ultimately folks are getting more sophisticated with how
- 24:08they're buying that medium. And we think the next iteration
- 24:11of that is how do you continue to get more sophisticated with
- 24:14not only the buying mechanisms, but the measurement and then
- 24:17ultimately feeding into things like real time optimization to
- 24:22better serve ads to the right consumer at the right time.
- 24:26And I think that's a, a medium that's ripe for disruption,
- 24:29obviously the fastest growing medium that's that's happening
- 24:34right now in media investment. So continued investment and and
- 24:38watching things change and in that space has been really
- 24:40interesting. Awesome.
- 24:42Well, look, thank you so much for your time, Daniel.
- 24:44It's been a real pleasure to chat to you and hear your hot
- 24:47takes on on what's going on in data collab.
- 24:49So appreciate it. Thank you.
- 24:51Cheers. Hey, we are at Live Ramp San
- 24:55Francisco, Rampart 2026 and I've talked with me, John Mangano.
- 25:00Thank you for joining us. And you're from Symantec,
- 25:04Symantec IQ? Yeah, it's semantic.
- 25:06Semantic and you've got a really fascinating specific ink that I
- 25:12really want to get into today and you're also podcastos
- 25:16yourself. So I think it's going to be a
- 25:17really cool back and forth on this top off and the topic is
- 25:22health data and specifically how that relates to Theta plan,
- 25:27voting, privacy also. So give us a little background
- 25:30to yourself 1st and haven't got into this niche.
- 25:34Well, you know, I for the most part fell into it where I think
- 25:37most people do in the health space.
- 25:40Jess was a marketer and had an opportunity to help a company
- 25:44work within the pharma space. And the need that the market had
- 25:48at the time was to be able to quantify and really justify
- 25:52their digital campaigns in the pharma space.
- 25:56Work for a company called Tomskirk, which did measurement.
- 25:59And we had to build the way that media which was a very unique
- 26:03data set was analyzed within the health data which has so many
- 26:10HIPAA regulatory privacy restrictions, but to turn that
- 26:15information into very compliant but insightful findings for
- 26:20those campaigns. And we were able to prove when
- 26:23you market to patients, they, they don't just talk to the
- 26:26doctors, but they get home all. Right.
- 26:29And and forgive my, I guess not Vigorance, but just not that
- 26:33depth of understanding is would you say the health advertising
- 26:37and picker and Uber, I guess community bikes and in order to
- 26:42use data in that space, would you say sort of where the rest
- 26:46of the industry needs to catch up to or is it just a totally
- 26:49second place? Well, with health data, there's
- 26:52a lot of very sensitive information about our level
- 26:55thing. It certainly should be protected
- 26:58and it and it is. And that doesn't mean it can't
- 27:01be used. So understanding that a certain
- 27:04population is set so large for people has a condition and what
- 27:08they have in common helps us as care providers figure out how to
- 27:14make them feel better. So it's important to be able to
- 27:16use that data. But in our space, we don't
- 27:20really care what an individual has.
- 27:22I mean, we do on an individual basis, right?
- 27:24But everyone as well. But for our research, we don't
- 27:26need it. In fact, individual data doesn't
- 27:28help us. It's knowing the big trends of
- 27:30them. There's thousands of people who
- 27:32have a certain reaction to a grant and do it when we get on a
- 27:36treatment that is a success because we know they're getting
- 27:40older. And and that, I guess, is a an
- 27:44interesting segue into what AI is doing to this space indicted
- 27:49for them. Now, I as a consumer have
- 27:51noticed a lot of, you know, just from the early days of using
- 27:55chat TPT, it was, you know, regardless of whether the
- 27:59information pointed back to me was good or not.
- 28:01There's more this I've I've had somewhere to talk about myself
- 28:07issues or maybe offload a picture or whatever.
- 28:09So I guess what what I'm interested in viewing your
- 28:12perspective on is how does how does AI change our health data
- 28:18gets used and what you see in that space from your.
- 28:22So on the side what health data can do disconnecting, right?
- 28:27So it used to be maybe 5-10 years ago you research a
- 28:31condition and you'd learn about the condition in general terms.
- 28:35In the new world you can not just research a condition using
- 28:40chat, but upload maybe your testing, yes upload what your
- 28:46history has been and chat. You can start to look at it and
- 28:50identify trends. So it could be that yes, you
- 28:54have this addition, but it also is tied to your BMI and your,
- 28:58the BMI has gone up lately, which is something I personally
- 29:02am dealing with. And it helps you understand like
- 29:04what the causes are versus 1 dimensional understanding.
- 29:09Because all of a sudden it's like, Oh no, I just need to lose
- 29:11weight. I don't actually have a
- 29:14condition like this. Or if I lose weight, the
- 29:16condition will reverse. In health data, it's the same
- 29:19thing. Instead of looking at the
- 29:20patient, companies can do their analysis using that information,
- 29:25that type of multi variant analysis that them put all the
- 29:29pieces together. Yeah, where I've had teams do
- 29:32it. Now you can have a machine do it
- 29:34in minutes, you know, and have that kind of result, yeah.
- 29:38And that's that's a really good point actually.
- 29:40And I also imagined that there's there's also a use case for
- 29:45looking at health data over its ion for an individual name where
- 29:50you know, for example, three, I say that is a lot of health
- 29:54conditions or a lot of health addictions on adequately dealt.
- 29:59With say by my general. Practitioner just because
- 30:01they're looking at I'm saying to them in that appointment at all
- 30:05what I've been feeling over the last.
- 30:08Year exactly. So if you look at a health care
- 30:10provider, they only see what maybe you tell them and how you
- 30:15might tell them how you feel maybe different based off the
- 30:19whole response. Some cultures you don't talk
- 30:22about your health and your name. Other cultures, you talk about
- 30:26it a whole lot. And it's very different, right.
- 30:28So you have that piece where the aunt can actually help the scorp
- 30:33or what? Someone to say.
- 30:34And I, I had this situation happening, dear friend of mine,
- 30:37she used to work for me and she resigned and she's from Russia.
- 30:41And yeah, resigning to someone you like is so hard.
- 30:45And then there's also the cultural differences.
