Creativity Is the Only Thing Left Tom Ollerton on AI-Eaten Marketing

(00:05) We're super excited to introduce Innovative Orchestrator, the first AI super agent for omni channel advertising. We're bringing humans and AI agents together so that marketers can conduct a symphony across channels to reach their most valuable customers. >> AI is not a matter of if, it's a matter of how and when.

### Show welcome + guest intro

(00:23) And we will help you solve that. Hey everyone, I'm David Burkwitz and welcome to another edition of AI Insiders with AI Markers Guild by Mark. I feel like we should have a tagline or slogan here, but no, we're not that official. Um, we do have some some pretty legit guests though and today is no exception uh because we have Tom Olton here.

### Shout-outs + community framing

(00:56) and Tom was actually introduced by a previous speaker, Dr. Cecilia Don. So, welcome back, Dr. Dones. And uh and and shout out to another past speaker, Katherine Montgomery. Always good to see you. And and just seeing like folks who uh we get to continue to learn from in the audience as well, which is always a great segue because these if you haven't been before, most of you have, but you know, they are uh they're more like community conversations.

### Setting up Tom’s session

(01:24) since these aren't webinars. And so Tom's got a few things to present. Uh he's from Automated Creative doing fantastic work out of London and beyond and and just uh yeah, as as soon as we got to start, you know, bouncing around ideas, I was like like we got to find time to go. And I I love those conversations that just start one-on-one.

### Bringing Tom on + light banter

(01:45) It's like we got to bring some more people into this room. And so Tom, glad you're glad you're in this room, >> mate. This is in the most intimidating thing I've done in ages. But um I will endeavor not to fall flat on my face completely. But you know, if I do, I can just close my laptop and we can forget this ever happened. >> Well, at least I don't I won't be the one to intimidate you.

### Audience roll call (where are you joining from?)

(02:04) It's some of these other folks who are staring at us. So, >> I look forward to it. Absolutely. And can I just ask where's where's everyone from? Loosely, can you just pop in the if I can you just do a little heart if I if I say like North America? >> Yeah, I mean that. Oh, North America. Thumbs up. Yeah. Okay.

### Regions represented + quick crowdwork

(02:23) Any any Oh, look at that. Eileen. Beautiful. Many Europeans. I saw Israel represented. Let's see. Uh, so we do have both coasts here and and the Gulf Coast or close enough to it. >> Any anyone from Northland here? >> That's so that's where I'm from in the northeast. So, uh, yeah. No Brits. Okay.

### Tom’s thesis: creativity after AI reshapes marketing

(02:49) So, I will slow down because >> I need need to use more more Z's than S's here. So, >> Z's. Oh gosh, that is going to be a push. But thanks guys. I really appreciate you not doing the rest of the internet and you coming here today. I really appreciate that. >> So, I'm going to give this talk basically comes around my belief that create creativity will be the only thing left once AI completely destroys all of marketing.

### Book origin story + interviewing senior marketers

(03:16) So this is what I I'm very passionate about and what I want to talk about today. Um but what I'm going to share is insights from this book I just got published. So I was approached about a year and a half ago to write a book. I was very surprised because I can barely speak, let alone write anything. Um and it was literally the most painful thing I've ever done.

### Writing process + what the interviews were about

(03:36) I had to get up between 5 and 7 every morning for a year and a half before my daughter gets up uh to write this book. Um, but the the highlight of the book really was interviewing 40 very senior marketers. So whether that was academics like Cessy or that CMOs or startup people or agencies, creative strategists. And I just had calls of them and I said, "Look, how do you take the smooshy lovely creative thing and combine it with the data thing? How how do you how do the how do those two completely different things go together? How does

### Why not write “an AI book” (it goes out of date)

(04:06) that work? Please tell me." and interviewed people like Roy Rory Sutherland um and Neil Patel some really interesting people for it. Honestly, I wanted to write a book about AI and I'll tell you why in a minute. But any book that I was published in, you know, July this year is going to be pretty much out of date.

### Choosing a durable frame: data + creativity

(04:26) And I've actually I read David's book on AI and marketing and he does a brilliant job of it at the start saying, "Well, he can't really do this. It's going to be out of date." And I David did a very elegant job of of um giving the 30,000 uh foot view of um AI and marketing. But I what I did I thought look if I write it about data and creativity it will be relevant to AI it will be relevant to quantum whatever all the rest of the things.

### Company background: Automated Creative

(04:49) So that that's that was the book I wrote and and it's available in bookshops and on the internet but the wider context is I I work for this company. Actually I'm the founder of this company. Uh it's called automated creative. So what we do and I'm not going to pitch the business to you today.

### What Automated Creative does (optimize ad creative with data + humans)

(05:07) We make and optimize ad creative. So we make all this stuff. We have a we have our proprietary tech that makes all of these ads and we do that using live data that comes back from the metas and the Googles the retail media of this world and combine that with human insight. So um we we founded this in in 2017 when we um uh when we kind of asked ourselves this question.

### 2017 bet: AI + agency services converge

(05:28) But like in 2017, we thought, well, well, what will happen if you combine AI with creative services? And so we we ran an event called I'll be back that looked at the intersection of creativity, ads, and AI. And my business partner and I, Alex, went, "Look, at some point, you're going to have AI and agency stuff squished together.

### Early belief in generative AI + building for the future

(05:47) So why don't we go and do that?" And that was all based on this thing that we got really excited about, which was called generative AI, right? So, we've been in that space since 2017 in one way or another. Um, and we try to build a business with this kind of future in mind. So, written the book. Um, it's coming up to Christmas.

### Uncertainty about the AI future

(06:04) Might make a beautiful present. I don't know. Depends how much you like someone. Um, and we and we've built this business. Um, so, so what how I feel is we're working into this uncertain AI future, right? There's a lot of people saying that this is definitely going to happen. and a lot of people throwing uh throwing sort of expert phrases around, but I'm really uncertain.

### Quote + “autopilot ads” vision

(06:26) I've been doing this for 10 years and I do not know what's going to happen next year. I've got a vague idea, but there's I saw this quote um and I'm sorry I'm I am going to read the the whole thing here, so I apologize. Let's get this thing out the way. And you're a business. You come to us uh you tell us what your objective is. You connect to your bank account.

