Cookies are out, context is in. People Inc.’s Jonathan Roberts joins The Big Impression to talk about how America’s biggest publisher is using AI to reinvent contextual advertising with real-time intent. From Game of Thrones maps to the open web, Roberts believes content is king in the AI economy.
Cookies are out, context is in. People Inc.’s Jonathan Roberts joins The Big Impression to talk about how America’s biggest publisher is using AI to reinvent contextual advertising with real-time intent.
From Game of Thrones maps to the open web, Roberts believes content is king in the AI economy.
Episode Transcript
Please note, this transcript may contain minor inconsistencies compared to the episode audio.
Damian Fowler (00:00):
I'm Damian Fowler, and welcome to this edition of The Big Impression. Today we're looking at how publishers are using AI to reinvent contextual advertising and why it's becoming an important and powerful alternative to identity-based targeting. My guest is Jonathan Roberts, chief Innovation Officer at People Inc. America's largest publisher, formerly known as Meredith. He's leading the charge with decipher an AI platform that helps advertisers reach audiences based on real time intent across all of People Inc. Site and the Open Web. We're going to break down how it works, what it means for advertisers in a privacy first world and why Jonathan's side hustle. Creating maps for Game of Thrones has something for teachers about building smarter ad tech. So let's get into it. One note, this episode was recorded before the company changed its name. After the Meredith merger, you had some challenges getting the business going again. What made you realize that sort of rethinking targeting with decipher could be the way to go?
Jonathan Roberts (01:17):
We had a really strong belief and always have had a strong belief in the power of great content and also great content that helps people do things. Notably and Meredith are both in the olden times, you would call them service journalism. They help people do things, they inspire people. It's not news, it's not sports. If you go to Better Homes and Gardens to understand how to refresh your living room for spring, you're going to go into purchase a lot of stuff for your living room. If you're planting seeds for a great garden, you're also going to buy garden furniture. If you're going to health.com, you're there because you're managing a condition. If you're going to all recipes, you're shopping for dinner. These are all places where the publisher and the content is a critical path on the purchase to doing something like an economically valuable something. And so putting these two businesses together to build the largest publisher in the US and one of the largest in the world was a real privilege. All combinations are hard. When we acquired Meredith, it is a big, big business. We became the largest print publisher overnight.
(02:23):
What we see now, because we've been growing strongly for many, many quarters, and that growth is continuing, we're public. You can see our numbers, the performance is there, the premium is there, and you can always sell anything once. The trick is will people renew when they come back? And now we're in a world where our advertising revenue, which is the majority of our digital revenue, is stable and growing, deeply reliable and just really large. And we underpin that with decipher. Decipher simply is a belief that what you're reading right now tells a lot more about who you are and what you are going to do than a cookie signal, which is two days late and not relevant. What you did yesterday is less relevant to what you need to do than what you're doing right now. And so using content as a real time predictive signal is very, very performant. It's a hundred percent addressable, right? Everyone's reading content when we target to, they're on our content and we guaranteed it would outperform cookies, and we run a huge amount of ad revenue and we've never had to pay it in a guarantee.
Damian Fowler (03:34):
It's interesting that you're talking about contextual, but you're talking about contextual in real time, which seems to be the difference. I mean, because some people hear contextually, they go, oh, well, that's what you used to do, place an ad next to a piece of content in the garden supplement or the lifestyle supplement, but this is different.
Jonathan Roberts (03:53):
Yes. Yeah. I mean, ensemble say it's 2001 called and once it's at Targeting strategy back, but all things are new again, and I think they're newly fresh and newly relevant, newly accurate because it can do things now that we were never able to do before. So one of the huge strengths of Meredith as a platform is because we own People magazine, we dominate entertainment, we have better homes and gardens and spruce, we really cover home. We have all recipes. We literally have all the recipes plus cereal, seeds plus food and wine. So we cover food. We also do tech, travel, finance and health, and you could run those as a hazard brands, and they're all great in their own, but there's no network effect. What we discovered was because I know we have a pet site and we also have real simple, and we know that if you are getting a puppy or you have an aging dog, which we know from the pet site, we know you massively over index for interest in cleaning products and cleaning ideas on real simple, right?
Damian Fowler (04:55):
Yeah.