- 30:48And I wrote down notes on what she told me and I I really
- 30:51couldn't understand why she was resigning.
- 30:53I even shared that with our head of HR.
- 30:55The note taking tool that I used in my iPad a month later came
- 30:59out with an AI interpretation module.
- 31:02So when we see what my friend really meant.
- 31:05And by then she had already had second and third conversation
- 31:08with me. And I I knew didn't interpreted
- 31:11it to exactly what she finally told me.
- 31:14You know a monthly so that is where you are when you think
- 31:18about that's kind of nice for the situation I had.
- 31:21But if you think of a doctor talking to a patient, having a
- 31:25tool that helped them understands the degree of
- 31:29expression of is really. I'm some type yeah the context
- 31:36within rich staff request or data .6.
- 31:39And then add that in with all the other specialists and, and
- 31:42the full history of what my health record tells me.
- 31:47So now it's looking at what every other Doctor Who might
- 31:50have seen the same symptom across the country and has found
- 31:54helps that doctor understand more.
- 31:57There's a lot of medicine is tied to my genetics.
- 32:01And you know, you and I have different genetics.
- 32:04My wife has different genetics. When a doctor is looking at it,
- 32:06they can't take all of that into account necessarily and they may
- 32:10not have the sample of experiences with patients with
- 32:13all the different types of Physiology.
- 32:16But now AI can pull the data passionately and start saying,
- 32:19oh, we've got little trends. Someone who is like John is more
- 32:23likely to do better on this brand of drug versus someone who
- 32:27was like my wife. And that makes such a just well.
- 32:29Absolutely, yeah. That's where I could define to
- 32:31see it going. And so to bring that back to Ram
- 32:34Hartman data collaboration, what are some of the, I guess, what's
- 32:38the one or two things that Someone Like You asked to deal
- 32:42with and think about more so than someone who's not in health
- 32:45data when it comes to data collaboration?
- 32:47So for me data collaboration is 2 pieces, rates to health data
- 32:50as well as the media or other type of data.
- 32:54And the challenge is they're often both very restricted in
- 32:57how you offer HIPAA with health and and just for a data
- 33:01ownership within the media side. Live Lamp has done a really good
- 33:05job at creating the environment, not just for a clean room for
- 33:09that data collaboration, but now having the ability to have third
- 33:12parties, great analysis associated going through those
- 33:17different data sets. So that's something that we're
- 33:19working with at Semantic is looking at helpful data that a
- 33:23client might have put into the clean room as well as media
- 33:26exposure from 20 different publishers to then come to the
- 33:30understanding as to how to optimize the campaign or what is
- 33:33working better. OK, that's like, yeah, that I
- 33:36guess that's a really. Fascinating area to, you know,
- 33:39tie all to harmless data or outcomes or public data, you're
- 33:44out. And ultimately, health marketing
- 33:50is successful if it means people are getting healthier.
- 33:53And using this health data, we can see the degree to which
- 33:57people who see ads talk to doctors and then ultimately
- 34:02start a new treatment and even see how long they stay on that
- 34:06treatment to see if they're being treated properly and their
- 34:09outcome is approved all within in a HIPAA compliant, anonymous
- 34:13way. So we know in general that
- 34:16people's health has improved as a result, but we don't know who
- 34:20it is, which is so important to write.
- 34:23Well, I mean that the. Pursuit of better personal
- 34:26health is a really great 1, so I really appreciate your time into
- 34:30your calendar. Thank you.
- 34:31Thanks for having me. All right, so we're at ramp up
- 34:352026 in San Francisco and I've got new point here from the NAI,
- 34:41which I want you to break down for me and help me understand
- 34:44what yes, but firstly, you've got a really interesting
- 34:47background. When I look at, you know, policy
- 34:52and privacy, there aren't too many people or sit across both.
- 34:58Lots of people have either privacy or policy and obviously
- 35:01they're bleeding together, especially with what's happening
- 35:03to the Jewel, so. Just give us a.
- 35:05Little intro to yourself and how you got.
- 35:07Yeah, thanks. Well, I'm so I'm Lee Ferrand.
- 35:10I'm the President of the CEO of the NAI.
- 35:12The NAI is a self regulatory organization and trade
- 35:16association that represents on technology in the, as you
- 35:20pointed out in the areas of consumer privacy.
- 35:24So I came to the NAI 11 years ago from AOL where I spent a
- 35:30dozen years, but hunting their advertising teams from a legal
- 35:34perspective, and then my last couple of years at AOL waiting
- 35:38their public policy. So that gave me a real insight
- 35:42into not only the business of attack and the technologies and
- 35:47the contracting and partnerships, but then as head
- 35:50and global public policy and it's over to understand policy
- 35:54implications of how policy and law making impacts.
- 36:01So at the NAR, moving over to the NARC.
- 36:06And given the AOL is one of the OG's, let's say it's been around
- 36:12since, you know, since certainly since I started.
- 36:15I agree with this as one of the incumbents and your role there
- 36:19was looking after. Yeah.
- 36:22So my first few years were spent representing the legal teams or
- 36:27the advertising teams from the legal political.
- 36:30So I did all of the partnership deals, the contracts and when
- 36:34that thing got to split this tune, it was data and his
- 36:38privacy representations of doing deals.
- 36:41Right. So it was almost, it was almost
- 36:43a byproduct of the job you had to get done, but you needed to
- 36:46sort this bit out. Yes.
- 36:48And then, and then they asked me to settle into the role of head
- 36:51of public policy. Yeah, policy intersected in
- 36:56really interesting way. It is something if you're
- 36:58talking about the privacy, one of the first things that I was
- 37:01involved with in that role was the reform government
- 37:04surveillance coalition. Governments are doing people's
- 37:11doing that and then whether or not you know, companies like in
- 37:15about but having a say how government is.
- 37:23So that was a really interesting had an opening experience and
- 37:28now you know you're Fast forward to today.
- 37:31Policy application is welcome to in the attack space.
- 37:38Absolute and like you know what you're saying there was.
- 37:42It feels like you've been grappling with not the same old
- 37:46topic. So it's not quite, not to say,
- 37:48but you've been definitely dealing with topics that are
- 37:52front and centre now a long time.