### Zuckerberg quote: “just give us your credit card”

(06:43) You don't need any creative. You don't need any targeting demographic. You don't need any measurement except you to read the results that we spit out. And I think that's going to be huge. I think it's a redefinition of the category of advertising. Now, this is what Mark Zuckerberg said. So, he's basically saying here, ah, creative stuff, it doesn't matter.

### Engineering vs human problem framing

(07:01) Just send us a picture of your products and we'll crank out this gen Genai stuff in the background. Just give us your credit card and you'll and you we'll just spit you out money basically. And so he is right if marketing is an engineering problem but for any of the marketing folk in the room will understand it's a human problem of which um technology plays a part.

### Business anxiety + platform automation risk

(07:24) So we're in this kind of odd space where and certainly as a business owner in the creative space it's literally terrifying that meta is going to we're just going oh we're going to automate that completely so we'll see if that happens or not. Um and then also the I heard this amazing phrase the other day it really stuck with me.

### “ChatGPT is a money pit” + economics of compute

(07:41) He said chat GBT is a money pit with a website over the top. It's not it's not making any money. Um you got to you got to think where you know the end user is using some kind of AI that's coming down a pipe from Chat TBT. Well, where's that compute power coming from? Where are those chips coming from? There's this um and the uh the economists on this call understand this far better than me.

### Capital needs + sustainability question

(08:03) But uh Chat GBT they were trying to raise what was it the the equivalent GDP of France and Germany put together. They were trying to raise the same amount of money as the US spent on on World War II. I mean, and this is this is like the biggest most successful startup ever. So, no one's actually making any money out of this stuff.

### What happens when investors want returns?

(08:18) So, what will happen when that shakes down? When the Microsofts want their money back, what will happen? And and yeah, they're expecting to lose 44 billion by the end of 2028 and and they Yeah, it's a crazy time that everyone to me. I see that number and I'm like, wait, that's it? Thanks, David. >> So, yeah, everyone's betting on this horse that hasn't actually worked out how to make a a profit yet.

### Prompting vs creative spark (experience of AI)

(08:44) >> Interesting. Um, and then you go is I'm almost embarrassed to to show you this, but you know, you've you've seen like the uh the chat GBT user versus the non-Ch, you know, for the sunshine and rainbows and unicorns that explode in your head at the thought of a great creative idea or the hair standing up on the back of your neck as you go, ah, that's the thing we're going to do.

### More work, not less (reading every word)

(09:05) That's the brilliant exciting thing. Not like am I going to write a really long prompt today? Right? You know, so and I think I don't I don't know what everyone else's experience of using AI is, but like now I'm finding myself going to do more hard work, right? Instead of going, "Oh, I'm just going to crank out this email to David and I'll just let the slot machine take care of it." Like, no, no.

### The talk title / framing question

(09:25) I'm going to read every word he said. I'm going to think about everything I want to say and I want to send it to him. So, I'll be interested to know later if people are going on a similar journey because I want this, not this. Um, so, so how are we going to use data and creativity to build your brand if AI kills us all? So, let's do a little bit of a history lesson.

### Oldest surviving advert (setup)

(09:46) Uh, you I nearly asked you, does anyone recognize this? >> Of course we don't. Of course, >> I share that quote all the time. >> It's an absolute banger. So, this 1477, right? So, this is the oldest surviving advert in the world. >> Wow. >> Not the oldest, but the oldest surviving, right? And it's about this book. Can I beg the differer? >> I'm I'm so sorry, Tom.

### Pompeii pedantry + defining “oldest branded”

(10:08) You're on a great role, but um >> I was in Pompei uh two summers ago and there's still an advertisement for one of the senators. It was like for his campaign and it's on a street corner. I'm being very pedantic at this point, but um >> my ads as I I don't know if you know um ads transparency tool. It lets you toggle between political ads and and brand ads.

### Back to the 1477 example

(10:35) So obviously this this all just point well taken >> you know so thanks thanks for the you death starred me do you know what a death starring is when someone has an idea and you find the one little thing that goes down the thing and explosive >> yeah I know I've been used to doing that my dying day anyway >> I I am hoping we have some like Mesopotamian scholar here who can prove us all wrong >> right so this is the oldest surv I branded. Okay.

### What the 1477 ad contains (price, CTA, viewability, sampling)

(11:05) So, my my oldie English isn't that great, but I'm going to point out some things. So, it's um it this here says good and cheap. So, it's a good price, right? And it's a and it's a guide for priests called Sarum Pie. Um and it says kind of go to this place to pick up this this book from Samuel Caxton's uh print prince works in Houndsditch in London.

### “Same as today” takeaway

(11:27) And the really interesting thing is it says that the book is written like in this font basically. It doesn't use the word font, but it's like it's it's basically saying when you read this book, it will be as read easy to read as this thing. Um, and then this says do not remove like leave this up.

### Modern marketing concepts in ancient ads

(11:43) Right? So, what um any of the marketing people in the room will recognize is that you've got good and cheap. You've got a a price point. You've got a call to action which is go to this shop. You've got do not remove this which is viewability. And then you've got sampling which is look the book is written in the same font or printed in the same font as as this notice.

### First banner ad (setup)

(12:02) Right? So, the one of the oldest surviving adverts is doing a lot of the same things that we're still doing today in 2025. So, fast forward a little bit. Does anyone know what this is? >> Oh, that that I know. I don't want to give that one away. >> Thanks, D. You see, Adam, you see David let me have the stage. It's good. I appreciate that. There you go.

### First banner ad details (AT&T, 55% CTR)

(12:18) >> This this is internet >> the first ever banner ad, right? And it had a click-through rate of 55%. But still pretty good. Um, so, uh, and this was for AT&T, uh, and it clicked you through to a website that talked about AT&T were doing something with some galleries. No, no one really knows. But actually, in a lot of ways, this is a much worse ad than one that was written, you know, several hundred years before.

