Introduction

So yeah, I could show you this sort of thing, but I know what you're for. I'll show you the dog. And let's be honest, I don't mind.

Setting the Stage

There's a fire exit down the back.

If you want to leave after that, it's fine. It's all downhill from there, I promise you.

A Memorable Dog Story

Look at this guy.

Second, who was the winner, like Lassie?

It's insane.

Who was better than that?

It took my sister four hours to get them to sit like that for that picture, and then she got it printed on a mug for me for my graduation. Prize possession, I can assure you.

Also, he had his picture in Pet World, a massive portrait picture of him, looking really glossy. He was regal. He was owning the pet store. It was fantastic.

The Time Constraint

So I think I've got 12 minutes left. I'll make the most of those.

A Humorous Anecdote

Now, I actually do have another story for you, if you'll bear with me while I actually create these three slides, Max. So I haven't been in Clerkenwell for a long time, and on the way here, I cringed a little bit.

So last year, after a long hiatus, I was back on the dating apps. And the first message I sent to someone said, oh, so I see you live in Clerkenwell. You might say I've been Clarking Well since 1987.

Ha ha. Baptism of fire. No response. Yeah, that's kind of the high point so far.

About Novella

So who are we? What do we do? Why am I here?

Mainly for the dog and the Clarking Well thing, but we are novella.

It might stun you, but I like to tell a story or two. and we've been really that's the heart of what we do.

Background and Experiences

I have spent 15 years in marketing, I worked at big agencies, I worked for American Express, I worked for Burberry and then I taught at Columbia and then at Imperial and I was staggered by what was being taught in these august institutions when it came to marketing. Fantastic theory, but zero practice.

I mean zero. They're great places, great people, but let's be honest.

Developing Simulations

1So last year, we started building simulations. The concept, I guess, doesn't land that easily, but I suppose if you think of a flight simulator, but for a job, really, it's like that.

Digital marketing is all about interfaces, data, making decisions, and it's about being fast. It's not about 10-year plans. That's not what people are doing these days in digital marketing.

It's optimizations. So we put them in the cockpit, I suppose. They're making decisions, they're dealing with live data, and they are learning from their failures.

That would be my thinking on it.

Exploring Simulation Types

Now, we have three simulations today. I'll whistle through this because I want to get to the demo. That's the main part.

Google Ads obviously dominates. It's still like 50% of global spend. Meta Ads is a further 30% of global spend.

So we've got two simulations there. You can see at the bottom, we sell universally.

So we've started working with some big insurers in the US who are using these to train people just to be better at marketing.

The third simulation though is, and you can see what's happened here, it's AI marketing, because everything has to be AI.

The first two simulations are really AI focused. They're co-pilot, there's all of this stuff, but we didn't badge it as AI.

So the third one, launching tomorrow, so wish me luck with the demo. If it doesn't go well, we're screwed tomorrow, right? It is all AI marketing focused.

AI Integration in Marketing Simulations

Now, I've got a detailed diagram.

I'm ready for the questions.

I would love to get into the detail of how this all works. But for now, I'll dig into the challenge.

So this is probably a bit small up there for you. So I'll zoom in on this bit.

Yes, it's no stories, right? And the idea there is we have an enhanced learning assistant called

Ella, it's a bit Knight Rider, isn't it really? So we call her Ella and she guides them through the whole simulation.

Now, I had a bit of a deal with my co-founder She was a senior data scientist at Spotify before she joined me to build this.

And she basically said, we'll build an AI simulation when you can explain to me what that means. And I couldn't really do it last year. I was just kind of saying, it matters.

You don't need to know AI skills. I mean, Josh does a much better job of it at Mindstone, right? They've built a whole curriculum around it and all of those great products.

I couldn't really pin it down. What do we mean by AI skills for non-technical people? I know in my gut that this matters.

But what I really found was I was trying to teach at universities, and I was teaching people that prompt engineering matters. But you see, without an incentive, they didn't care.

They were still doing the same lazy stuff. They were just taking pictures of things, sending them to chat GPT, and saying no until they got an answer that they liked.

And people would tell you, oh, Gen Z is the wave of the future, all of this. They're the laziest users of AI you will ever see. They're staggeringly bad at it. And they need an incentive.

So yes, we have a challenge introduction. They're going to work well, working well.

Challenge Introduction

They're introduced to this company. They hear about the challenge. They hear about the customer personas, all of that marketing stuff.

But I'm not here to talk about that necessarily. I want to dig into the real thing.

So I have to zoom back out a little bit and go back to the first step here. This will all mess up when I'm like that.

Making Strategic Decisions

1Ultimately, we are challenging the student here or the professional, whoever wants to use this thing. They have a choice to make at the bottom. They need to decide on their market segment. So who do I want to speak to with my marketing campaign?

They will not know. That is the point.

There are two AI tools that will help them, but they're not magic. If they don't use the AI well, they will not get the answers. And the incentive for them is clear. If you use the AI well, you could get an A in this class. They care about that.