- 37:55Yeah. I think we've, I mean, it's
- 37:56interesting that sometimes we have the same conversations that
- 37:59we've been having from the guys, right.
- 38:01And it all boils down to what data are you using?
- 38:05Do you understand where it's coming from?
- 38:07Do you understand the use cases around it and are you disclosing
- 38:10that, right. So that's a lot of very basic
- 38:15premises or principles that a lot of.
- 38:20Right. And what do you so with that in
- 38:22mind, what do you what would you say is something that's his can
- 38:26existence in your mind a problems challenge?
- 38:30You know his face that you feel just has moved too slowly.
- 38:35Just need more more attention. Yeah, so I think, you know, if I
- 38:40could boil oil the ocean down to one small, you know, come.
- 38:45I would say unexpected uses of dating in our space is the
- 38:50biggest threat and the biggest motivate for a lot of policy and
- 38:57lawmaking is coming down on our age.
- 39:01So no so collective consumers data you want to show them a
- 39:06relevant path, but there's all these other unexpected uses that
- 39:11I feel everyday consumers. They are also guess that data
- 39:17gets enforced for all these lot of potentially selling content
- 39:21and their advertising data to the government.
- 39:27And so, so that is converging and it's really interesting
- 39:31about it in the last probably less.
- 39:33Yeah. And and I guess it does say,
- 39:40does unexpected also cover callers and AI maybe like is
- 39:46that your, is that in real house unexpected or not?
- 39:49Well, it depends, I will say and in the clinic lawyer responses,
- 39:53it depends, right. It depends one part of what how
- 39:58you educate consumers them data uses is through your right.
- 40:03So your privacy, well it's very basic and and other ways that
- 40:07interact with your users to help them understand that you're
- 40:12using their data. Purchase data to serve
- 40:16additional relevant ads so that we're using the interests to
- 40:19serve them ads. If they're like auto intender,
- 40:22you kind of know that and shouldn't have auto ads
- 40:25interested to them as they're going through that purchase
- 40:28funnel. And if you're not disclosing
- 40:31that, that's when you know, you hear people speak in the policy
- 40:36world, like the creepy factor, right?
- 40:38My phone is listening to me or I just had this conversation with
- 40:43a friend at their house about a new product and I came home and
- 40:46now I'm getting social media for that for sure, right.
- 40:50So that seems creepy to folks and, and, and therefore perhaps
- 40:55unexpected, but I think there's a lot that the ad tech industry
- 40:58can do to to debunk that, doing a better job of explaining what
- 41:05data they're, you know, how people expel data, you know, how
- 41:09they, how, what, what data are we using and how we're using
- 41:14them so that they'll feel more comfortable and trust their,
- 41:18their, you know, their Internet in search all of what?
- 41:21Absolutely. That's a really good point.
- 41:24Well, you know, throughout my career, I've always felt that
- 41:26the say do gap of this stuff from a Pinterest perspective is
- 41:30it's quite wide in the sense that people who say they want
- 41:34privacy and then they act as if they don't want privacy or they
- 41:39use products in a way that is. Yeah, I think it's there's
- 41:43really a give and take, right. So I think first of all, a lot
- 41:46of the questions do do consumers want privacy?
- 41:50If you were asked, do I want privacy, you're never going to
- 41:54say no to that. But if you're asked, am I, are
- 41:57you willing to, to give up a little bit in your data for the
- 42:02benefit of a free and independent year?
- 42:06The answer would be a lot different.
- 42:08So, so part of it, it depends on the data question, but I also
- 42:12think there's tremendous benefits from data we have in
- 42:17the sides that can help humanity, that can help
- 42:20consumers, that can help brands, that can help publishers.
- 42:23And so being able to do that in a responsible way while not
- 42:28going to the end extreme of using data for things that you
- 42:31don't want me to lose data is where that is where the NAI and
- 42:36some of the other obviously folks that are trying to to
- 42:40distill from. OK, and that brings me to.
- 42:44I guess a. Really interesting question for
- 42:46the President. Of the nai.
- 42:48Which is AI and how AI intersects with data separates
- 42:54it kind of black boxes it a little bit and you know that
- 42:59sort of took on battle about the pre D factor as well.
- 43:03I see in today or sorry, I see implementations of AI LED data
- 43:10extrapolation usage that that make things more previous.
- 43:13Yes, far also don't come under the same privacy scrutinizations
- 43:19as the data itself. And so I'm interested to hear
- 43:22from you about what keeps you up at night to AI and all the time.
- 43:26Well, yeah, there's a lot going on.
- 43:28You certainly are. But I think if you step back and
- 43:30look at it from a macro level, it's from a policy.
- 43:33There's like a public policy perspective.
- 43:36There's a lot going on with AI. There's global profits to show
- 43:40in the AI space. So there is a a hesitancy to
- 43:45regularly you've seen that from the current U.S. government
- 43:49dictating constraints. They don't pass any regulation
- 43:52on AI. There's great there's arms races
- 43:55with their with other governments about how they're
- 43:58using and deferring the AI resources you have.
- 44:02And then one of the things that the US has always benefited
- 44:06from, I think is a sense of touch regulation rather than you
- 44:11check people. It's one of the reasons that we
- 44:13have all those news out when he's being found rule and
- 44:16developed in the United States. So I think then there's a
- 44:19there's a hope that we can preserve that.
- 44:22So you start with that premise and then you say, but what what
- 44:24about these unexpected or unknown losses of data?
- 44:28And I'll say that the same laws that apply to use of data for
- 44:33advertising or other uses that they decide and privacy may also
- 44:39apply to Jr. So that's the base, right.
- 44:41The FTC Act, which is the federal law that prevents unfair
- 44:46or deceptive trade practices or acts, it's insulators, applies
- 44:50to AI in the same way that it applies to all of the other
- 44:52technologies we have. So that that is very technology
- 44:56neutral and all the other privacy laws that have been
- 45:00passed in the States and international and, and sure all
- 45:06of the other privacy rules that other countries have passed.
