How to Use ChatGPT Code Interpreter to Analyze Your PPC Campaigns

Episode Description

In this episode, Sam shared some really interesting data analysis: using an actual customer export from Shopify – loading it up to GPT-4, asking it to conduct an RFM analysis of the data, visualizing the outputs, and then exporting the customer segments.

He spoke about how this took his agency only a fraction of the time to accomplish that which would take a data scientist hours.

Sam also shared some thought-provoking, contrarian views on AI and how we should do PPC in general.

Tune in to learn:

- Why it’s time to rethink paid search

- How to use ChatGPT code interpreter for campaign analysis

- When should you do influencer marketing

and more

Resources

• Sign up for Sam’s newsletter here: https://samtomlinson.me/newsletter/
• Follow Sam on LinkedIn: / digitalsamiam
• Follow Fred on LinkedIn: / frederickvallaeys

Episode Takeaways

1. Rethinking Paid Search:
   • Paid search should be broadened beyond Google and Microsoft to include platforms like YouTube, TikTok, and Amazon, which are increasingly used for search functions.
   • Marketers should consider the entire ecosystem where potential customers might search, including niche or less traditional platforms.
2. Using ChatGPT Code Interpreter for Campaign Analysis:
   • ChatGPT’s code interpreter feature can democratize data science, allowing marketers to perform complex data analysis without needing a data scientist.
   • This can be used for audience segmentation, profitability analysis, and refining marketing strategies based on detailed customer data analysis.
3. When to Use Influencer Marketing:
   • Influencer marketing is more suitable for companies with larger budgets that can afford to test less predictable marketing channels.
   • It should be viewed as a supplemental strategy rather than a primary method due to its unpredictability and lower success rate for consistent ROI.

Additional Insights:

• Marketers should integrate insights from various platforms to tailor their strategies to where their audience is most active.
• Continuous learning and adaptation to new tools and platforms are crucial as the landscape of digital marketing evolves.

Episode Transcript

SAM TOMLINSON: A significant portion, I think, of marketing investments, especially in smaller companies and companies that have limited budgets, is predictability, right? It’s, we have a certain bucket of money that we can allocate to do this, and we have a certain set of goals we need that money to do. Wait, what did you just say?

Oh definitely can never see myself using Optmyzr ever again.

FREDERICK VALLAEYS: Hello and welcome to another episode of PPC Town Hall. My name is Fred Valles. I’m your host. I’m also the CEO and co founder at Optmyzr, a PPC management tool. Today we have a great guest. We have Sam Tomlinson, who is one of the deepest thinkers, I would say, in the PPC world.

Whenever I have a conversation with him at a conference, he’s he opens up my mind to things I hadn’t thought about before. Some really interesting data analysis. He also has some contrarian views, so he’s always fun to talk with and sort of understand where he’s coming from. And it takes him a different perspective, I would say, than most PPC practitioners who are very marketing heavy.

STEM also has a bit of a financial background, so a lot of that comes into play. So let’s find out today what he’s thinking about in PPC, how we can all do better. And let’s talk about some of the AI that he’s using. And let’s get rolling with another episode of PPC Town Hall. What’s your hesitance on influencer marketing?

SAM TOMLINSON: I just think the,

a significant portion, I think of marketing investments, especially in smaller companies and companies that have limited budgets is predictability, right? It’s We have a certain bucket of money that we can allocate to do this. And we have a certain set of goals. We need that money to do. And when you think about influencer marketing, one of the biggest things is if it hits, yeah, great.

You will likely do that. The hit rate’s relatively low. It’s not, there’s not the predictability and the ability to, to forecast returns and to accurately plan a business around an influencer centric strategy. I think is. tenuous at best. And most likely for most companies, it is just non existent. So to me, it’s like, okay, great.

Influencer marketing is a great supplement. It is a great appetizer, but no one goes to dinner at a fancy restaurant for the appetizers. I

FREDERICK VALLAEYS: had a nice birthday dinner. The appetizers were pretty good. I could have probably filled up on appetizers. You can make a nice dinner. I hear your point, though. I mean, clearly, and this is why PPC keeps winning, right? It’s because it’s the most measurable, it’s the most accountable. And but, but ultimately I think there’s also a little bit of backlash and, you know, are we spending too much focus at the bottom of the funnel?

Probably. Right. And so we should be doing more branding, we should be doing more influencer, but it’s hard to measure. So yeah, it makes sense. Like that’s probably the last place then that you spend your money. And I guess that’s your point, right? Like if you’re a bigger company, then absolutely some of your budget should go towards that.

But if you have say a 10, a month budget, then it’s probably wiser to just spend that where you can actually measure it, continuously optimize it.

SAM TOMLINSON: And where you’re going to get, I mean, where Forecasting, right? Like we’ve had, but I think the other bigger picture of this is, you know, maybe people just aren’t thinking about search broad enough.

FREDERICK VALLAEYS: What do you mean by that?

SAM TOMLINSON: Like, I mean, I think every time you say PPC, right, people immediately go Google paid search. That’s what he’s talking about. Maybe a little bit of Microsoft because, you know, we have to pay attention to the company that has 5 percent market share. Sorry. All the big people who are listening to this.