### “You will click here” + CTA commentary

(12:42) It's just it's blank. You are going to do this. It's it's a kind of rarely rarely used technique these days like you will you will click here. You know, there's not many called call to actions that say yeah. Anyway, and then moving forward uh slight a bit a bit further forward. Sorry about the slop here. We're getting the point across.

### Amazon desks + recommendation algorithm origin story

(13:00) So, in Amazon's original office, um you you know better than me where that actually was. They didn't even have desks. They had doors. They had like doors on on on on bins and they were working up doors and there was like a coffee machine and the carpet was kind of all disgusting. was covered with coffee and I can't remember the name of the developer but in in his spare time what he did is he he wrote um a a recommendation algorithm that basically said if you bought this book you might also like this book and when Jeff Bezos heard about this he

### “We’re not worthy” moment + impact

(13:33) uh he saw it in action and came into the room and he knelt down on his knees and judging by um uh the age of some of the people in this room you all remember rains world they're going to you know we're not worthy moment. So you he did he did a we're not worthy moment uh to to this this guy um because basically this algorithm but you like this book you might also like this book went on and on and on and obviously it's developed massively but it's probably in terms of revenue the most successful marketing campaign of all time and in

### Recommendation engine scale + revenue

(14:03) terms of cash through the till it's probably you know and obviously grand derivation of the original thing but that this recommendation algorithm was was a really creative use of data, right? That's what this talk was about. >> My information is dated, but at one point when they were still primarily books, it accounted the recommendation engine was close to half of their sales, >> right? Yeah.

### Too much data (even in 1998)

(14:28) That I've heard um yeah, similar similar. >> So, so that's that that that's creativity and data together. Um and then gosh, one of you guys know there's a lad called Jim Stern in the US. He's he's head of the US Analytics Society or something. Sorry, David, if you're here or something wrong. Yeah, he he he did a he did a he was in a book.

### Biggest challenge: “there’s so much of it”

(14:49) He was I interviewed him and he said in 1998 he said to 50 um brands, what's your biggest challenge with digital data? And can you guess what anyone said? Come on, Adam. You can't be quiet now. I've asked you a question. >> Not understanding it. Not knowing what is coming from or what does it mean? >> Okay, any other guesses? >> It's in the format and it's not what we need.

### Data without action is distraction

(15:14) No, what they came back with was there's so much of it. There's so much data in 1998 like oh like oh like so long ago >> I got too much data. How much you got now? It's like everyone I interviewed for the book they're like how many departments creating data? Everyone goes oh everyone's got so much data. Everyone's got so much data everywhere.

### “How much actually makes us do something?”

(15:34) Um so and one of the things I learned from writing the book is um is this is if data doesn't inspire action it's a distraction right so how much of the data that we have got even before you consider AI is how much of that data are we scratching our chins going h and how much of actually is making us do something right that data is very abundant it's everywhere but how much of it is actually making us do a thing or we just got it because we can get And the really weird thing for anyone who doesn't work in Adland is we do this

### Fake data for awards (Cannes Lions anecdote)

(16:09) really weird thing even if there's no data we like we we pretend that there is and one of the people I interview for the book I mean some of you guys will will be familiar with with Kand Lions is the very peak the very peak of the advertising accolade tree the the star at the top of the uh the awards Christmas tree and there was a a journalist who was um covering the event and there was a campaign that ran in Latin America that was banking the unbanked through some kind of telephone network. I can't quite can't quite

### Campaign “never ran” + the lesson

(16:39) remember what it was. Um and so the journalist, she went and spoke to the CMO and said, "Look, can I interview you about this campaign? It's really interesting." And the CMO said >> it never ran. >> It it this campaign never ran. The agency did it and then they said, "Can we enter it for awards?" And the CMO was like, "Yeah, sure.

### Data “shadows of people”

(16:58) " Like knock yourself do I'm not paying for it, but you know, go for it. So, so even in a world where there is this incredible amount of data, advertising people want to pretend that there is data that doesn't actually exist just so they can win awards. So, we have this kind of odd odd relationship with data. Um, and uh I interviewed Rosie and Ferris Jacob.

### When things get big, all you have is numbers

(17:18) You probably got some >> old friends of mine. Yeah. >> Yeah. I worked with Rosie. Yeah. >> Um, so we interviewed those guys for the book and they opened my mind. Um, and they they talked about this this idea when things get really big, all you have is numbers. If you got a large organization or a country, business, whatever it is, at a certain point, all you can do is talk in numbers because you can't go out and speak to literally everyone in the organization and all the clients.

### Smoothing/rounding hides the edges

(17:45) It all kind of comes down to a spreadsheet. And what that means is stuff gets smoothed and rounded. And I think that is one of the big dangers of AI is that if we're just dealing in in large numbers with the numerical data sets, it's going to round things off when it's the edges that matter. So interesting case of point. Does anyone know who this is? Any guesses? North Americans thought you should have a swipe at this.

### McNamara’s fallacy (optimizing to one metric)

(18:07) >> He looks like a general. >> This is Robert McNamara who was he was the CEO of the poor motor company and he was drafted in to run the military operation in Vietnam. Um kind of odd choice. So, um, one of the one of the errors that Far Ferris, um, and Rosie talked about was the idea when when businesses get distracted by one data point, right? They go, but we're just going to optimize uh, uh, to this thing and we're not going to we're not going to um, we're not going to um, think about anything else. And so it's it's

### “Basket of metrics” to avoid blindness

(18:38) quite gruesome this this stat but Robert McNamara is called the Magnamara's fallacy is that he said we will win the Vietnam war if we increase the number of uh enemy casualties if that number keeps on going up eventually we will win right if that that goes north at some point we just have to win but what he didn't do was understand the politics was understand the mood of the nation and there was all of these softer metrics that weren't weren't looked at and so he just focused on this kind of very gruesome bloody metric when actually

### Abundant data ≠ important data

(19:10) there's a lot more things that worked into it and Rosie in the book talks about having a basket of metrics. So how do you have a bunch of different metrics that all related to each other that work together to help you grow the business or whatever your your goal is because people are making this mistake that abundant data doesn't equal important data.