- 45:08So I think though the the use, the understanding what data is
- 45:14being used and what it's used for and are you disclosing it
- 45:16properly or shielding it properly responsible use of
- 45:20data, those premises and principles will still apply for
- 45:24they are the tricky part is that it's much more as you mentioned,
- 45:29it's much more right. And so it's harder to understand
- 45:32the see that that besides all the other implications of
- 45:38publishers data striped in unfair ways.
- 45:42We've all kinds of other competition issues with that
- 45:44method. I think, you know, we are going
- 45:47to reach especially advertising when we're talking about for
- 45:53programmatic, for advertising facilities, you're really going
- 45:57to get to a point where you're going to have to make sure that
- 46:00it just about our user understands the data flows and
- 46:05also understands where the data is from, whether that partner is
- 46:10collecting that data filled with a safe form, right or a
- 46:13responsible you want. And so that is where I think the
- 46:17rubber meets the road. Is there going to have to be
- 46:20privacy scoring or data scoring or are we going to, you know,
- 46:25descend the resolution along with this data was collected and
- 46:28properly. So so it's OK to use this, but
- 46:32it's not OK to use this. So I think we're just starting
- 46:35that journey in AI governance, but I but we should start with
- 46:39the kindness of basics. He's teachers, Paula's voice
- 46:44still apply to dating generally, regardless of if you're using
- 46:47it. So that's a really interesting
- 46:51point in terms of teaching your agents about to data.
- 46:55I think the concept of that is a really interesting one for the
- 46:59next year last couple. Of.
- 47:02To set that standard not at when she's the human using it, but is
- 47:06the agent using it properly. But that's fascinating.
- 47:09Sounds like you got a lot on your.
- 47:11Plate there's a lot on our plate yeah I think you know looking
- 47:14ahead to the future you know one of the things in the at least in
- 47:17public policy things they going well now looking at is the
- 47:21divergent in the way different governments or administrations
- 47:27look at these things right so the prior administration had a
- 47:31much more regulatory. Have you like felt this to
- 47:35follow our FTC and share and now it's now planner a new
- 47:39administration's backed off. So we'll see where that pendulum
- 47:42focuses in the next few years. But there are definitely
- 47:45different talks on tonight just for that.
- 47:47So I think that's a variable factor as well.
- 47:53Wow, OK, this has been a really illuminating and fascinating
- 47:56things with the stuffing. Thank you so much.
- 47:58Yeah, thank you. It was nice to talk.
- 48:00To fantastic. Thanks.
- 48:03All right. So we're here at Ramp Up 2026 in
- 48:05San Fran and I'm with Scott AD tech therapist and we are
- 48:10talking a little bit about publishers and data
- 48:13collaboration and what that sort of entails over the next couple
- 48:17years. Scott, you've got some really
- 48:18interesting takes on this topic and what content businesses are
- 48:24finding perilous, what they need to do more of.
- 48:26And so talk me through your your thesis, I guess.
- 48:29Absolutely. Well, let's go on a little bit
- 48:30of a journey here. So we're going to do some go
- 48:32through some examples and you'll see where we wind up at the end.
- 48:35The first thing was I came out a couple weeks ago after leaving
- 48:39IBALM and hearing some confidential stories from
- 48:42sources and it drew me to the conclusion that using LLMS is
- 48:50funding organized crime and state funded terrorism.
- 48:55OK, this is a big call. Let's let's dive into this a
- 48:57little. Bit and I, I will say like
- 48:59there's a bit of a leap and it's not all of it.
- 49:01And I'm not saying don't use AI. I'll tell you why I'm bringing
- 49:04this up. No, for sure.
- 49:05But the way that it happens is that in order to scrape content
- 49:09and sell it in a marketplace, you need large scraping
- 49:13networks, which are essentially headless browsers using
- 49:17residential IPS and proxy locations to visit websites,
- 49:22scrape the content, and then send it back.
- 49:24In order to build an army of those, it would be very
- 49:27expensive and very difficult, but these armies actually exist
- 49:30already. Sounds familiar?
- 49:32Sounds familiar? Ad fraud.
- 49:34So a lot of the companies who commit ad fraud have those exact
- 49:37same pieces and are now using them instead of spreading
- 49:41malware to go and scrape publisher websites and to build
- 49:45large corpuses of content that can be sold to LLMS for training
- 49:49purposes. There's two inherent problems.
- 49:51One, that content was not paid to the publisher in any way and
- 49:56two, some of those companies that have those scrapers,
- 49:59especially the ones that come from LOD fraud organizations are
- 50:03the money goes back to organized crime and state funded
- 50:06terrorism. So that's.
- 50:09Bad. And people just need to
- 50:11understand that stealing content from publishers is not a
- 50:15victimless crime. It's like the handbag market,
- 50:19right? There is a illicit market for
- 50:21handbags and it's not a victimless crime.
- 50:23Sure, Fendi doesn't get as many sales, but also what happens
- 50:26from the money that those organizations make and where
- 50:29does that go? And I can promise you that is
- 50:31used for things that are way worse than counterfeiting
- 50:34handbags. Yeah, absolutely.
- 50:36Look, it's a really, it's a drastic example.
- 50:38But the fact that I think we all understand that whenever new
- 50:42technologies emerge, good actors emerge, but bad actors tend to
- 50:46emerge pretty quickly as well. And so, you know, what's, what's
- 50:50your key call out here for, for content businesses and the way
- 50:55that the way that they've got to sort of change their mindset.
- 51:00A little bit over the next year. Yeah, the, I sort of break it
- 51:05into the what's in your tool set for a publisher content creator,
- 51:09because the ultimate goal is to compel the LLMS to license
- 51:13content legally through marketplaces so a publisher can
- 51:17get money directly and build that into their business plan,
- 51:20right. Currently today, some of the
- 51:21small LLMS are licensing content, but the big ones, open
- 51:27AI and Google and Perplexity will throw them in.
- 51:30There are not. And why would they want to do it
- 51:37when they are taking something for free and selling it for a
- 51:39very large markup? It's a great question, right?
- 51:44Like we have to compel them. So that's the language I use,
- 51:47compelling the LMS to the place. And it's not one single thing
- 51:51that will do it. But there's a series of sticks
- 51:54that exist in our tool kit. And I'm going to talk about
- 51:56carrots just after this, right? Because it can't be only sticks.