But I mean, let’s be real. Microsoft is a second place company. That’s okay. I’m Microsoft shareholder. I’m very wealthy, but I’m in second place. I’m not really very wealthy, but I’ve done very well on that particular side. Yeah,

FREDERICK VALLAEYS: that stuff has been amazing. What? Yeah, the stuff has been amazing. Amazing. Not on the back of its Not on the back of search.

SAM TOMLINSON: And honestly, not on the back of surface either. Billions of dollars in service advertising for 2 percent market share. I mean, Microsoft is, I think the last product Microsoft probably won was Office, but they have been consistently very good across a number of different verticals. But anyway, you’re saying

FREDERICK VALLAEYS: that PPC is too narrowly defined?

I think it’s too narrow

SAM TOMLINSON: to think, right? Like, because to me, when you think about it, like, we’ll, Search is about two things, right? It’s about answer discovery. It’s about task fulfillment. Well, how task fulfillment I think is the part that everyone’s pretty clear about, like we do branded search. We have product specific search we have, et cetera.

Right. But there’s this whole ocean of queries that people are going to different places and search is kind of unbundling itself. Right? You’ve got more and more people searching on YouTube now. YouTube University is a real thing. It’s a great thing. That’s paid search. You get in front of people who are looking for X, Y, or Z or have A, B, or C problem.

That’s paid search. TikTok is one of the most popular search engines among those under the age of 28. That’s paid search, right? Amazon, number one shopping search engine on the planet, that’s paid search. Those are all paid search. They look different and they function different, but they’re all paid search, right?

Same thing for retail media, right? Fastest to, you know, one of the fastest growing segments of the advertising world in terms of dollars invested year over year is retail media. That’s all paid search. It looks different and functions different and seems different. But fundamentally it is paid search.

And when you think about it, it’s like, okay, well, maybe paid search marketers just need to think a little bigger outside of the Google box and say, well, where are people actually going? How are people getting the answers they want? And how do I participate in that in a way that’s helpful to my company or my brand or my client?

FREDERICK VALLAEYS: And so at Wachowski, your agency, how do you act on that? So, and because you said Microsoft 5 percent market share, I think in the U. S. it’s actually a little bit bigger. Good job,

SAM TOMLINSON: Microsoft.

FREDERICK VALLAEYS: Give him some credit. But, but still, I mean, okay. I’m sorry.

SAM TOMLINSON: 8%, 8%, 8%. Oh, so, and

FREDERICK VALLAEYS: then you got TikTok and then you got Amazon.

Amazon’s fairly big, right? Especially if you sell a product, but, but how do you make decisions on where are you going to invest? And then the other thing that we see is. Because Google’s been around so long and people have their processes and workflows and they have their tool sets and they have their APIs and their scripts, like, you can like scale these campaigns really easily, but then you get into something a little bit newer that maybe doesn’t have that great API, doesn’t have a great tool set around it.

Is, is it worth the effort? Sometimes. I mean, so I think for what we think about it, you know, I think, well, a big part of what we’ve done and a big part of what I’ve tried to instill in people is for a long time. Paid search has been about marketing to keywords, right? We’ve been so obsessed with keywords and it’s like, well, behind every keywords, a person and behind every, every person is doing this for a reason.

SAM TOMLINSON: No one goes to Google and searches for something out of the fun of it. You do it for a purpose. And I think if you start to, when you think about budget allocation, when you think about strategy allocation, when you think about it, is it worth it? Well, a lot of what we do starts with audience research, right?

We do use, you know, SparkToro. We use BrightEdge. We use Moz. We use Ahrefs. We use focus groups. We have data scientists that actually do run like market research studies. But fundamentally, it’s got to start with people, right? It’s got to start with that. And if you start there and you see like, okay, well, this is how this audience is behaving.

And we, for a lot of clients, even do ongoing like pulse studies just to see how things are changing because you can start to see pretty clear trends. Like, Hey, more and more people in this sector are going to this place. Okay. Well, that’s interesting. We might want to think about including that 10, but when we talk about initial budget allocations and strategy, a lot of it’s rooted in.

Well, where do people get information? Where are people going? Who do people trust? And what does that tell us about them? Sometimes the answer to that is do more paid search. And sometimes the answer to that is, do you feel just really live on Instagram? So maybe we should figure that one out.

FREDERICK VALLAEYS: And I mean, I find that fascinating because TikTok I hear a lot of Instances where the younger generation uses that as a search engine to learn how to do something or they go on a vacation.

They’re like, what should I visit? Very funny example recently. We took our repair to Hawaii with us and we were literally five minutes walking from this little shopping center with a nice cafe and all of these things. And by day five, she’s like, Oh, I had no idea that that place was like just down the road.

And I’m like, what? Did you not look at Google maps? Like, did you not research like on the map where we were? And it’s, it’s, it sort of seems like, no, they went on TikTok and they were like, I’m in this place, what is there to do? And so if it didn’t exist on TikTok, then it just didn’t exist. But, but I think of TikTok as like this consumption of a feed.

Right. But, so it sounds like. The videos live there in perpetuity. And the point of making, maybe making these TikTok videos is not so much to be in the feed, but to be findable when somebody is looking for that.

SAM TOMLINSON: I think that’s part of it. I mean, I have a home services client that we do TikTok ads for, and it’s been amazing for younger homeowners.

Because that’s where they go for answers. That’s their thing. So yeah, I think part of it is obviously you want to, whatever, go viral. Very low probability. Win the social media lottery.