Jonathan Roberts (04:55):
This doesn't seem like a shocking conclusion to have, but the fact that we have both tells us both, which also means that if you take a health site where we're helping people with their chronic conditions, we can see all the signals of exactly what help you need with your diet. Huge overlaps. So we have all the recipe content and we know exactly how that cross correlates with chronic conditions. We also know how those health conditions correlate into skincare because we have Brody, which deals with makeup and beauty, but also all the skincare conditions and finance, right? Health is a financial situation as much as it is a health situation, particularly in the us. And so by tying these together, because most of these situations are whole lifestyle questions, we can understand that if you're thinking about planning a cruise in the Mediterranean, you're a good target for Vanguard to market mutual funds to. Whereas if we didn't have both investipedia and travel leisure, we couldn't do that. And so there's nothing on that cruise page, on the page in the words that allows you to do keyword targeting for mutual funds.
(05:55):
But we're using the fact that we know that cruise is a predictor of a mutual fund purchase so that we can actually market to anyone in market per cruise. We know they've got disposable income, they're likely low risk, long-term buy andhold investors with value investing needs. And we know that because we have these assets now, we have about 1500 different topics that we track across all of DDM across 1.5 million articles, tens of millions of visits a day, billions a year. If you just look at the possible correlations between any of those taxonomies that's over a million, or if we go a level deeper, over a hundred million connected data points, you can score. We've scored all of them with billions of visits, and so we have that full map of all consumers.
Damian Fowler (06:42):
I wanted to ask you, of course, and you always get this question I'm sure, but you have a pretty unusual background for ad tech theoretical physics as you mentioned, and researcher at CERN and Mapmaker as well for Game of Thrones, but this isn't standard publisher experience, but how did all that scientific background play into the way you approached building this innovation?
Jonathan Roberts (07:03):
Yeah, I think when I first joined the company, which was a long time ago now, and one of the original bits of this company was about.com, one of the internet oh 0.1 OG sites, and there was daily data on human interest going back to January 1st, 2000 across over a thousand different topics. And in that case, tens of millions of articles. And the team said, is this useful? Is there anything here that's interesting? I was like, oh my god, you don't know what you've got because if you treat as a physicist coming in, I looked at this and was like, this is a, it's like a telescope recording all of human interest. Each piece of content is like a single pixel of your telescope. And so if somebody comes and visit, you're like, oh, I'm recording the interest of this person in this topic, and you've got this incredibly fine grained understanding of the world because you've got all these people coming to us telling us what they want every day.
(08:05):
If I'm a classic news publisher, I look at my data and I find out what headlines I broke, I look at my data and I learn more about my own editorial strategy than I do about the world. We do not as much tell the world what to think about. The world tells us what they care about. And so that if you treat that as just a pure experimental framework where this incredible lens into an understanding of the world, lots of things are very stable. Many questions that people ask, they always ask, but you understand why do they ask them today? What's causing the to what are the correlations between what they are understanding around our finance business through the financial crash, our health business, I ran directly through COVID. So you see this kind of real time change of the world reacting to big shocks and it allows you to predict what comes next, right? Data's lovely, but unless you can do something with it, it's useless.
Damian Fowler (08:59):
It's interesting to hear you talk about that consistency, the sort of predictability in some ways of, I guess intense signals or should we just say human behavior, but now we've got AI further, deeper into the mix.
Jonathan Roberts (09:13):
So we were the first US publisher to do a deal with open ai, and that comes in three parts. They paid for training on our content. They also agreed within the contract to source and cite our content when it was used. And the third part, the particularly interesting part, is co-development of new things. So we've been involved with them as they've been building out their search product. They've been involved with us as we've been evolving decipher, one of the pieces of decipher is saying, can I understand which content is related to which other content? And in old fashioned pre AI days when it was just machine learning and natural language processing, you would just look at words and word occurrence and important words, and you'd correlate them that way. With ai, you go from the word to the concept to the reasoning behind it to a latent understanding of these kind of deeper, deeper connections.
(10:09):
And so when we changed over literally like, is this content related to that content? Is this article similar in what it's treating to that article? If they didn't use the same words but they were talking about the same topic, the previous system would've missed it. This system gets deeper. It's like, oh, this is the same concept. This is the same user need. These are the same intentions. And so when we overhauled this kind of multimillion point to point connection calculation, we drastically changed about 30% of those connections and significantly improved them, gives a much reacher, much deeper understanding of our content. What we've also done is said, and this is a year thing that we launched it at the beginning of the year, we have decipher, which runs on site. We launched Decipher Plus Inventively named right? I like it. We debated Max or Max Plus, but we went with Plus.