- 51:59Carrot and the stick. Exactly right.
- 52:01OK, so the carrots that are in your toolbox are private
- 52:05litigation, right? Government regulation.
- 52:08Yeah, competitive action model, dehydration and public
- 52:15sentiment. So that last one where we talked
- 52:17about organized funding, organized crime, that sort of
- 52:21goes into the public sentiment. One.
- 52:23Another one for that is if we defund journalism by removing
- 52:27traffic from The New York Times and The Wall Street Journal and
- 52:29whatever your favorite newspaper is, they can't pay for
- 52:32journalists. They can't give them safety
- 52:34equipment, and that becomes a problem.
- 52:36And it's sort of degrades society.
- 52:38And so society should care about that.
- 52:40That's one thing that's weighing in the minds of Google and open
- 52:42AI of like, how does the public see feel us taking this content?
- 52:48Of course. And and you mentioned, sorry,
- 52:50you mentioned 1 slightly more technical one called model
- 52:53dehydration. Could you elaborate on that?
- 52:56Does layman's term that one? Model dehydration is when the
- 52:59content literally is not available to these models.
- 53:03And you can think of it becoming dehydrated, right?
- 53:06If the breaking news is not in a model, it becomes like Chachi PT
- 53:12was three years ago where it only was trained up to 2022.
- 53:15Very frustrating, right? And so if you start removing
- 53:18large chunks of information from these models, yeah, they become
- 53:23dehydrated and they become relatively useless actually.
- 53:26Like think about how much real time information you need in
- 53:29one, not just to do coding, but if you want it to be an agentic
- 53:33browser, if you wanted to read the news, if you wanted to
- 53:35update you and do these things, they need real time information.
- 53:39So model dehydration is publishers can employ that by
- 53:43moving from an allow list of, I'm sorry, moving from a block
- 53:48list of, blocking all the scrapers that come to them to an
- 53:53allow list and only allowing certain robots onto their site.
- 53:58I see. Now you're going to block some
- 54:00good ones, and then you're going to have a conversation with.
- 54:02Them I was going to say you'll end up blocking by default.
- 54:04Yeah, do not block your ads dot TXT robot from the IAB because
- 54:08that will delist you from all the marketplaces as soon as you
- 54:11do that. So allow that one and several
- 54:13others. You still have to grapple with
- 54:15the question of like, should we allow the Google crawler or not?
- 54:18And there's a lot of popular opinion out there that says
- 54:20actually we should be blocking the real Google search tool
- 54:25because they are taking that data and it is being fed into
- 54:28Gemini and it's sort of coming around the back and it's still
- 54:31in the Google index and they are training on it.
- 54:34And like, if you're getting very little Google traffic, why
- 54:38should you be feeding your information to Google?
- 54:39I feel like a lot of ad tech crawlers are going to see their
- 54:43their inbounds traffic dropping. Yes, yeah.
- 54:48But I think from a consumer's perspective as well, I think as
- 54:50AILLMS move from being a novelty to being a utility and something
- 54:55that's ingrained in your workflow, this bit of model
- 54:58dehydration is really important because I need fresh information
- 55:02to do the latest work, to know that my sources are real, that
- 55:05they're fresh. And you know, two years ago and
- 55:08maybe I didn't need or didn't think about, I didn't think to
- 55:10ask if my information was fresh whereas well and I so.
- 55:13I think that might be something that.
- 55:16You know the big content businesses out there that they
- 55:18can, they can hold back. It's not to ransom, but they can
- 55:20hold that as a precious. Especially news publishers,
- 55:23Yeah, the long tail SEO ones, like that's not a big deal for
- 55:26you, but the news publishers that is, and that comes into
- 55:29competitive action as well. Like what's stopping Google
- 55:33today from moving into a licensing model?
- 55:37Very little. I have a theory that they could
- 55:39use Google Search Console and Adsense to do the counting of
- 55:44queries and then pay publishers out directly.
- 55:47Those two mechanisms exist for them.
- 55:49They are what they are for. And I think making that code
- 55:51update isn't that hard for a company like Google.
- 55:55So they could do it. Now let's assume that they
- 55:57actually do it. As soon as Google starts paying
- 56:01publishers for licensing content, Google publishers will
- 56:05use that blocking mechanism to block an open A an open AI and
- 56:10immediately model dehydration starts on open AI.
- 56:14The snowball rolls. How quickly do you think Open AI
- 56:17will move into a licensing program if that happens to them
- 56:21at Internet scale? Not just onesie twosies of a
- 56:24Reddit and New York Times and the and the little deals.
- 56:28And I think that's where competitive action then pulls it
- 56:30right. And then you obviously have
- 56:31private action of everybody suing them and government
- 56:33regulation whenever the DOJ shows up.
- 56:35The UK is has some interesting lawsuits going on and unlike
- 56:39GDPR and privacy law, content rights are much more of a global
- 56:45problem for LLMS. Sure, because you can't say like oh the the
- 56:50European version of our LLM doesn't contain this content but
- 56:53our US1 does. That melting pot is like, well
- 56:57we're well past soup and Stew and down.
- 57:00We're down into like fine soup at this point.
- 57:02So momentum creates momentum in that way, yeah.
- 57:05And so to to. I guess to round out the carrot
- 57:09and the stick. Now we haven't.
- 57:11Got to the carrots. So talk to me about the carrots.
- 57:13So licensing content through marketplaces has benefits.
- 57:17One, there's solving a bunch of your legal problems, your public
- 57:21perception problems, and then also the data is structured for
- 57:27LLMS to consume it. Imagine going to a website and
- 57:30how long it takes you to read it and find it in all the boxes and
- 57:33the places and the recipes are here and there, and some of it's
- 57:35hidden in JavaScript. When you get it through the
- 57:38marketplace, it's all written in a markdown format that's very
- 57:42easy for the LLMS to ingest and consume.
- 57:45And then you're also moving from a let's scrape the Internet to
- 57:48find out what happened yesterday to a let's query the sources
- 57:52where we know that there are updates being made like
- 57:55newspapers and use more of an API driven system to get that
- 58:00information back in that structured format.
- 58:03And the cost is going to reduce significant.