### Dashboard addiction + missing what’s hard to measure

(19:28) Anyone in the marketing industry um will be able to fully understand that you you can get dashboards you can look at dashboards all day. You could probably look at a a different dashboard every hour for seven, eight, nine hours a day, right? And that's abundant data, but that doesn't mean it's important. And what Ferris and Rosie were were telling me a lot about was that people make the mistake that if something's hard to get, if data is hard to get a hold of that it see it's not as important.

### Assuming unmeasured data “doesn’t exist”

(19:56) And some people make even worse mistake that if you and just because you can't get that data, they assume it doesn't exist. So, we we are very quick in marketing to go, "Hey, here's this lovely dashboard. Everything looks really shiny. It's abundant data, but is it the most important data?" Um, and I'm uh Cersei is on the call.

### Focus group story: small print + glasses

(20:14) I'm going to embarrass her really, but she was also in the book um and one of the people who inspired me to write in the first place. And she she talked about when she was uh very new to advertising and what this brand was trying to do was to take print ads and kind of turn them into turn them into um uh online ads.

### What spreadsheets can’t capture (watching real people)

(20:31) And so what they did is they had a focus group and there was this uh this lady who had um you know not not such great eyesight and she was looking at the ads and the people what do you think of these ads? Would you buy this blah blah blah blah and she was having to take her glasses off to to read the ads to see uh to see what it actually said because the print was too small.

### “AI wouldn’t catch this” + point about observation

(20:51) And Cessy was like well look there's there's no spreadsheet. There is no deck. There is no data point that would have covered that. She had to sit in the room and watch that person have to take her glasses off. So, how how often as marketers are we just looking at the dashboard and no further, right? >> So, we're get we're training AIs to look at our numbers and crunch them for us.

### Video as richer training signal (world modeling)

(21:11) But in this instance, AI wouldn't have been able to do anything because it wouldn't have been able to look at a person uh examining the text. And I I saw a brilliant article um the other week about um is it was it um the guy from Meta who's um who's saying that the future of generative AI will all be about video because you can you can read a text that say when a glass falls off a table it will smash on the floor but actually watching that on a video tells you much more about the real world than something in print which is will be a

### Data is “shadows of people”

(21:41) really interesting development of AI um but anyway so slightly disruptive but data is isn't a comprehensive record of the truth it someone said to me once in the when I was researching the book they said data is the shadows of people but we love data, don't we? It's so clean. It fits in a spreadsheet.

### Goodhart’s Law (when measure becomes target)

(21:58) We can show the CFO, but it isn't. It's just shadows. It's not the it's something that happened. It isn't what isn't what was going on in the mind of the person. So, the next thing that I learned from uh doing the research for the book, and this is this is a a guy called Goodart, and this is Goodart's law.

### Cobra bounty story (gaming incentives)

(22:15) He says when a measure becomes a target, it ceases to become a good measure. Right? So, how many of our targets have we given our teams? They say look just deliver on that target and everything will be great. But then what everyone starts to do is game that figure to hit the target. Right? So a great example of this was in British colonial India.

### Cobra farms + the unintended consequence

(22:33) Uh there was a real problem with cobras cobras. They were getting everywhere causing a problem killing lots of people. So what the Brits did, what they uh decided to do was to give a bounty to anyone who brought a cobra skin to them. Right. >> So brilliant. Ah, they're bringing they're bringing they're bringing bringing.

### Bounty removed → snakes released → worse outcome

(22:53) But what happened is the locals much smarter than the Brits set up cobra farms and bred cobras to get the bounty. So then the Brits discovered this and then got rid of the bounty deal. And then what happened? Any guesses? The farmers just let all the snakes go back into the wild and then the population of cobras went through the roof in the area.

### You only get data back on what you test

(23:14) So what they did, the mistake they make, they said like look all we need is lots of dead co that we need cobra skins. That's the goal. So what they did, they gained it, produced loads of cobra skins, but essentially it all kind of went back in their face. So a a big big part of our belief at automated creative is that you only get data back on the things that you test, right? So AI will only give you data on the things that you've run.

### Whiskey ad example: “sorry” performs best

(23:39) AI will only give you insight to the things that are actually in your data set, but it's not going to tell you what to test outside of that data set. So concept who this is for but we worked on a whiskey brand a little while ago. Um and so our methodology at automated creative is to test the hypothesis right.

### Hypothesis testing with many ad variations

(23:56) So what we do is we use AI and autom automation to gen generate very many ads uh to test things like the which visual elements and which written elements within the ads are driving the outcomes that the brand want. So we were selling whiskey as a gift. So we tested things like occasions, new year, holidays, uh things, reasons like uh to say thank you and people.

### Unexpected insight + “human tangential thought”

(24:18) So mom, dad, brother, sister. So we tested this huge range of visuals, huge range of different messages. And two years running, the best performing messaging theme overall was ads that said sorry. So every time there was an ad that said apologize with a whiskey, make it up to them with a whiskey. Nothing says you really like a whiskey.

### Optimizing the wrong thing if you never test outside the box

(24:37) that would in in in in different markets would perform really well, right? So, had we not tested those things, we wouldn't have got data on those things. So, if we're just pointing our AI at the information that we have to get it quicker and cheaper and all the rest of it, actually, unless you're applying your human tangential thought to that, there's every chance that you're going to be optimizing the thing that shouldn't exist.

### Dating app story (talk to customers)

(25:00) So, getting pretty near the end here, but I interviewed a lady called Barbara Galiza, who is a performance marketer uh out of Amsterdam and told me a story I I'll never forget. It was incredible. So, she was working for a dating app um where women could meet other women and her cost per acquisition of new users started getting really expensive.

### “Just scale the best ad” vs real conversations

(25:21) So, so what the classic the the data only person would just be going, "Right, well, what's the best performing ad? Let's put all the money behind that." But she decided to have conversations. So she spoke to a large group of existing customers and said, "Why do you use this dating app or why do you use this app?" And the message that came back really surprised her.

### Insight: users wanted friends (not dates)

(25:44) And the the respondent said, "We use this app to make friends. This is a dating app. Why would you use a dating app to make friends?" And so she said, "Interesting. uh how how do you choose someone to be friends with on this app? And and she and the respondent said, "Oh, we we do it by looking looking at people." So, you make friends by looking at people.