(10:59):
And what this says is we understand the user intent on our sites. We know when somebody's reading content, we have a very strong predictor model of what that person's going to need to do next. And we said, well, we're not the only people with intent driven content and intent driven audiences. So we know that if you're reading about newborn health topics, you are three and a half times more likely than average to be in market for a stroller. We're not the only people that write about newborn health. So we can find the individual pages on the rest of the web that do talk about newborn health, and we can unlock that very strong prediction that this purchase intent there. And so then we can have a premium service that buy those ads and delivers that value to our clients. Now we do that mapping and we've indexed hundreds of premium domains with opening eyes vector, embedding architecture to build that logic.
Damian Fowler (11:56):
That's fascinating. So in lots of ways, you're helping other publishers beyond your owned and operated properties.
Jonathan Roberts (12:02):
We believed that there was a premium in publishing that hadn't been tapped. We proved that to be true. Our numbers support it. We bet 2.7 billion on that bet, and it worked. So we really put our money where our mouth is. We know there's a premium outside of our walls that isn't being unlocked, and we have an information advantage so we can bring more premium to the publishers who have that quality content.
Damian Fowler (12:24):
I've got lots of questions about that, but one of them is, alright. I guess the first one is why have publishers been so slow out of the starting blocks to get this right when on the media buying side you have all of this ad tech that's going on, DSPs, et cetera.
Jonathan Roberts (12:42):
I think partly it's because publishers have always been a participant in the ad tech market off to one side. I put this back to the original sin of Ad Tech, which is coming in and saying, don't worry about it, publishers, we know your audience better than you ever will. That wasn't true then, and it's not true today, but Ad Tech pivoted the market to that position and that meant the publishers were dependent upon ad Tech's understanding of their audience. Now, if you've got a cookie-based understanding of an audience, how does a publisher make that cookie-based audience more valuable? Well, they don't because you're valuing the cookie, not the real time signal. And there is no such thing as cookie targeting. It's all retargeting. All the cookie signal is yesterday Signal. It's only what they did before they came to your site, dead star like or something, right? The publisher definitionally isn't influencing the value of that cookie. So an ad tech is valuing the cookie. The only thing the publisher can do to make more money is add scale, which is either generate clickbait because that's the cheapest way to get audience scale or run more ads on the page.
(13:57):
Cookies as a currency for advertising and targeting is the reason we currently have the internet We deserve, not the internet we want because the incentive is to cheap scale. If instead you can prove that the content is driving the value, the content is driving the decision and the content is driving the outcome, then you invest in more premium content. If you're a publisher, the second world is the one you want. But we had a 20 year distraction from understanding the value of content. And we're only now coming back to, I think one thing I'm very really happy to see is since we launched a cipher two years ago, there are now multiple publishers coming out with similarly inspired targeting architecture or ideas about how to reach quality, which is just a sign that the market has moved, right? Or the market moving and retargeting still works. Cookies are good currency, they do drive performance. If they didn't, it would never worked in the first place. But the ability to understand and classify premium content at web scale, which is what decipher Plus is a map for all intent across the entire open web is the thing that's required for quality content to be competitive with cookies as targeting mechanism and to beat it at
Damian Fowler (15:15):
Scale. You mentioned how this helps you reach all these third party sites beyond your properties. How do you ensure that there's still quality in the, there's quality content that match the kind of signals that makes decipher work?
Jonathan Roberts (15:32):
Tell me, not all content on the internet is beautiful, clean and wonderful. Not all
Damian Fowler (15:36):
Premium is it?
Jonathan Roberts (15:36):
I know there's a lot of made for arbitrage out there. Look, we, we've been a publisher for a long time. We've acquired a lot of publishers over the years, and every time we have bought a publisher, we have had to clean up the content because cheap content for scale is a siren call of publishing. Like, oh, I can get these eyeballs cheaper. Oh, wonderful. I know I just do that. And everyone gives it on some level to that, right? So we have consistently cleaned up content libraries every time we've acquired publishers. Look at the very beginning about had maybe 10 to 15 million euros. By the time we launched these artists and these individual vertical sites were down to 250,000 pages of content. It was a bigger business and it was a better business. The other side is the actual ad layout has to be good,
Damian Fowler (16:29):
But
Jonathan Roberts (16:29):
Every time we've picked up a publisher, we've removed ads from the site. Increase, yeah, experience quality,
Damian Fowler (16:33):
Right?