- 58:06Got it. And then you pay publishers and
- 58:08we continue to fund journalism. There you go.
- 58:10So there are some carrots that are out there, but I don't think
- 58:13that the carrots are enough to compel them, which is why I
- 58:16start with your stick strategy and remind them that carrots
- 58:19will be there when they show up. Fantastic.
- 58:21I love that. I love the the walkthrough and I
- 58:23love the carrots and the sticks analogy.
- 58:26Thank you so much, Scott. It was a pleasure.
- 58:27To talk to. You you're very welcome and
- 58:29credit to Danny at Ozone for the Carrot edition.
- 58:32So thank you very much. Thank you.
- 58:34Chris Pleasure. OK.
- 58:38So we're at ramp up in San Francisco 2026 and I'm very
- 58:42lucky to be speaking to Shaylee Singh, the Chief Operating
- 58:45Officer and EVP of Product for the IB Tech Lab.
- 58:49Is that correct? Yes.
- 58:50Fantastic. OK, that obviously that sounds
- 58:52like an absolutely huge role, a very important one and one
- 58:55that's probably keeping you up at night with a lot of different
- 58:59topics. So let's, I'd love to hear your
- 59:01take on, you know, where you've come from to get to that role
- 59:05and you know what, what is actually your sort of number one
- 59:10thing that you're, you know, looking at right now.
- 59:14Yeah. So I've been here at IB Tech Lab
- 59:16for 10 years, since pretty much the founding states.
- 59:19This is my 11th year running at IB Tech Lab.
- 59:22Started working here as a director of product, looking
- 59:25after mobile stuff and AD products.
- 59:29And then kind of of course now I look at everything and then I do
- 59:34pick up like every year like one or two things that I really want
- 59:36to focus on. So it keeps me stimulated enough
- 59:40mentally and gives me some hands on experience so.
- 59:43I'm sure you're not short of things to to think about at the
- 59:46moment. So what What did you pick?
- 59:48What was the thing that you have chosen as I guess your flagship
- 59:53or project for the next? Year.
- 59:55So this year, there's a couple of things like that I'm working
- 59:58on to kind of carry over from last year.
- 1:00:00One is the impact of AI on the industry as such, of course,
- 1:00:03what are different things in that?
- 1:00:04And the other is the project that I was just telling you
- 1:00:06about, like we're trying to move the stack from the browser and
- 1:00:09feed from the browser restrictions to take it to the
- 1:00:12CDN and the server side product where you can manage all of your
- 1:00:15advertising. So those are the 2 main things.
- 1:00:17Within the AI, there are two areas that we've been looking at
- 1:00:20that I've been focusing on. One is the impact of AI on
- 1:00:24publisher revenues, the impact of AI search and drops.
- 1:00:27The you know people have seen up to 3040% reduction in traffic of
- 1:00:31course from search and that means has an impact on everyone
- 1:00:35like publisher lose revenue. It also has an impact on
- 1:00:38advertising industry like the brands cannot find these new
- 1:00:40incremental users that they typically find on the open web
- 1:00:44and then drops down QPS for everybody else in the middle.
- 1:00:48So we're trying to, we have a project on that called com
- 1:00:51content monetization protocols. So you're trying to, because the
- 1:00:54value exchange moves up from the destination Click to informing
- 1:00:58the LLMS. Yeah. So we're trying to put together
- 1:01:00a set of standards and protocols where it's easy for LLMS and
- 1:01:05publishers to transact on the content and publishers to get
- 1:01:08paid compensated for that value exchange that they've lost.
- 1:01:12Right, so it's like a, it's a framework to enable publishers
- 1:01:16to understand how to set up their stack.
- 1:01:19Yeah, it's in order to best. To get out three-step like 1 is
- 1:01:22understanding your bot management like which bot brings
- 1:01:25you value, which doesn't and how to manage that.
- 1:01:27Second is to set up licensing like how do you actually
- 1:01:30understand the license for each crawler that comes to your site?
- 1:01:33And then third is of course getting paid by how to you know
- 1:01:36what the transaction looks like, How do you invoice them?
- 1:01:38How do you build them of? Course and and does that, you
- 1:01:41know, out of interest Google Keep flip flopping on their sort
- 1:01:45of cookie deprecation and I guess that sort of plays into
- 1:01:48how you think about the standards you set behind
- 1:01:51publisher monetization as well. Is that one of the factors?
- 1:01:54It used to be the case, but after all the back and forth,
- 1:01:57like, I don't think we'd think about it anymore.
- 1:02:00No, yeah. I mean, you know what?
- 1:02:01Everybody so we do think about signals in general because
- 1:02:05because of, you know, privacy regulations that are restricting
- 1:02:09some of the signals that come through because of the privacy
- 1:02:12restrictions placed by platforms and browsers like, you know,
- 1:02:15there's no third party cocaine about 42% of the web in US, for
- 1:02:19example. In some regions it's even higher
- 1:02:22and then there's ad blocking and then there is ITP restrictions
- 1:02:26that somebody has placed. And so we do think about that.
- 1:02:31So how do you enrich the signals?
- 1:02:32What what are the other signals that can be passed that will
- 1:02:36enable a better quality decision on on on bidding?
- 1:02:39OK. So surely I want to talk a
- 1:02:41little bit about publisher monetization and publisher
- 1:02:46futures. You've been busy with the
- 1:02:49projects inside the IB Tech Lab, you know, regarding publisher
- 1:02:54CDNS and how signals are getting propagated from the CDN in order
- 1:03:00to better have the publisher take control.
- 1:03:01Can you talk us through what you're doing with the IB there
- 1:03:04and why that's so important for publishers?
- 1:03:06Yeah. So what we're doing, the
- 1:03:07project's called Trusted Server. So what we're, the original
- 1:03:11vision of the project was to to solve the problem that
- 1:03:16publishers face due to the restrictions of the browsers for
- 1:03:20privacy reasons. So no third party cookies and
- 1:03:23then ITP restrictions in Safari, which kind of stop and and
- 1:03:29restrict the traditional methods of monetization like you know,
- 1:03:32that publishers use on the web. So they will kind of block the
- 1:03:36trackers, block the calls that are being made from the browser
- 1:03:40and, and anybody who goes can look at that list.