### Repositioning + usage data nuance

(26:06) And so, she realized that actually, because this is what 10 15 years ago, that really what was going on is that the users of this app had a had a a stigma about the app, about online dating, about meeting people online. So what Barbara decided to do was to change all of their messaging, all of their positioning to make this app about making friends.

### People reveal why (AI can’t replace that)

(26:28) So when people landed landed on the the website on the the opening screen of the app, it was all about community and meeting people. And when she looked at the the usage data, no one used any of that stuff. They used all the dating stuff. So what she did, she had a conversation before, she expected better conversions.

### Investigative vs evidential data (serial killer story intro)

(26:46) So yes, AI is very powerful. Yes, it can give us lots of abundant data very quickly, but what it can't do is look beyond and go and have real conversations with real people. So, finally, and this is a really really hard uh story to tell because it's really not suitable for work. So, I'm going to I'm going to try and do this in a in a vaguely polite way. Right.

### Car facing the wrong way (pattern as clue)

(27:09) So, at the end of the the 70s and the early 80s, uh there was a serial killer called the Yorkshire Ripper in in the UK. And Rory Sutherland told me this story. Now, the way that they caught him has a really interesting lesson for marketing people in in an AI world. So, serial killer was killing sex workers, right? And so at the time the way it used to work is you would you would meet a sex worker in a car and you would and you would drive the car somewhere quiet and you would point the car at the wall.

### Police stop + why orientation matters

(27:41) And the reason you point the car at the wall is so that the seats that uh the people were sat on would obscure the the relationship that was happening. I think that's suitable for work. I think that's I'm doing doing okay here. >> Great. So, so what happened is the the two policemen saw a car and they noticed that the registration plate didn't quite match the the the the type of car that it was.

### Bathroom break + returning to investigate

(28:04) So, they went up to knock on the window and they saw a relationship happening. So, which was illegal. So, they they arrested uh the guy and the guy said, "I need to use the bathroom. Can I just go over and use the bathroom in that bush?" And the police said, "Yes, of course you can." So, they took him back to the station. They were processing him.

### Investigative signal → evidential proof

(28:21) And then one policeman said to the other policeman, they said, "Did you notice something unusual about the car?" And and one of them said, "Uh, no. What was that?" Said, "Ah, the car wasn't pointed towards the wall. The car was pointed the other way around." So, can anyone think why the car was pointed the other way around? >> He wanted to see >> anyway. I like that. Thank you.

### Weapons found + concept takeaway

(28:43) No, it was so they could get away quickly, right? No. blood. So they went back to the the the where they picked the guy up and then went to the bush where he'd gone to the toilet and they and they found like murder weapons, right? And that that was what convicted uh that was what convicted the York auction ripper.

### Investigative vs evidential data (definition)

(29:00) Now what um Rory Sutherland talks about is investigative data versus evidential data. So the car being faced the wrong way, you can't put someone in prison for 37 years to life for that because it has investigative value. It's like how can you go further? What else is there? The um the evidential data was the the um the weapons in the bush, right? So, so to his point like if we don't allow these exploratory procedures, we're only exploring a tiny part of the possible solution space.

### Curiosity beats “scanning” (AI isn’t enough)

(29:33) So AI is quick, it's cheap, it's powerful, it can do all these things, we can replace all of these people. But is it really is it just giving us some kind of scanned evidential value when really we need to go and look further? So it's my belief that AI is very powerful, but it isn't quite as powerful as being curious and going the extra mile yourself.

### Practical takeaways (bullet list)

(29:52) Um, so some practical takeaways. Um, so it's balancing quant with cost. Numbers will tell you what happened, but don't uh but people reveal why. Don't just stop at the LLM. Um, just because just because the answer was quick and abundant doesn't mean it's right. What did someone say? Excuse my language, but this some ad guy said that arrives at the speed of sound is still So, you know, sometimes you need to go further. excuse my language.

### Measure what matters + don’t let AI flatten the weirdness

(30:15) Um, and measure what matters. What what are you actually going to act on? Um, you know, there's big numbers, but hide the real insight. So, don't look at what the data is telling you. Look at what it's hiding from you. What is AI hiding from you by smoothing the edges? And data should be fuel, not a cage.

### Closing + invite to community

(30:31) And and embrace those anomalies and those odd signals that spark creativity. Don't let AI flatten out the weirdness where the great ideas live. Um, I host a little mini community of people who've um, seen this presentation and read the book. So, if you'd like to join that, you can do. But that is me. So, I am now expecting to get torn apart by Adam and the rest.

### Q&A kickoff (rabbis + serial killers)

(30:51) >> Tom, I'm going to start and then I'll I'll I'll let the wolves have you. >> Um, and I'm going to bring up two of my favorite topics, rabbis and serial killers. I promise not not one and the same because I I think there's a parallel here. So I I I was at this uh uh lunch round table with the rabbi where someone asked him uh uh about God's omniscience.

### Defining omniscience + link to AI prediction

(31:20) You all right there? >> So what does omniscience mean? Sorry. >> So knows everything, right? And so and so can so so if humans have free will, how can you Yeah. Then uh then how do you reconcile that with God who is a future? and a god who um knows the future. And this I promise I'm getting somewhere about AI in a second.

### God as “complete dataset” (metaphor)

(31:44) Uh we don't have to go too deep down theological hole here, but his response was God basically has all the data of what happened. And so so God has much better luck at being able to predict what's going to happen because God sees everything going on all over the world. and uh everything and and knows everything that has happened up until this point.

### Prophecy vs educated prediction

(32:10) So, so something that sounds like a prophecy is actually just a very educated prediction. Um, going from there, what what I wonder is if you're potentially giving AI the short shrift here, because if AI, for instance, if you're using AI to crack a case of of other serial killers, a um topic that comes up a lot with a name like mine, uh, then a AI then presumably AI knows that example and it knows the examples of any published police prince precinct all around the world in places that you and I have never even heard of. And so it's got all of this.