Jonathan Roberts (16:36):
Because we've audited multiple publishers for the cleanup, we have an incredibly detailed understanding of what quality content is. We have lots of, this is our special skill as a publisher. We can go into a publisher, identify the content and see what's good.
Damian Fowler (16:54):
Is that part of your pitch as it were, to people who advertisers?
Jonathan Roberts (16:58):
We work lots of advertisers. We're a huge part of the advertising market because we cover all the verticals. We have endemics in every space. If you're trying to do targeting based on identity, we have tens of millions of people a day. It'll work. You will find them with us, we reach the entire country every month. We are a platform scale publisher. So at no point do we saying don't do that, obviously do that, right? But what we're saying is there's a whole bunch of people who you can't identify, either they don't have cookies or IDs or because the useful data doesn't exist yet. It's not attached to those IDs. So incremental, supplementary and additional to reach the people in the moment with a hundred percent addressability, full national reach, complete privacy compliance, just the content, total brand safety. And we will put these two things side by side and we will guarantee that the decipher targeting will outperform the cookie targeting, which isn't say don't do cookie targeting, obviously do it. It works, it's successful. This is incremental and also will outperform. And then it just depends on the client, right? Some people want brand lift and brand consideration. They want big flashy things. We run People Magazine, we host the Grammy after party. We can do all the things you need from a large partner more than just media, but also we can get you right down to, for some partners with big deals, we guarantee incremental roas,
Damian Fowler (18:26):
Actual
Jonathan Roberts (18:26):
In-store sales, incremental lift.
Damian Fowler (18:29):
So let's talk about roas. What's driving advertisers to lean in so heavily?
Jonathan Roberts (18:34):
Well, I think everybody's seen this over the last couple of years. In a high interest or environment, the CMOs getting asked, what's the return on my ad spend? So whereas previously you might've just been able to do a big flashy execution or activation. Now everybody wants some level of that media spend to be attributable to lift to dollars, to return to performance, because every single person who comes through our sites is going to do something after they come. We're never the last stop in that journey, and we don't sell you those garden seeds. We do not sell you the diabetes medication directly. We are going to have to hand you off to a partner who is going to be the place you take the economic action. So we are in the path to purchase for every single purchase on Earth.
(19:19):
And what we've proven with decipher is not only that we can be in that pathway and put the message in the path of that person who is going to make a decision, has not made one yet. But when we put the messaging in front of it of that person at the time, it changes their decisions, which is why it's not just roas, which could just be handing out coupons in the line to the pizza store. It's incremental to us, if you did not do this, you would have made less money. When you do this, you'll make more money. And having got to a point where we've now got multiple large campaigns, both for online action and brick and mortar stores that prove that when we advertise the person at this moment, they change their decision and they make their brand more money. Turns out that's not the hardest conversation to have with marketers. Truly, truly, if you catch people at the right moment, you will change their mind.
Damian Fowler (20:10):
They'll happily go back to their CFO and say, look at this. This is working
Jonathan Roberts (20:15):
No controversially at can. During the festival of advertising that we have as a publisher, we may be the most confident to say, you know what? Advertising works.
Damian Fowler (20:27):
You recently brought in a dedicated president to lead
Jonathan Roberts (20:30):
Decipher,
Damian Fowler (20:30):
Right? So how does that help you take what started out as this in-house innovation that you've been working on and turn it into something even bigger?
Jonathan Roberts (20:39):
Yeah, I think my background is physics. I was a theoretical physicist for a decade. Theoretical physicists have some good and bad traits. A good trait is a belief that everything can be solved. Because my previous job was wake up in the morning and figure out how the universe began and like, well, today I'll figure it out. And nobody else has, right? There's a level of, let's call it intellectual confidence or arrogance in that approach. How hard can it be? The answer is very, but it also means you're a little bit of a diante, right? You're coming like, oh, it's ad tech. How hard can it be? And the just vary, right? So there's a benefit. I mean, I've done a lot of work in ad tech over the last couple of years. Jim Lawson, our president of Decipher, ran a publicly listed DSP, right? He was a public company, CEO, he knows this stuff inside a and back to front, Lindsay Van Kirk on the Cipher team launched the ADN Nexus, DSP, Patrick McCarthy, who runs all of our open web and a lot of our trade desk partnerships and the execution of all of the ways we connect into the entire ecosystem.