- 1:03:44So that was the original idea to move all of this work to the
- 1:03:47server side where you can actually publishers can have
- 1:03:49full control, right? And they can give controls to
- 1:03:51their partners to do the things that they want to do.
- 1:03:54But then if you look at the other thing that we were talking
- 1:03:56about, the impact of AI and the, the traffic on the web today is,
- 1:04:04is more crawlers and more bots than humans.
- 1:04:07So in 2024, the 1st. Wow.
- 1:04:09The web became 51% crawler traffic on non human traffic and
- 1:04:14a lot of that jump has been because of AI, because all these
- 1:04:17agents companies like. That's a terrifying step.
- 1:04:19Yeah. So they need like they need all
- 1:04:21the machines need all of this data for sure to process and to
- 1:04:24train and to infer results for users.
- 1:04:27So basically the actions move to CDN, all of that happens to CDN
- 1:04:29that's really recognize what the user agent is, whether it's a
- 1:04:32human or not. Got it.
- 1:04:33If it's a human, you want to give them a nice looking HTML
- 1:04:36page and with all the JavaScript and everything on that page.
- 1:04:39But if it's a crawler, you want to just give it a markdown file,
- 1:04:42just give it the content you don't want to, You can optimize
- 1:04:45all of the things there, right? And we, we also found like, you
- 1:04:48know, ad blockers like if you, if you're on the server side, if
- 1:04:51you're already filling all the ads on the CDN and then sending
- 1:04:54to the browser, then you can bypass the ad blockers, which
- 1:04:58just means like 15 to 20% of the traffic.
- 1:05:02Which is quite meaningful like that turns into revenue that's.
- 1:05:05Straight dollars like yeah. And then straight out of eye on
- 1:05:08your investment and moving things to the CDN.
- 1:05:10Yeah. And CDNS like traditionally
- 1:05:11being looked at a place to distribute Edge files on the
- 1:05:15edge, faster delivery of files, but they're far more capable
- 1:05:18today. Like they've got the framework
- 1:05:20called Wazam and Wazi, which can actually execute server like
- 1:05:22functions. OK, So what that does is like in
- 1:05:26a in a browser, technically everything is a single thread.
- 1:05:29So if a Javascript's written before, first it gets done, then
- 1:05:31the next one, then the next one, right?
- 1:05:34And the server, it's all you can do parallel processing.
- 1:05:36So I can go out to like I can go out to pre bid, I can go out to
- 1:05:40Google open bidding, I can go out to Amazon publisher service
- 1:05:43at the same time get all the process everything back, package
- 1:05:46it and push it back to the browser as a single page, which
- 1:05:49is also. Like better for the page better,
- 1:05:52right? We haven't measured all of that
- 1:05:54yet. We're still testing all of that,
- 1:05:55but you've already seen like initial improvements and all of
- 1:05:58those things. It sounds, I mean, it sounds
- 1:06:00like not a no brainer in that way, but it sounds like the
- 1:06:04closest thing to a no brainer for a publisher whose traffic is
- 1:06:07declining that they would take more control and monetize more
- 1:06:12securely of the customers that they do have.
- 1:06:16You know, if they've, if they've, if they've got less
- 1:06:18visitors, then it's really important to treat every one of
- 1:06:21those visitors like you know them and, and securely and, and.
- 1:06:27It's not happening today, but once LLMS agreed to compensate
- 1:06:30publishers for the content, then that transaction will also take
- 1:06:33place and recorded at the CDN. Yes, that's.
- 1:06:36Going to happen. So CDN is something that you
- 1:06:38know, publishers and at tech really needs to focus on and
- 1:06:42make the best use of. It's got the technology, it's
- 1:06:46got like there's many business reasons to be there and then
- 1:06:52technically it's proficient to execute what you need to do.
- 1:06:55Wow. Yeah.
- 1:06:56I mean, that's definitely something that I think you know,
- 1:07:00more publishers need to hear and get levelled up on because it
- 1:07:05sounds like a really robust solution for the long term as
- 1:07:09well for for their health. Well, we've talked a little bit
- 1:07:12about bots, but I want to, I want to talk about one topic,
- 1:07:15which is a gentic advertising or a genetic ad tech, programmatic,
- 1:07:20whatever you want to call it. This must keep you up at night a
- 1:07:23little bit because you've obviously spent so much time
- 1:07:25thinking about data and AD tech. But now you've got a world where
- 1:07:29potentially humans are not doing the thing, agents are doing the
- 1:07:32thing. And so when I say the thing I
- 1:07:35mean. You know, a media buy or.
- 1:07:36A placing a bid or an optimization or whatever.
- 1:07:39And So what? What is going on in your mind
- 1:07:43when it comes to AI agents and how you need to change how you
- 1:07:49adapt your frameworks? They must be things that maybe a
- 1:07:53still grey area to. Not that grey anymore.
- 1:07:56Like we started working on it sometime last year.
- 1:07:59OK, so we've been noodling with this like understanding MCP
- 1:08:02server A to a protocols and how that applies to advertising.
- 1:08:06We actually launched our full agent take road map.
- 1:08:10So we've taken what you've done is it's got kind of three
- 1:08:13layers. So what is happening is like
- 1:08:15you're going from automation to autonomous.
- 1:08:18So few things are really going to be critical for those
- 1:08:20autonomous operations that happen.
- 1:08:22One, the first is like people have to change their role from
- 1:08:26being operators on a dashboard and entering line items to
- 1:08:29architects and governance like giving the guidance, giving,
- 1:08:33giving the architecture guidance to the to the agents.
- 1:08:37And to do that, to execute on that agents would need, you
- 1:08:41would need to trust the agents that they don't hallucinate.
- 1:08:45This is the right agent that my agents talking to and that it's
- 1:08:48not some kind of spoof agent come up from somewhere.
- 1:08:51And then you would need interoperability that if
- 1:08:54somebody's got different types of agents, they have different
- 1:08:56systems at the back end of the agent, they can all talk to each
- 1:08:59other. So the work that we do have been
- 1:09:02doing for 10 years, 11 years is actually becoming even more
- 1:09:05important to have the standards in place.