FREDERICK VALLAEYS: Yeah. But I also think that’s maybe the. Flaw in the older way of thinking when we think about social, we think about going viral and like, that’s the purpose, but no, the purpose here is that people just do very basic searches and you don’t have to have the coolest you know, the water bucket challenge or anything.

Like, it doesn’t have to be that it just like, how do you fix a leaky toilet? I mean, you could get some business

SAM TOMLINSON: from that. Correct. I mean. I was talking to, you remember J. D. Craylor, right? From Moira, and yeah, so J. D. and I were actually just talking, and he mentioned how he found a niche video of a guy fixing a 20 year old HVAC system that happens to be the same HVAC system that he has, but this guy’s wore a GoPro, and he walked through this process, and he’s like, this, this saved me hundreds and hundreds and hundreds of dollars, because I was able to watch this video of this obscure thing that’s super old.

But now, if I ever need anything, I know exactly where to go. Like,

FREDERICK VALLAEYS: it’s And that’s the thing. So, yeah, I mean, you would do that search on Google. And if there was a great YouTube video explaining it, you’d obviously find that through Google. Correct. But you wouldn’t find the TikTok videos. You wouldn’t find the Instagram videos.

You’d have to go into those channels, into those apps. I mean, I’m really curious about this home services client. So like, what kind of videos are these organic videos? Do you advertise for them? Oh, we

SAM TOMLINSON: advertise. So we’re just promoting some of our, we do have organic videos that are posted, but right.

It’s like a, you know, window door, window replacement from there. Right. So we’re just posting videos about, you know, anything from how to know if it’s time to replace windows and like what that process looks like the economics behind there’s a lot of times. You know, in many cases, if you’re a younger homeowner, you may honestly have never even had to think about this, right?

If you’re 22, 25, 28, lifetime of windows is supposed to be between say 15 and 30 years, depending on the quality, it’s entirely possible that your family you’re in your lifetime. It’s never happened. So you don’t even know what to do, but now you’ve. You know, obviously with the larger macroeconomic and housing situation here in the U.

S., you might have had to buy a house for, maybe didn’t have updated windows, and now you’re looking at it and you’re like, all right, well, what do I do about this? My heating goes out of control. My, I got drafts coming in, you know, whatever it is. And now you’re like, okay, what do I do about this? Enter us. Hi, here’s one.

Here’s like how you could do like a temporary fix. Here’s. You know what this means in the long term and here’s Why if these things are happening consistently across multiple different windows, maybe maybe you should think about getting windows replaced because you know the the heat loss potentially is It’s more than what you pay on the globe.

FREDERICK VALLAEYS: I’m kind of curious, so I mean, like the typical consumer, so obviously a homeowner. So I guess you’re speaking to frustrations of that homeowner. You’re coming at it from the angle of like energy bills too high, drafts in your house, like pain points that someone understands and then bringing it back to, Oh, well, it could be your windows are.

Not energy efficient enough. I mean, that’s okay. That makes sense. Now you’re big on data and analytics, right? So how do you go about did? Yeah. So how do you go about measuring that? And how do you work with clients when they might be a little bit more of the mindset of like, Oh, did this video go viral?

Like, how do you bring it back to like what matters?

SAM TOMLINSON: So, I mean, for us, there’s probably, there’s two ways we can do it. Right. You know, one is for a lot of our clients, we allocate a percentage of their spend to tests. And we agree on what those parameters look like and what those tests are, but those aren’t, those dollars aren’t, it’s, it’s, you’re paying to learn.

So it’s like, all right, let’s, you know, continue to evolve our program. So, you know, like our TikTok thing has come out of our test budget. So that was easy. The second thing, honestly, for them is. It’s leads in the door and we’re, you know, original source tracking was pretty solid. It’s not perfect and it’s certainly not the whole thing, but it’s been a very, it’s painted a very clear picture.

And then as part of our sales and intake process, we also followed with like, okay, well, how’d you hear about us? Why’d you do that? Kind of like the team’s actually legitimately good about asking questions. So props to them for that. And the data validates it that, you know, younger homeowners are disproportionately turning to things like YouTube and Tik TOK.

Even Instagram to at the very least understand their questions more because that’s just the content that they’ve grown up, they’ve grown up on. That’s the content they’re familiar with. They’re, they’re happy to listen to a podcast or to watch a video or to see a social post as an answer to their question.

They don’t need tender links. Like, you know, I feel old now, man. You made me feel old. Like I, I do, I I enjoy my 10 blue links. Thank you very much.

FREDERICK VALLAEYS: I do too. But I also kind of enjoy the, the generative, the generative portion above that, that takes those 10 blue links and distills it down. Not having to click, having to scroll.

Like I love it.

SAM TOMLINSON: Yeah, that part’s been, I’m slowly warming up to that being a useful thing. Because there are still times when I read it and I’m like, that’s just wrong. And then I look at it and I’m like, yeah, that’s not right. No, no, no, no.

FREDERICK VALLAEYS: So that’s interesting because I mean the generative results on Google or on Microsoft, they, their job is not to generate from a blank page, but it is to take in those top organic results, the top paid results, and distill that down into a narrative.

So I haven’t seen that many of the still gets, yeah, I got confused on time series data. Like, if you ask for, like, you know, what are market shares in certain periods? It’s like, it was this and this. I’m like, well, that wasn’t a question friend. The question was. You know, what was market share in 2022 versus 2023 and you gave me 2021.