So two things they can use, I'll use Samira because she is a bit more succinct. Customer personas, so we have 300 data points that are fed as context to, we're using open AI, we're using 4.0 at the minute, but we are going to use open source soon, we're going to build our own essentially and host it so we can control the costs a little bit better, but we're using 4.0 at the minute and they can ask these customers questions. What do you think of my company? And they know what the company is. They have an opinion of it. So she says, yeah, I'm constantly looking for new ways to improve engagement. It's very appealing to me. She should say that because she is the target segment they should go for.

So we plant learning paths within the context as well, so that the user might be led towards certain things, they maybe have a few red herrings, a few cul-de-sacs that they go down along the way, and then we lead them back out of those with the feedback at the end.

AI as an Enterprise Co-Pilot

So the main thing here is at the top. Now, we're speeding this up, so I'll just go a bit quicker with it now. I'll do this in the background.

This operates like an enterprise co-pilot. So they can see on the right what the sources are that the AI is using. They can, but they will not open these files and have a look at the contents. Never going to happen. in a billion years, but they could.

So we've used really descriptive titles so that they get a sense. But not all of these documents are relevant. Enterprise Ghost Strategy, well, great, but this company targets SMBs, so maybe don't listen to the AI if it pulls its sources from there. Financial Wellness Focus Group, well, we don't target finance, so don't listen to that document.

So it will always cite its source at the end so they can see where this information comes from. The AI has a lot of context. It's also trained on a lot of educational principles and concepts related to marketing, so it's able to understand, right, what are the right things to point this user towards? Where should I lead it away from? And if they ask it the right questions, they will understand where the audience should be at the bottom.

Now, I've preselected this so that we can get to the more visually exciting parts, like this one, which has a pie chart.

Using AI for Budget Allocation

The second stage, they've got to decide, where do I want to spend my money? I've got $150,000 and I don't have a clue. Well, if you ask the AI, it should be able to help you.

So it can explain marketing theories to them, the 95-5 rule might mean something to some of you. maybe to 5%, 95%, maybe not so much. It's quite a specific marketing thing, but we're taking a bit of the load off the professor here as well. So they can use this to teach multiple weeks of classes too.

It does operate a bit similarly to the one before. This is where they would decide whether they want to spend their money, but I want to show you the final part because it's the most exciting.

Now, it's interesting listening to, Sophia was talking earlier about how long it took her to get those demos right.

We wanted to give the students just access to 4.0 and say, create your own images, create an ad campaign, it'll be great fun. We were still getting the six fingers on a hand sort of situation and the brand logo was just being morphed into all this weird stuff. So it's not quite controlled enough for us yet.

Ad Campaign Creation

So we used it and curated a list of images and they select from them. they perform differently. So our simulation engine looks at the images they've selected, the audience they're targeting, and the predictive likelihood that that audience would like that image, and that affects their performance.

So here, I've preselected these. They can use the AI as well. It works like Google PMAX, if you've seen that.

And there are good and bad prompts here. So if they select a bad prompt, they'll get bad outputs. Select a good prompt, you've got a pretty good chance of getting some decent outputs.

So if I use this one, the AI will go and create me 10 ad copies that I can pick from. But these aren't as good, to be honest.

But anyway, I want to show you the simulation actually running, and then the co-pilot and the results, which is the big part of all of this.

So my contribution to design is that gavel you're seeing. I insisted on that. I nearly delayed the second product launch, because I didn't know that. I didn't know it was slowing everyone down that much.

But anyway.

Evaluating Simulation Results

Now they can see the results of their campaign. But they can see their ROI, how much revenue they got, how many deals they got from the campaign they launched.

This will make more sense to the user.

They've seen the video. They've been in the class.

But the big thing I want to show you is what has just popped up on the left. So automatically, the AI will tell them, and it's gonna be tiny, I'm sure, so I'll zoom in quite a lot more on that one. Well, at that level, I can see it.

So on the left, you can see it'll tell you three things that you did well and three things that you did not do well. And it is very, very specific.

Retargeting, 142 marketing qualified leads, an impressive ROI. Display, 30.5% ROI. Things that didn't work well, industry events, imbalanced budget distribution.

Now, we haven't launched this product yet, but we're already getting requests from companies saying, can you build one of those for our data? Why can't you do this with live data? Because what's happening is we're getting better at understanding the context that we need to feed the AI the whole way through the simulation.

AI-Driven Feedback

So every time the user is making a choice, it's being built up into a package that is sent to the AI. So it now understands these are all the choices that they made. These are the results that they've got, and then it joins the dots between them and explains things to them.

And they can ask anything, like, which image worked best for, I'll say, awareness. And this should bring up some interesting stuff here.

Comparing to Other Simulations

Now, there are a couple of other simulations in this space. They were built 15 years ago and they show it, but they just don't explain anything to anyone. They just show the results and that's it, which doesn't really seem good enough for me.

So for LinkedIn ads, your first image consistently performed well. It got a click-through rate of 1.2%, strong performance. For display ads, you got 0.3%, et cetera, et cetera.

This is kind of wasted in an educational environment. So yes, that is hopefully going to help them navigate all of this rather well.

Looking Ahead

We go live at Kellogg Business School on Thursday and Columbia Business School next Tuesday. So wish us luck.

Conclusion

And if you took nothing else from it, you saw a nice dog. So thank you all.