- 1:09:08And when we looked at it, we realized that we have all the
- 1:09:11standards that you need to leverage.
- 1:09:13We may have to build some more new things to support the
- 1:09:16business processor. But for now, we think like the
- 1:09:18all the standards that we build, they're actually more
- 1:09:22meaningful, more critical, more important than ever before.
- 1:09:25Absolutely. In this world.
- 1:09:27Yeah, especially like my mind is racing as you're talking about
- 1:09:30this, because the implications of not having standards in a
- 1:09:34world where agents are doing things, it doesn't sound great.
- 1:09:38It sounds like things can get out of control at a much faster
- 1:09:42pace than you. Get hallucination.
- 1:09:44You can have agents like so that's the thing with autonomous
- 1:09:48like the amplification is much larger.
- 1:09:51If it's going in the wrong direction for whatever reason,
- 1:09:54it amplifies that much faster as it amplifies the accuracy and
- 1:09:58efficiency as much better. Absolutely.
- 1:10:00So that's what we did. So we built like a three layer
- 1:10:03architecture like we're calling it AMP, AMP for the AA.
- 1:10:07So agent take advertising, management protocols, OK, which
- 1:10:10is a collection of three kind of major things.
- 1:10:14So at the bottom is like the foundational of the execution
- 1:10:17layer, which we're calling ARDA for agenting real time
- 1:10:20framework, which allows you to take the outputs of the agents,
- 1:10:23the decisions agents have made and plug them into the supply
- 1:10:26chain. OK, so the, the plumbing and
- 1:10:28execution is not going to change.
- 1:10:29You still have RTB, you still be doing bidding, but how to make
- 1:10:32it work faster, make it work smarter?
- 1:10:35So the ARTF is that layer, which kind of plugs into the supply
- 1:10:38chain, lives in the host platform servers.
- 1:10:41So they can act really fast with very low latency.
- 1:10:44Like with the container technology, you get like 80%
- 1:10:48improvement in latency. So you have this luxury of time
- 1:10:51to make more quality decisions. The second layer is a management
- 1:10:54protocol, which is like the decision making layer, which is
- 1:10:57where you do your media allocation, which is where the
- 1:10:59seller agent can start doing their optimization and being
- 1:11:02able to provide a better rate or, or or understand the
- 1:11:06inventory better or exchange inventory better.
- 1:11:09So that is built on top of like our standards.
- 1:11:11We already have like that open direct that that ex sales.
- 1:11:15We have the deals objects, so we can be able to do deals through
- 1:11:18that. And we have the RTB.
- 1:11:19So if you have to set up a campaign on ADSP, you can just
- 1:11:21push it with the RTB objects and all of the combines with the
- 1:11:25adcom, which are the underlying objects and everything.
- 1:11:27So the agents can use that as the schema, standard schema and
- 1:11:31then talk to each other and have much higher accuracy than
- 1:11:35otherwise. They have a lot of the people
- 1:11:37like started questioning this like last year that LLMS
- 1:11:41understand English. So I don't need standard, I can
- 1:11:43just tell whatever it knows. But if you go and read
- 1:11:45anthropics documentation on MC protocol, the first paragraph
- 1:11:48basically says you need a standard schema.
- 1:11:50You need to provide the LLM the right context, and nothing can
- 1:11:53provide the right context better than the standard schema that
- 1:11:56the industry is used for like 15 years.
- 1:12:00So that's the management layer. And then the third layer is the
- 1:12:03trust where we just launched yesterday at the agent registry.
- 1:12:06OK, So people can come and register the agent.
- 1:12:08We do the verification of the agent.
- 1:12:10It can be an MCP server, it can be an A2 agent.
- 1:12:14It may need authorization. They can declare that, or they
- 1:12:16can provide a token for general discovery, but then we will hit
- 1:12:20their agent, We'll understand the capabilities.
- 1:12:22We'll list all of them. We also match it to the data
- 1:12:26that we already know about the industry, like the TCFID or your
- 1:12:29GPPMSPID. So then you have a verification
- 1:12:32like this is a company that's already been certified to
- 1:12:36operate under GDPR, so I can actually work with this agent.
- 1:12:39Of course, they'll be commercial paper before you open up your
- 1:12:42system to anyone, but at least you can do a search.
- 1:12:45You can list, see like, OK, I already work with these
- 1:12:48companies and so that's the trust layer.
- 1:12:51So you're creating all these three layers, the trust, the
- 1:12:53interoperability and the execution.
- 1:12:56Yeah, it sounds, it sounds really robust.
- 1:12:58And I can imagine in a weird way in an agentic world once
- 1:13:03deployed or these frameworks deployed or once the initiatives
- 1:13:06are are embedded, they'll get followed.
- 1:13:10And that's why they're more important in an air world,
- 1:13:12because they'll get followed more systematically.
- 1:13:15More systematically, the agents will incorporate them and they
- 1:13:19will actually make the agents and the LMS even smarter.
- 1:13:22Yeah. Because now they'll Start
- 1:13:23learning how these how people operate on these standards.
- 1:13:28And then they they get much more smarter that they get much more
- 1:13:31accurate over time. Yeah, that's how we reach like
- 1:13:33that. I think we won't reach
- 1:13:35autonomous stage like right away.
- 1:13:36First to be like pilots, control pilots.
- 1:13:39Then it'll be like human in the loop kind of operation for some
- 1:13:41time and then we will reach like a full lead and you're
- 1:13:44comfortable like people start OK, I can let this part of my
- 1:13:47agent take framework run autonomously because it's never
- 1:13:51made a mistake in the last one year.
- 1:13:52Of course. Yeah.
- 1:13:53I mean, look, I mean that that's an incredibly exciting road map
- 1:13:57and it's, it's really cool to hear it from you specifically
- 1:14:02because you've obviously come from, you know, a lot of tenured
- 1:14:06background in this space, but the path you're charting
- 1:14:09towards, I guess autonomous is, is really exciting to hear.
- 1:14:13So thank you so much Hayley, for, for talking with us and
- 1:14:17we'll, we'll talk to you next time.
- 1:14:18Yeah. My pleasure, thanks for having
- 1:14:20me. Thanks so much.