SAM TOMLINSON: Why? So it does get some of that wrong. Yesterday I was researching iPads. I want to buy one for my kids. And I was like, well, it helped me understand the difference between iPad air, iPad, And iPad mini. And so I was like, okay, iPad seems like the cheapest ones. It has great for the kids. So don’t need a lot, but there’s 21, 2021 like what’s the difference?

FREDERICK VALLAEYS: And it gave me a good explanation of the 2021, but the 2022, it was struggling. It was like, well, maybe it has a different processor. Maybe it has a better screen. But clearly it was on an outdated model and it couldn’t help me that last bit. And so that’s the other issue that we still see there.

Damn SGE. It’s getting better. And that’s the thing. I mean, think back to Google, right? When I started at Google, when you started advertising on Google, they would index the web. It would take several months to run a fresh index. So when you search for something and that landing page was updated yesterday, like if you happen to be outside of the update cycle, you’d wait two months for that to show and look at Google now, like within seconds of something happening, it’s on the search engine.

Yeah, that’s going to happen with generative to like, it’s already a getting there. Yeah, it’s going to get there.

SAM TOMLINSON: I mean, I think it’s. I think Google is uniquely positioned to win in that particular space, which is a very unpopular take among the AI bros. But

FREDERICK VALLAEYS: why do you think they’re uniquely positioned?

SAM TOMLINSON: I mean, I think the, the biggest challenges with LLMs, right. Are, you know, hallucinations, it’s kind of the model becoming a little untethered. It’s, you know, all these questions about, you know, what are you generating versus, is it, Is accuracy, right? Fundamentally, who has the world’s biggest collection of data that they can use to ground that model?

In fact, right? Who has the world’s largest index of entities that understands the relationships between certain entities at a better level than anybody else? You know, who has access to arguably more location and business data than anybody else on the planet, Google. Okay. That seems pretty good. And then who also has, this is going to make the open AI.

I mean, all the respects in the world, open AI, but like Google’s AI team is. It’s amazing. They’re, you know, these are the people that figured out Go. These are the guys that are going to win, you know, the next iteration of the gaming things. These are the people that built the transformers and some of the initial concepts that OpenAI turned into ChatGPT.

FREDERICK VALLAEYS: Exactly. So they bought DeepMind, the company that initially beat the game of Go and then AlphaGo. So that was mind blowing. And then they were the people behind the transformer models. So you can look up the research papers and then it’s OpenAI that commercialized it. In the beginning for nonprofit purposes, but now obviously for profit purposes.

Yeah.

SAM TOMLINSON: Everybody’s a nonprofit right up until it’s better to be profitable.

FREDERICK VALLAEYS: Yeah. I mean, I honestly don’t know what, what was the reason for being a nonprofit in the beginning?

SAM TOMLINSON: I don’t know. Probably some altruistic thing.

FREDERICK VALLAEYS: I mean, now Sam Altman is seeking 7 trillion to make chips. To build the next models.

SAM TOMLINSON: Yeah, just don’t take that from foreign entities or all of a sudden you become a problem. Yeah, exactly. Also Australians a lot. Yeah. That’s more than the values of Microsoft and Apple combined. So that’s a, yeah, some change. Yeah. Maybe, maybe, maybe through a Google too. Maybe it’s a couple of them. You probably could.

Yeah. I mean, Apple’s market cap is what, 2 trillion, give or take? Yeah. They’re all like 2, 3 trillion at this point, which is amazing too, because just a few years ago, like we were looking at the first trillion dollar company and now it’s just like, wow. And just like you said, Microsoft stock and Nvidia stock and anything connected to AI is, is basically off the charts at this

point.

What is it? The Magnificent Seven?

FREDERICK VALLAEYS: Is it? Yeah. Well hey, but so like, yeah, I think the contrarian or not, I guess you have the contrarian view on Google going to dominate. And I like the reasons that you’re saying, I mean, they do have access to all the world’s information. And if you think of an LLM as taking that information and sort of the ranking system behind Google, and then you feed that to the LLM to digest it for you and make it easier to consume and maybe have a conversation with it like that, that makes sense.

But so I tried Gemini. Which is the rebranding and the rebranded Bard. Right. And that’s the thing. Like, it feels like rebranded Bard as opposed to an evolution on Bard. Like I’ve tried it. I’m pretty disappointed.

SAM TOMLINSON: It’s not great on the text. It was actually solid from like a code interpreter standpoint.

It did some good stuff.

FREDERICK VALLAEYS: Yeah. So let’s talk about that. So code interpreter, right? Like, and this, we had some fascinating conversations where we met in Berlin. But code interpreter was open AI’s plugin, which was in beta initially. So if you wanted it to write some Python code or do like an analysis or an audience analysis, you could use it for that.

For that, now it’s part of the core GPT 4 model. It’s a built in plugin. But yeah, talk about Code Interpreter and similar capabilities in Gemini and what what’s interesting. I mean,

SAM TOMLINSON: so to me, I think it’s just a democratization of data science in an interesting way. That part’s helpful. I think it’s also just doing in some ways to data science, why it.

Editors did to parts of web development, right? You, you don’t need to be a specialist to launch your website. Obviously if you want a great website, you need to have a specialist. But if you just, anybody go to Squarespace these days and drag drop point, click, go team, and I think for a long time, right, the barrier to smaller, to midsize companies using data in the way that these.