(21:38):
Ran product for AppNexus. Sam Selgin on the data science team wrote that Nexus bitter. I've got a good idea where we're going with this and where we should go with this and the direction we should be pointed in. But we have seasoned multi-decade experience pros doing the work because if you don't, you can have a good idea and bad execution, then you didn't do anything. Unless you can execute to the highest level, it won't actually work. And so we've had to bring in, I'm very glad we have brought in and love having them on the team. These people who can really take the beginnings of what we have and really take this to the scale that needs to be. Decipher. Plus is a framework for understanding user intent at Webscale and getting performance for our clients and unlocking a premium at Webscale. That is a huge project to go after and pull off. We have so many case studies proving that it will work, but we have a long way to go between where we are and where this thing naturally gets to. And that takes a lot of people with a lot of professional skills to go to.
Damian Fowler (22:43):
What's one thing right now that you're obsessed with figuring out
Jonathan Roberts (22:46):
To take a complete left turn, but it is the topic up and down the Cosette this summer. There isn't currently any viable model for information economy in an AI future. There's lots of ideas of what it would be, but there isn't a subtle marketplace for this. We've got a very big two-sided marketplace for information. It's called Google and search. That's obviously changing. We haven't got to a point to understand what that future is. But if AI is powered by chips, power and content, if you're a chip investor, you're in a good place. If you're investing energy, you're in a good place of the three picks and shovels investments, content is probably the most undervalued at the moment. Lots of people are starting to realize that and building under the hood what that could look like. How that evolves in the next year is going to really determine what kind of information gets created because markets align to their incentives. If you build the marketplace well, you're going to end up with great content, great journalism, great creativity. If you build it wrong, you're going to have a bunch of cheap slop getting flooded the marketplace. And we are not going to fund great journalism. So that's at a moment in time where that future is getting determined and we have a very strong set of opinions on the publishing side, what that should look like. And I am very keen to make sure it gets done. You sound
Damian Fowler (24:17):
Optimistic.
Jonathan Roberts (24:19):
A year ago, the VCs and the technologists believed if you just slammed enough information into an AI system, you'd never need content ever again. And that the brain itself was the moat. Then deep seek proved that the brain wasn't a moat. That reasoning is a commodity because we found out that China could do it cheaper and faster, and we were shocked, shocked that China could do it cheaper and faster. And then the open source community rebuilt deep to in 48 hours, which was the real killer. So if reasoning is a commodity, which it is now, then content is king, right? Because reasoning on its own is free, but if you're grounding it in quality content, your answer's better. But the market dynamics have not caught up to that reality. But that is the reality. So I am optimistic that content goes back to our premium position in this. Now we just have to do all the boring stuff of figuring out what a viable marketplace looks like, how people get paid, all of this, all the hard work, but there's now a future model to align to.
Damian Fowler (25:23):
I love that. Alright, I've got to ask you this question. It's the last one, but I was going to ask it. You spent time building maps, visualizing data, and I've looked at your site, it's brilliant. Is there anything from that side of your creativity that helped you think differently about building say something like decipher?
Jonathan Roberts (25:42):
Yeah. So I think it won't surprise anyone to find out that I'm a massive nerd, right? I used to play d and d, I still do. We have my old high school group still convenes on Sunday afternoons, and we play d and d over Discord. Fantasy maps have been an obsession of mine for a long time. I did the fantasy maps of Game of Thrones. I'm George r Martin's cartographer. I published the book Lands of Ice and Fire with him. Maps are infographics. A map is a way of taking a complex system that you cannot visualize and bringing it to a world in which you can reason about it. I spent a lot of my life taking complex systems that nobody can visualize and building models and frameworks that help people reason about 'em and make decisions in a shared way. At this moment, as you're walking up and down the cosette, there is no map for the future. Nobody has a map, nobody has a plan. Not Google, not Microsoft, not Amazon, not our friends at OpenAI. Nobody knows what's coming. And so even just getting, but lots of people have ideas and opinions and thoughts and directions. So taking all that input and rationalize again to like, okay, if we lay it out like this, what breaks? Being able to logically reason about those virtual scenario. It is exactly the same process, that mental model as Matt.
Damian Fowler (27:12):
And that's it for this edition of The Big Impression. This show is produced by Molten Hart. Our theme is by loving caliber, and our associate producer is Sydney Cairns. And remember,
Jonathan Roberts (27:22):
We do not as much tell the world what to think about. The world tells us what they care about. Data's lovely, but unless you do something with it, it's useless.
Damian Fowler (27:31):
I'm Damian, and we'll see you next time.