### “Seeing around corners” concern

(32:56) Same thing with these ad campaigns. Yeah. Then uh if you know so if it knows that this one thing happened here like it can make inferences and some of these things that might be unexpected to us but if it has all of the data and can tease out what are the most relevant parts then at some point like won't it feel like AI is actually seeing around corners like like how do we not get there and that we then like there's still some room for creativity but it's almost like we're kind of more and more marginalized on

### Response: AI doesn’t have the full data context

(33:32) that sense. >> So, so stick sticking with your your long theological point at the start. If the rabbi is correct, do God has a complete data set. >> He he has a an unbiased unfiltered 360 left, right, and center unfiltered data set, right? >> Um whereas AI doesn't. And this comes back to my point before about um AI is the is the shadows of people right so for example um you know purely in a AB test between two ads there's a red one and a blue one right and so there's a data set that the red did better than the blue right but unless you know that

### Context matters (why an ad stands out)

(34:14) um within that ad there's a attractive person going hey it's 50% off >> then it's incomplete data set right and that's even at an ad level right so if you then look at the context within which that ad is being seen. No one has that data point, >> right? So, you're imagine you're flicking through Facebook or Instagram, whatever, and you you're seeing lots of yellow things and then you and then a red ad pops out, right? You go, it's the red ad that worked.

### AI “hoovering data” question

(34:42) It was the ad that wasn't yellow. >> God has a complete data set. Marketers have a a woefully inefficient. But what I'm wondering is as soon as there's some case study in Paraguay that someone didn't respond to the ads because she couldn't see them. Uh and and that there's some other case study that comes up from some agency that publishes this as an award submission for Mosamb beek that no one in this room would been likely to see.

### Limits of generalizing across contexts

(35:10) But AI is able to to hoover up more and more of that data. Then it then doesn't it become more able to make enough of those inferences even if it's always going to potentially miss something for two different markets for two different brands two different times of year two different objectives different audiences different platforms different everything no it can't make that correlation the only thing that was the only thing that was saying was that they were ads and that's not a that's not a strong connection >> well all right well well well look I I

### Outliers matter (behavioral science)

(35:41) yeah I know I want to push David, you brought God into it. I'm going to, you know, I'm going to have to come back strongly. >> Oh, oh, no. This this is great. Well, well, this is also I, you know, I I wanted your uh point of view here. Uh so actually in because you know as a behavioral scientist the the majority middle is boring and plain vanilla and all the all the all the grist where your your good ideas and handwriting on the wall of why something is bad is in the is in the outer limits of the data set.

### Synthetic audiences (topic shift)

(36:14) It's in the outliers where you get your real insights, not in the not in the Joe average middle. >> And well, exactly. And that's also why the one other thing I I'm sorry I I did want to steal you for one more second, Tom, uh that I wanted to ask your opinion on, I'm sure you have them, is synthetic audiences.

### Synthetic audiences: useful but “easy button”

(36:35) And have you looked into this? And >> yeah, you know, we we um we have synthetic audiences as part of our part of the the way that we work. Now, my the big what I'm most down on AI for >> is when people go, >> it's great because it's quicker and cheaper. >> Right? So that story, I'll come back to your question at the start.

### Fashion brand example (synthetic models cut costs)

(37:03) Um so I was in Amsterdam. I was meeting a very successful uh person in the creative field at a at a a famous fashion brand you all know. And that individual sat down across from me and got their laptop out and there was the 10 images of models wearing clothes. I don't know anything about fashion so whatever. They look like clothes to me.

### “Leveling the playing field” argument

(37:19) I'm sure they were very cool. Um and he went right. So that that would have been 10 shoots, 10 crews, 10 lots of flights, blah blah blah edit model rights. And that used to cost me 700 grand euros. But now because of AI and having synthetic people, the clothes weren't synthetic. They were real clothes. It cost 35 grand.

### Everyone gets the same advantage

(37:41) And he was he was going, "Yes, 750 grand. No, it's only cost 35 grand." Woo woo. I said, "Look, all you've done is you've leveled the playing field because all your competitors will do exactly the same thing. Every go from 700 grand to 35 grand." And then their Tom and David fashion company who only had 35 grand now is at a level creative playing field with the biggest people in the game.

### “Easy button” critique

(38:01) Right. It's like someone getting on the Euro Star for the first time from the UK to Paris and getting off the train going, "Yes, I got a train." and forgetting that there's a thousand other people on that train. Right? So, that is the mistake that people are making with AI. They're going, "I can now do the thing I was already doing quicker and cheaper.

### Synthetic audiences: quick, cheap, but incomplete

(38:22) " But guess what, guys? For $50, so can everyone else. >> Everyone else is doing it. It's not just you. We're all on LinkedIn. So, to your point about synthetic audiences, brilliant. What a great quick cheap way to do a slightly worse job of the thing you were going to do anyway, right? So yeah, like we have them and it's great. So if you want to get a feel for what moms in, you know, rural Manchester might feel about a napper very quickly, yes, you can use a synthetic audience.

### Real observation still matters

(38:47) But to this my point, my story before about Cersei looking at a human being and seeing the glasses come off, it wouldn't capture that because what we want to do as someone I saw on LinkedIn the other day made me laugh. They said advertising loves pressing the easy button. If there's an easy button, we all look at who's got the easy button.

### “The joy comes from effort and listening”

(39:04) Let's press let's press it, right? But the the joy, the goodness, the empathy comes from effort and listening to people. What are this? As I said before, what is the data hiding from you as well as telling you? >> Well, I'm I'm asking too many questions, so I'm going to stop, but I'm loving this.

### Invite more questions

(39:23) Uh, who else wants to ask Tom something directly? There's an amazing chat going on, Tom. I'll have to send you out. >> Oh, really? Okay. Follow. Yeah. Oh, Adam, go for it. >> Yeah. So, Tom, this is awesome. Uh, thank you for uh putting up with my introjections. Thank you for taking us through all of this. um about uh 10 years ago uh tried to start up this company very early stages but the general thesis was if you have these ads you could deconstruct you know the creative um into 2 plus you know uh a taxonomy of 2 plus attributes that really matter and then you could predict

### Taxonomy question: can AI attribute what works?

(40:01) and instead of GCO you'd use these many attributes and you'd have all these massive permutations and it strikes me that how you know AI can do that um at much grander scale. Are to what extent do you currently use or do you believe in a future where you are going to be able to have a reliable taxonomy of um of what's working in an ad um or a reliable taxonomy of attributes and then kind of derive it from there.