Platforms want you to use that data. Was it, you didn’t have a data scientist, like you didn’t have somebody on the team who understood RFM segmentation or could run a multivariate regression or who can do any of this, if you wanted any of it done, you know, your best option was to pay a data science firm, a hundred, 200, 500, a million dollars to go do it.

Well, small advertiser, that’s not in the budget. So to me, code interpreter is just like, great. All of those tasks that. We’re previously unavailable to so many people. Now it’s as easy as literally export Shopify thing, upload to Code Interpreter, say, you know, let’s segment this data and it will prompt you through the entire process of segmenting that audience, you know, whether it’s, you know, doing an RFM model, doing a K means you know, doing a random forest, whatever you want to do, it talks you through it.

It, it democratizes data science in a really fascinating way.

FREDERICK VALLAEYS: Yeah, exactly. And so even if you don’t understand what those different models might be, the beauty is that you can ask it. You can say, tell me more. Why would this one be useful versus that one? What’s the difference? How much data do I need for this one versus that one?

Right. Yeah. And it’s literally like having a conversation with a data scientist one who’s very patient, by the way, and then he turns around and like, once you’ve made your decision, doesn’t have to go back to their office and spend days building the models. It just spits it out and then you still have like.

SAM TOMLINSON: There’s still drift, like the model, you know, every once in a while, it gets a little, but, you know, to your point, you can, you can save it once they write the code for you, you can save that, that, you know, snapshot of that particular thing. It’s a great, this is.

FREDERICK VALLAEYS: And that’s an important point, right? Generate from Python code for you.

Like you said, there could be drift. It could be hallucinating. So it could make mistakes. So the output that comes out of it, you do have to validate it. And then one mistake that we’ve seen people make is they think, Oh, next time I want to do a similar analysis, let me go back and ask the same questions, but let’s use the same prompts, but that’s dangerous because you have drift and drift just to explain to some of our listeners and viewers.

But basically the model keeps evolving and debates in the model keep changing. And so a correct answer today could very well be an incorrect or the same problems could give you an incorrect answer tomorrow. There’s no, it’s not like a calculator where you punch in the same numbers and it just uses the same logic gates to get you the correct answer.

It’s an evolving model, and that’s where you have to be dangerous. And so what you’re saying is generate the python code, validate it. If you think it works, then actually install it. We can reuse it in the future, which also saves you a little bit, the cost of regenerating the same code.

SAM TOMLINSON: Yeah. Which I pay for chat premium.

We’re okay. Spending some of Sam Baldwin’s money. As you said, I was getting 7 trillion.

FREDERICK VALLAEYS: Exactly. So you’re saying, I mean, basically pay for the 20 premium plan. Now, I am curious though, right? Because in the agency or like anything that we want to do at scale, like, do you have tips and tricks for, do you use the API?

Do you use Sheets plugins? Like how do you use

SAM TOMLINSON: everything? I use the Sheets plugins. I haven’t played too much with the API yet, but it’s on the to do list. Okay. It’s been a bit of a busy start to the year. I still honestly use the web interface probably more than I should just because it’s bookmarked and it’s easy and I can just pull it up, drag, drop, say what I want and it does it.

I should use the API more. It would make my life easier.

FREDERICK VALLAEYS: And now Gemini. So you said Gemini has a code coding ability as well. And that makes sense, right? So Google has long said that they, one of the strengths of their models is coding and different languages, different programming languages, as well as different, like spoken languages.

So have you tried that and how does that differ from the open AI code interpreter?

SAM TOMLINSON: I think in terms of. The overall experience, the Gemini has been impressive in the sense that it’s getting to the right, it’s getting to very close to accurate, if not very accurate code faster with less revisions. In fairness, it’s only been what, two weeks since we announced Gemini was a thing, so I haven’t like, that’s.

Not a time, time for drift, but it’s

FREDERICK VALLAEYS: like a mid February or early February announcement, depending on when you watch.

SAM TOMLINSON: So yeah, this is mid February. It’s been two weeks since we announced Gemini and it’s been, I mean, I would say it’s been very good at the data science task. It’s been very good at writing basic code.

I’ve used it for like, you know, some tag kind of like triggers. I’ve used it for regular expressions. I also just hate writing regular expressions. It makes me angry. So I let Code Interpreter do it. Or I try to

FREDERICK VALLAEYS: get angry at regular expressions. And if you don’t know what regular expressions are, then you know, you’re welcome,

SAM TOMLINSON: right? To me, it’s just been a very, on the code side, it’s been very buttoned up. The code has been compact, it’s been useful, it’s been accurate. It’s very, very few issues with it. With ChatGPT, I would say it’s been 85 to 90 percent accurate. I think Gemini on the code tasks has been 90, 95%. So to me, that was a big, that was a big win.

I think on the text generation, like the actual response generation, OpenAI is still, is still ahead there. And like you said, it’s been a little underwhelming, but then again,

GPT 3. 5 was not great. Like when GPT 4 even started, it was, Glitchy, we’ll call it, right? But the whole point of the models is they get better as adoption increases, as data velocity through those models increases, you know, as user feedback is able to be collected and used to either adjust or reinforce or remove parts of the training, it gets better.

So I think if you look at, I think it’s a bit of a disingenuous comparison to say, okay, Gemini today, 12, 14 days post launch. Okay. Versus GPT 4, eight months post launch,

it’s just not a fair comparison. I know, I know, but then also I feel like, I mean, Google has had BART for a long time and Google Like you said, through Google brain, they invented the transformer. They’ve been working on this. They just held it back because they were too more concerned and open AI about.

FREDERICK VALLAEYS: The hallucinations and sort of the things that these models could do and say that might be incorrect. And kudos to Google for not trying to spread misinformation in the world that’s maybe seeded by these large language models. But at the same time, I think it’s not entirely fair to say that Google has started later, like they just But you make a very good point, which is, listen, if you need code written, maybe Gemini It’s great.

If you need a blog post written, you might want to go to GPT four. And I think that’s one of the things that we as humans still bring to these models, right? As we understand, okay, this model is really good at that. That one’s pretty good, but it’s also 10 X cheaper. So if I’m going to do this at scale or through like a sheets plugin, because by the way, now you have to pay for these sheets, plugins for, I don’t like it.

Who likes paying, right? Nobody likes paying, but sometimes you just have to say, I’m sorry.

SAM TOMLINSON: I was going to say, I mean, credit, you know, pay for Optmyzr. That’s pretty good. Right?

FREDERICK VALLAEYS: Yeah, yeah,

SAM TOMLINSON: yeah,

FREDERICK VALLAEYS: exactly. Well, so, and that’s the other so for us at Optmyzr and thank you for mentioning it, but we do bring in AI into the tool because we fundamentally believe that people don’t necessarily want to change their workflows.

They want to keep working in the tools they already know. And that should have AI capabilities added. So if you need new headlines, Optmyzr will suggest those to you. If you want to have a conversation about an account, you can do that through Optmyzr Sidekick, which is using OpenAI’s models, but but it grounds it in the truth of your account.

And we do a lot of the calculation to say like, okay, by the way, is CTR. That’s clicks divided by impressions. Like if you ask GPT 80 percent of the time, it does it correctly. And then 20 percent of the time it does it wrong. And now you go to your client and you tell them some like completely wrong CTR.

Yeah, you’re, you’re getting fired. That’s not good. That’s not, that’s not great, Bob. I know. So so tell us, so you, you brought up the example of Shopify data and bringing that to do some segmentation. Is there a cool example that you’ve done maybe using Google ads data directly? Is this something our listeners could try today?

SAM TOMLINSON: So, I mean, I would say the closest we’ve done with that would be using our own audience data. So CRM data, obviously, you know, on the Google ad side, right? It’s. The importance of your own data is getting bigger and bigger, right? We’re, we’re playing more and more with demand gen campaigns. We’re doing more and more with some of those lookalikes.

Well, I would say probably one of the cooler examples is, you know, we took audience data, like a whole customer profile, right? Used. GPT to segment that into different buyer types into different personas and cluster that and then uploaded each of the clusters as a unique audience, which actually gave us much better performance on the lookalike segment, which was cool, because obviously, you know, lookalikes, the more homogenous The seed, the more relevant, the lookalike is most, it’s more likely to be.

And those are quite sensitive to the initial condition. So that was fun. I think you could, that’s an, it’s an easy application, but it’s a good one. You know, take your customer data, build segments, right? I think so many marketers just. Take all customers, throw into Google, go team. Well, you could add a single step into that process and get 20, 30, 40, 50 percent better results in terms of efficiency, in terms of lower cost per lead, in terms of higher return on ad spend, however you’re measuring that, by just saying, hey, chatGPT, hey, Gemini, here’s a list.

Can you tell me, run an RFM model on this? Run a K means cluster on this. Give me back something. Or, you know, the other easy one you can do if you have an e commerce store is. Especially for Shopify, all of your user data is in there, right? Hey, Shopify, give me this list without the lowest 30% of clients.

Or, Hey, Shopify, or hey, you know, GPT, Gemini, whatever. Here’s a list of my actual costs, which is another fun thing you can do. Here’s the people and what, here’s the SKUs they bought. Here’s the cost for each sku. Tell me which customers were the most profitable. Which customers did I lose money on? What things do those customers have in common in terms of source?

In terms of what ad platform they came from in terms of what their purchase path looked like. You can find a ton of insights that from like a paid search, paid social standpoint, you can then leverage pretty easily.

FREDERICK VALLAEYS: Yeah. Very cool. All of a sudden. And so. Go ahead. Sorry. So the, I mean, you talked about code interpreter, right?

But just to be very clear, anyone who’s using that today, you’d simply go to GPT four and there’s a little logo for uploading a file. You upload that file. That basically starts the code interpreter ability. And then it it has this little arrow at the end of the response. If you click on that, that’s where you see the actual Python code.

So this is not highly obvious what’s happening, but that’s where you’d have to click to see that it had done some code generation. And actually used statistical models and actual programming to come back with an answer. And where it’s always been a little bit annoying to me, I don’t know if they fixed that.

So you can tell us if you know they have, but you do this analysis and now you get these segments. Right. And then you’re like, well, now tell me something in common, like search terms, these types of people might’ve looked for, if that is not present in your data set. It has a hard time going from like code interpretation in Python to going back to the large language model capabilities.

And so sometimes you kind of have to split up that task to get it all the way to where you need it to be.

SAM TOMLINSON: Yeah, that’s a hundred. That has not changed. It is. But that same task, right, would have taken a data scientist three, four or five hours potentially.