### Example (3+ people implies “family”)

(40:35) So the example we were given was if you have three or more people in a travel ad, it implies family um family creative, it performs better for leisure travel campaigns and like people sometimes would have four or family or focus. It's like, no, you just need three or more people. And that distinction, if you run some correlation with outcomes on a campaign, is something you could back into if you've done enough with the ML feedback loop, you know, enough attributes to pattern match against the outcomes. Is this something that

### Predefined taxonomy vs emergent patterns

(41:06) ultimately doesn't need a predefined taxonomy? And you guys believe that AI will make this possible. I guess it's sort of a near infinite number of permutations to predict outcomes of performance-based messaging of any kind. >> Well, what an excellent practitioner question. Thank you. So, so there's a couple of roots to that.

### Visual recognition exists, but relevance is the issue

(41:28) I'll probably forget what one of them is, but so you know, visual recognition has been around for a long time. Um, and things like Amazon's recognition and you know, Google's and they've all seem readily available through an API. Um, and so we've got like a reams and reams and reams of this data. Like you know, we've been going since 2017.

### Does the CMO care about “dogs in ads”?

(41:46) Um, so yes, and there's a lot of a lot of suppliers in the space that that use that stuff, right? So there there's an argument that, oh, you just let the AI do it, right? So an AI would probably recognize a pair of Air Jordans or a beach or the color red. Um, things, right? They would recognize things like, "But yeah, does the does the CMO of does the CMO of Mars really want to know whether dogs work in his or her ads?" Right.

### Joy example: humans can’t even agree

(42:19) Well, unless dogs the central to their strategy, which they may well be. >> Yeah, I was going to say, >> so so AI will get better and better and better and better and better at that stuff. But I'll tell you a story about when I used to work in agency site, uh, an agency called We Are Social. I wasn't in those conversations, but they they were talking I had a conversation with one brand, I think it was Cabri, and they and they were having months and months and months of discussions about what the word joy meant.

### Current approach: define “strategic tags” with the brand

(42:42) How how would you describe joy? Like we all know what joy means, you know, we feel joy, but then when you try and button it down, they go, "Well, well, joy is it's just I can see Eileen. It's quite tricky, right?" So, so the idea that a an AI can understand something as subjective as joy when it's the humans who are going to train that AI can't even agree themselves.

### Reporting back in brand terms

(43:05) So our approach at the moment and this may change based on what happens with the technology is we go to the brand what is it you want to know about this audience that you've never been able to find out and what is the brand trying to go go from here to here sales awareness whatever it is and then what we do is we create what we call strategic tags that are relevant to that brand strategy and then we we can tag up ads based on those strategic tags.

### Human + machine to do what wasn’t possible before

(43:30) So then when we report back to our client through our LLM, the brand's able to go right um is is a is it rational or emotional in terms of messaging that works really well. But rational emotional means something to this brand, but means nothing to that brand, right? They're cats, they're cats and dogs.

### Best practice = copying (anti-pattern)

(43:46) These guys are rational and emotional. So there is definitely a future where AI could do all of this stuff and none of none of us will have jobs and we'll be working in the party or something. But even in the case of Joy or the dogs and Eminem like it it could assemble a very manageable universe of insights conclusions and then the CMO could quickly look through the 10 or 20 insights and say oh well those few I I'll take out right away because we don't need dogs in an Eminem app.

### What can AI enable that was inconceivable before?

(44:17) >> I would I would love to see a CMO that was going to go through a list of um tangential insights. Yeah. Yeah. Yeah. Good luck with that. Um Um, so yeah, so it's like at this point in time, it's it's the human plus the machine to do something that we couldn't do before AI, right? To my point about the the the fashion brand, cool, you can save $700,000, but what can you do with AI now that you couldn't do before, right? And that that is what marketers are not getting at this point in time.

### Not “cutting your way to growth” with AI

(44:47) They're going, I'm going to save time and money and keep the the CFO happy while the the cost of living crisis crushes over. I'll just well but you can't cut your way to growth and you're not going to cut your way to growth just using AI. The why I got into this business in 2017 was like what is possible creatively with AI was inconceivable and that's what that's why I get out of bed in the morning.

### Momentum + why these conversations matter

(45:09) That's why I show up to these kind of things is because I love meeting people who have the same vision and belief as them. >> Yeah. I I mean I I am curious though also like beyond Yeah. Beyond major brands, you know, for the Do you have a chip shop on the corner? Can do you have uh one of those? Yeah.

### SMB question: does Zuckerberg win for the long tail?

(45:28) Yeah. For the >> No, I'm in London. Yeah. Yeah. The green stuff. Yeah. Cool. >> So So So for your local chip shop, right? uh uh then Yeah. If Yeah. They just want to spend a dollar on an ad and get someone to come in and buy $2 worth of chips then like it like like does Zuckerberg's thesis win there or not really. >> Yeah, absolutely.

### Small business reality: they want to do the craft

(45:52) So, so me and Adam, we're going to set up the Adam and Tom burger company, right? Because we love making we love burgers, veggie burgers, burgers, whatever it is, right? So, like all we want to do is sit in front of a hot grill and mix up different types of meat to make burgers is what's the right kind of may burger burger burger.

### Long tail economics + AI as “marketing autopilot”

(46:08) We love it, right? The last thing we want to do is marketing. >> Oh my lord. We don't want to like what an agency or write a brief. We're we're artisans, right? So most of Facebook's money comes from the longtail from the chip shops, not from the not from the the blue chips, right? They obviously spend the the most per client, but actually all the all the longtail.

### Conclusion: great for those who don’t want to market

(46:29) So I think there's an amazing use for AI currently is for helping people do marketing that don't want to do marketing, right? you know, like I'm not a finance and ops guy, right? So any any technology that helps that happen automatically. So I can focus on the things I do, which is talking to people, meeting people, learning, communicating, or trying to um so there's all things that we don't want to do.