FREDERICK VALLAEYS: Oh, yeah, absolutely.

SAM TOMLINSON: Gemini will crank that out in three or four minutes.

FREDERICK VALLAEYS: Well, and then to your point that, you know, people just upload their whole list and they shoot it on this one extra step. And I think that’s where we’ve all just been conditioned to like that one extra step that was your half million dollar data science project, or that was like waiting five hours for the data side.

And then that’s, if you’re lucky waiting five hours, I mean like give it. Waiting a week for that data to come back. Right. So that’s why we didn’t do it. But now, now it’s available for 20 a month and three minutes of your time, like having a quick conversation and and it spits out the data in a structured format, which you can then bulk upload back into Google ads or whatever platform you want.

SAM TOMLINSON: Exactly.

FREDERICK VALLAEYS: And you’re saying you see these amazing gains from it. So, so I think it’s

SAM TOMLINSON: been, it’s been very, very promising. Obviously, you know, we’re still early on deep on demand. I think that’s an hit or miss. But I love lookalike audiences. I think we’re finally getting away from the the broad bros.

So I’m happy to see that there is kind of a, you can’t just put everything on broad and YOLO anymore.

FREDERICK VALLAEYS: Yeah. Well, you know, audiences, it’s it’s kind of a social thing, but it’s in search and it works for a reason, right? I mean, PPC looks more and more,

SAM TOMLINSON: you know, social. It’s more and more like a DSP every day.

FREDERICK VALLAEYS: Yeah. And like you, you brought up the point too, that you do these focus groups because you’re, you’re fundamentally trying to understand who are my customers, who are my prospective buyers, and that’s a little bit to point that over 20 years of Google ads and being so metrics driven, like we may be forgotten about the human element and the person on the other end of that click, and that’s what you’re saying.

Now we can go back to that, but we can also. Layer that on top of AI generative, do some really cool stuff with it. So I think that’s really bringing it full circle. So yeah, I love all of that stuff. So Sam why don’t we do a couple of rapid fire questions here? Oh, good. Here. Let’s do those. All right.

So the first one, what’s something you wish you would have known before you started PPC marketing?

SAM TOMLINSON: Honestly, how much I didn’t know about creative, like how important creative was to it. It’s taken me a long, it’s been a long journey for me to go from, yeah, yeah, yeah, creative is like 20 percent of the equation to, oh yeah, creative actually is a much bigger part of the equation than maybe I wanted to admit.

But, you know, you.

FREDERICK VALLAEYS: Yeah, makes sense. It’s one of the few things you can still control. So even a bigger deal today than it was Yeah, a couple of years ago. What’s one common PPC myth that you would like to debunk?

SAM TOMLINSON: There are so many of those. Where do you want to start with this one? Oh bid modifiers. I hate, I hate the misconceptions around bid modifiers. Bid modifiers and most smart bidding are not compatible. You cannot upload a bunch of zip codes and be like, with my target CPA campaign, I’m going to target the zip code. That’s not how that works with bid modifiers.

You can target them? Sure. But it, what, almost all bid modifiers, except for minus 100 percent on device are ignored by smart bidding. So whenever you put in something stupid, like, Hey, this campaign, it’s using smart bidding. I’m going to use bid modifiers. All you’re doing is wasting your time. So please stop it.

Yep.

FREDERICK VALLAEYS: One of our most popular blog posts, infographics, which bid adjustments are compatible with which Bit strategies cause yeah, there’s a lot of bit strategies to choose from and like, it’s, it’s kind of weird which ones are compatible, but for the most part, you’re right in the modern day and age of automated bidding with these bid modifiers.

Yeah, that means I mean, the, the, this is going to have an obvious answer. So you’re not allowed to say open AI or chat GPT, but what is your favorite AI tool and how do you use it?

SAM TOMLINSON: Favorite AI tool. That’s not one of those. Let’s see,

FREDERICK VALLAEYS: I have one. So I use a talk notes dot IO discovered very recently.

SAM TOMLINSON: Yeah, I have I do know that I have Otter AI, so it’s just like, does the same thing for my zooms. That’s been a game changer. I like that one.

FREDERICK VALLAEYS: So tell people a bit more about what Otter AI

SAM TOMLINSON: does. It just joins, so it joins all of my meetings, all my zooms. And it’s basically just like a virtual assistant and it jots down, it basically just transcribes the conversation and summarizes it in bullet points and sends them to you.

So if you’re ever in a meeting, obviously if you’re like, you know, back to back to back zooms and there are specific details, You go back and listen to a whole recording of it, you can literally just pull this up, it tells you when a certain part of the conversation happened, you can, you know, read the, the actual transcript, or you can go listen to that part of the conversation and just, you know, It’s just a big time saver for me in the sense like I don’t feel compelled to take notes and I don’t have to worry about, Oh my, like what, what was that?

What was that number? Where was that list? What was that file name? Whatever.

FREDERICK VALLAEYS: And I think it’s also brilliant with selective hearing, which I think we all suffer from. At some level. So well, you’re making a face there like you don’t, I’m sure. Oh, I do. I, I a hundred percent. But, but yeah, I mean, so like to, at the end of the call, it’s like, Oh wait, that’s what the client wanted.

That’s what they asked me like that completely in one ear and out the other, because my mind was somewhere else, right? So yeah, I love otter AI and that sort of technology. Yeah. That what is one important skill marketer should develop to stand out in 2024 and beyond?