### For big brands, not fully autonomous (yet)

(46:50) And so for smaller long-term chip shops, mom and pop shops, I think you guys call them in North America. Yeah. Brilliant. Right. Instead of them not making burgers or following the thing that they love, they could just press the advertising button and all these ads appear and they got whether they're any good or not probably doesn't matter because they just want to spend time doing the thing that they love.

### Question: scale vs edge insights (movable middles)

(47:10) So absolutely Jai formemes that's a fantastic solution. But for the the the McDonald's, the Bose, the Mars, the Jack Daniels, the Formula 1, the Wreckits, the PGs that we represent, I don't see it at this point to to be entirely dependent on those formats. Love it. Who Who else got something for Tom? >> I guess I'm curious um and first of all, thank you Tom for the innumemerate insights as always.

### Tension: outliers vs moving the middle

(47:45) Um I'm curious about a few of the things mentioned in this conversation. So we see attention with um understanding that the most interesting things, the wonderful things, the unicorns and puppy dogs and rainbows never happens at the mean. It always happens at the edges. And so finding those uh unique points of information that we don't get when we average to the mean, um that's what's exciting for us creatives.

### Push/pull: scale for growth while finding diamonds

(48:13) However, for those of us who've been in in town for a bit of a while and were familiar with Joel Robinson with the movable middles, for those of us marketers who want to sell more stuff, we got to move the middle. And so, I'm curious about that tension and your thoughts on how do we navigate that kind of push and pull.

### Answer: humans spot patterns, but marketers must avoid “average”

(48:34) We need to do scale, but we also want to find the diamonds in the rough. So I say this a lot and no one ever listens to me but maybe this is going to be the thing we're listening this might this may well be no sorry people people are very polite and they smile and not so right the reason that the human species has evolved to the level it has is because we can spot a pattern right so like oh this dangerous thing happens over there let's not go people who die go over to the dangerous bit like we'll go over here we're great we we've

### “Pattern matching” leads to bland best-practice ads

(49:06) evolved quicker than also not an anthropologist ist. So laugh at me as much as you like, but loosely in like, you know, knuckle dragging salesman of a way, that's what I understand, right? We spot patterns, right? However, that's the worst thing we could do as a marketer. That's the absolute worst thing we could do.

### Burger ad example: average = invisible

(49:21) So, back to mine and Adam's uh burger restaurant. This is definitely going to happen. Um what we could do is we could spot the pattern. We go, "Okay, Adam, you go and find 100 uh like um burger ads. I'll go and find 100 and then we'll find the commonalities and then we'll have like the ultimate burger ads and it'll be right there.

### AI will also converge to the mean

(49:38) " or we'll go to the digital intern chatbt or the rapper of Reddit it's also called and we go what makes a great burger ad so what does what does uh what does chat GBT do goes well what are the most successful burger ads let's pull in all of those attributes together and squish them all together and we go and we're dropping up and down our burger shop and we've got the best ad ever possible because it's the average the mean the middle of everything right but what's going to happen no one's going to notice it because it looks like every other burger ad so as humans We've

### Stand out by being “nonhuman”

(50:07) evolved because we spot patterns. This the most powerful marketers in a world where any ad could be made by AI will be the ones that are able to make the thing that's different. So great example of this recently is the water category, right? It's all about mountains and volcanoes and purity.

### Liquid Death example (category disruption)

(50:25) Liquid death comes in a bloody can and looks like an energy drink, right? They they they did what the guys at Liquid Death didn't do is go well let's you know let's get a really fancy agency to find the median like what's best practice it's one of the things we hate in our business is this idea of best practice in my school best practice was called copying right so what brands go hey let's best wait let's do best practice right so they squish all these people together and they go right we're doing best practice no no you're just doing the

### Dad in formula ad example (testing the unexpected)

(50:55) same as everyone else in the category right So your cost acquisition is going to go through the flipping roof because you wallpaper you you fit in. So Cesy to answer your your question I think is is to go like if you if you want to avoid that middle and go to the edges you have to learn to be a nonhuman and stand out.

### Only get data on what you test (again)

(51:14) How do you stand out and not fit in? We were working on a um on a like a a infant formula product years ago um that was aimed at moms on Facebook and we found out that the best thing you can put in an ad targeted at a mom on Facebook is a dad. We tested a whole bunch of different things.

### Closing Q&A + community link request

(51:33) Best practice says but what's what's best practice for a informula ad? Have a mom and a baby. You're so beautiful. No, no, have a dad. Cuz guess what? It stands out. It cuts through. It's different. So to my point before, Cesy, it's like you only get data back on the things that you test. So if you want to get get to the good stuff, the unicorns as you said and the rainbows on the edge, you've got to think about the stuff.

### Wrap-up + thanks + next steps

(51:54) You've got to you got to think in a way that the machine can't so you can test them and then learn. Anyway, you've got me all excited now. Um sorry, >> Tom, I think you got a lot of us very excited. Can you, speaking of which, uh share the WhatsApp code again or put the link in the chat? Uh just >> um I'll I'll uh Yeah, I'll I'll I'll send it around to you.

### Closing remarks + goodbye

(52:12) You >> Okay, great. cuz we we'll make sure and I know there's some folks eager to join and uh uh if they haven't read the book yet that I'm sure there'll be a >> yeahong the way I'm I'm going to expect an email from Amazon saying that you know they're closing down some servers because of the orders for the book and you know I got you >> well well this is the crew to make it happen so uh >> and Adam don't actually make that ad because what you're going to have is the dog's breakfast not a burger shop >> details details Well, well, and Tom,

### Host sign-off + gratitude

(52:43) I'll make sure to send you the chat. It's been lively. So, thanks everyone for contributing uh behind the scenes as well. It's been a lot of fun for me. Uh Tom, hope you come back and join us sometime. So, appreciate you. >> Yes, I will. Um thank you so much for the questions and the opportunities and thanks for the introduction, guys.

### Final farewell

(53:00) Have a a wonderful rest of the day and uh and a beautiful end of the year. >> Thanks, guys. Thank you so much, Tom. This is awesome. >> Wonderful holidays and uh see everyone next week. But this has been terrific. Thanks again. Appreciate >> CeCe.