SAM TOMLINSON: I’m gonna say the same one I’ve said for a while, which is they need to understand how finance works.

I think so many marketers have just grown up in an era where there was the illusion of accountability versus actual financial accountability. And we’re finally getting to that point where, you know, the, the marrying of marketing and finance is finally starting to happen. I think people on the financial side are recognizing the importance of growth.

for talking to me. And how some of the metrics marketers have can be valuable. And I think some marketers, you know, I think DTC has had to grow up a little bit faster, maybe than other industries are finally starting to say, you know what, I need to, I need to get away from row ads. I need to get away from the click through rates and impressions and clicks.

And I need to start thinking about impact to the underlying organization and what that looks like and how much of that is incremental versus how much of that is not. And I need to start making better decisions because, you know, fundamentally, I think marketing is an exercise in capital allocation, right?

Can’t do that without understanding finance.

FREDERICK VALLAEYS: Yeah, yeah. And then some really great blog posts that you’ve done. These are a little bit older, but I think you had one on the recession. Like what happens in the TPC in a recession, you put your foot or you like back off and two different outcomes. So some good, where can people find your stuff by the way?

What’s the best place to

SAM TOMLINSON: website? So I’m on Twitter. It was usually the one I responded to the most, but I have a newsletter every week. I think you read it sometimes, Fred.

FREDERICK VALLAEYS: Oh, you see my open when I open it? Of course you do. Yeah, you’re,

SAM TOMLINSON: you’re, you’re one of the circled ones. But it’s Sam Tomlinson.

me slash newsletter. Very original, I know.

FREDERICK VALLAEYS: We’ll put that in the show notes. And then final, RapidFire. You’ve talked about a number of platforms, but what’s one of the more emerging ones that you’re most excited about?

SAM TOMLINSON: I really like Perplexity. I’ve been playing around with that one for a minute. That to me is it’s a really interesting use case in terms of, you know, how it’s, Unlocking information discovery using AI, but that I think is just an interesting platform.

I’m also playing a lot more with retail media platforms than I ever had before, because I just think they’re both fascinating and criminally underutilized.

So that’s been interesting because on the one hand you have like perplexity, which is wildly advanced. And on the other hand, you have retail media networks, which are Google in 2014. Love you guys, but like, let’s be real. That’s the tech that’s currently there. And it’s like,

FREDERICK VALLAEYS: you

SAM TOMLINSON: know, one, one foot in the past and one foot in the future.

And you’re trying to figure out how these two things come together. And it’s going to be interesting.

FREDERICK VALLAEYS: Great. And then we, we had one audience question, so we’ll finish with that. But it’s speaking of one foot in the past, one in the future. All right. How do you make sure you got both feet in the future? And how do you stay up with all of the latest developments and any tips on that?

SAM TOMLINSON: It’s a never ending I read way too much, I would say. I mean, I, in terms of staying up with trends, I think a lot of it is. Refining where you get information from so for me, that’s been a continual process of seeking out sources that maybe other people aren’t looking at and looking at sources might be outside of marketing.

So if you know, Benedict Evans, he writes a newsletter every week. I subscribe to that 2 PM. Which is web Smith. They write something on commerce and trends. It’s more big picture. Quite a few financial ones that do, you know, stock analysis trend analysis, earnings call write ups, et cetera. So you can get some good info out of those.

Trade clubs, search engine, land, search engine, journal, the Optmyzr.

FREDERICK VALLAEYS: Let me, let me interrupt you. And final, final question, but given that you’re so into finance and reading stock blogs and that all what are you buying today?

SAM TOMLINSON: What am I buying today? What did I buy? So let me tell you, let’s find out what did I buy this morning?

Yeah, we do. Ooh, we

had a good day. I bought some deposit, DPST regional Bank, ETF, and bought some Alibaba. Love. Charlie Munger. Bought Alibaba around 68, bought deposit at 68. And I still keep buying a few, still keep leveraging into Google and Microsoft. Just dollar cost averaging myself up, but

FREDERICK VALLAEYS: at the end of the day, it’s going to

SAM TOMLINSON: be

FREDERICK VALLAEYS: what financial advice.

This was not financial advice.

SAM TOMLINSON: Do whatever you I’m not a licensed broker. Have whatever you

FREDERICK VALLAEYS: want. Have fun. But that’s what Sam bought this morning. So that’s what I bought. Anyway,

SAM TOMLINSON: we’re up 5 percent today. So I didn’t do something. I did something right today.

FREDERICK VALLAEYS: I don’t play it day by day. It’s just. too depressing some days and happy other days.

But like, anyway, Sam, always fantastic talking with you. Thanks for sharing with our audience. And if you’ve enjoyed this conversation, want to see more of them, please give us a like, subscribe at the bottom and join us for future episodes. Sam Sam Tomlinson dot M E. That’s where you’ll find his blog. You can subscribe to his newsletter and he’ll be speaking at many conferences in Europe some in the United States.

By the time you watch this, he may have already spoken, but he’s a regular on the circuit. So look him up. And if you want to catch him in person, he is actually out there in the real world, which is awesome. I myself look forward to grabbing a drink with you at Munich. Munich. Yes. Thanks Fred. So We’ll see you for the next one.

Take care. Sounds good. Let’s do it