Frederick Vallaeys on why digital marketers will still have jobs and what they’ll look like in an AI world
"I think that this is full of opportunities for PPC, but it's going to be different from what we've been doing for the past 10 years."
We all know automation is dramatically changing the way we approach PPC. From setting up to managing and optimizing campaigns, our roles are evolving. In his new book “Digital Marketing In An AI World,” Frederick Vallaeys, former Googler and co-founder of campaign optimization platform Optmyzr, discusses why understanding the fundamentals of PPC matter more now than ever, what machine learning can (and can’t do) and the skills and capabilities marketers will need.
I sat down with Vallaeys at SMX Advanced in Seattle last week to chat about the book. We discussed his time as an early AdWords team member and one of its biggest advertisers, and why he’s optimistic about our various roles as digital marketers in the age of automation. Have a listen or read the interview below.
This interview with Frederick Vallaeys, CEO of Optmyzr, has been lightly edited and condensed.
Let’s talk about it on a very basic level: What we’re talking about when we talk about AI, when we talk about machine learning and when we talk about automation and what marketers need to really understand about what these terms mean that gets thrown around so much.
At a super high level, you have to understand where the technology is coming from and how fast it’s evolving. And fundamentally, understand that this is a very real change that’s going to have a big impact. But it’s also not as all-encompassing as a lot of people might think.
So if you look at traditional media or movies, AI is like these humanoids that do basically everything, and that’s not where we’re at. In PPC, it’s mostly machine learning, targeting very specific problems like, “Hey, how do we set the right bid if we have a ROAS goal.” Because we have to put a CPC bid because that’s how the Google auction works right?
So the automation is very specific to that. But then also understanding, I think, at some level the history of it. So AI is actually not that new in Google Ads. So quality score, I mean, that’s like over a decade old. That was the original machine learning system — figuring out what’s the predicted click-through rate on a certain keyword and ads combination for a certain query.
Which you had a hand in building.
Yeah, I was on that team. So it was kind of a fun part of the book too — lending that insight or perspective and telling some stories of, you know, playing hockey with the founders and cross-checking Sergey [Bring, Google co-founder] and then still be lucky enough not to get fired that day.
Talk a little bit about your history at Google and some of the highlights of the stories you have.
First of all when I was joining Google, I had put a pretty short list of companies I wanted to work at, and Google was sort of up and coming but it wasn’t that big of a deal, yet — which is why I could play hockey with Sergey, and I cross-checked him because I didn’t really know who he was at the time. Then he became really famous.
The acquisition of Urchin. Flying down to San Diego and getting to meet that small team, that eventually became Google Analytics. And that’s one of the most foundational things in online marketing nowadays, right? What would we do if we didn’t have these numbers?
Some other highlights: So I actually started advertising while I was working at Google because I saw the light. I was like, “Oh my God, people are spending $30,000 dollars a month — and that back in the day made you a Tier 1 advertiser, that was like the biggest of the biggest.
You were able to be an advertiser, which informed how the products got developed.
Because there’s sort of that gap between the people building these tools and people using these tools, and you helped kind of make those connections.
Right, and you see that even here. I’m at SMX, and there is a keynote from Google. It’s a great vision that’s being put out there, but there’s a certain gap between that vision and how do you get there. And so that was kind of my role in Google. I started advertising and became a pretty big advertiser and affiliate advertiser, and I don’t know which of my keywords are converting. And guess what? Conversion tracking didn’t exist back in the day.
We all had this concept of, yeah, PPC works great, but really, we still didn’t know what was actually working. We were just throwing money at it. And so, I built my own little conversion tracking system. Then the team took note and said, “That’s pretty cool. Maybe all advertisers should have that. Can you go talk to the product team?” That was one of my early product things.
So you say in the book, thinking about the skills that we need to have, that many of us don’t need to understand kind of the technical details, but do you need to understand what the “smart” systems are doing, and not doing, which I thought was really important. What factors do PPC marketers who are not in the weeds on the technical side of this need to understand about the factors and ways that these tools and systems are working to be able to make better decisions and make sure the machines are doing what they’re actually supposed to do?
And that’s one of the big premises and, I think. One of the big things PPC marketers are going to have to work on going forward is helping the machine learn by giving it more data about your business. And sometimes that’s scary, right? You talk to plenty of businesses that say, “We don’t want to give Google more data because you’re going to do something nefarious with it.” I worked at Google. It’s not a nefarious company. I don’t think it’s changed since I left in that regard. And ultimately one of the safeguards is it’s still an auction, right? So, you know, you give extra information, and all Google really does with it is it tries to increase your bids when it thinks there is a higher likelihood of it converting and vice versa.
But now, I think the foundation or the fundamentals of PPC are still so important to understand. I talk to people, and they are using tROAS bidding. I ask them, “Well, how do you go from a target ROAS to an actual CPC based on your conversion rate and values?” And they can’t do the math. And if you can’t… I don’t expect you to do it on the spot, but you should be able to think through that and actually kind of explain and then be like, “Well, if it’s using historical conversion value per conversion or per click, what kind of data do we have that maybe informs certain audiences that perform better or worse, and what can we do with that? Should we supplement Google’s understanding by building a new audience that takes a look at something that matters to us?”
Because that’s one of the fundamentals of machine learning, that it uses historical data to make predictions about the future. And so the kind of big shift is thinking about your data in a new way. And one of the things you talk about is informing these algorithms with more data, and that there are gaps that Google has. So when teams are talking to each other about, OK, we want to use smart bidding — and maybe it’s not even Google’s Smart Bidding, its any smart bidding, it’s a third-party tool — and you figure out, OK this is the data we have, and this is our goal. Can this help us achieve our goal?
So, what are the steps to thinking through that process and thinking about, “What are the gaps?” How does somebody come back to the office and say, “This is our challenge. How do I either do this on my own or work with teams because I don’t have access to this data right now?”
Right, and that’s really one of the big things that PPC experts are going to play a role in, and I don’t think it’s a simple answer, but that’s kind of why we will have jobs going forward. It’s just our jobs have been redefined because of this new question. There are several problems with it. One of them is Google doesn’t really tell you what data it has or what data it uses. So one example I gave was for quality score. At one point, we said, “Hey, we should look at the lunar cycle and figure out if that impacts predicted click-through rate.” And we found that system-wide, it really didn’t – so we decided not to use it as a factor. Now, if you’re a tarot card reader or a psychic, you know, maybe it does matter.
I know, when I read that I thought it also maternity wards. So many … deliveries happen based on the full moon. Any maternity ward nurse will say, the full moon is when we’re full. So I was thinking, Oh, that’s a great example of, you know, on the whole, lunar cycles may not have mattered, but there are things like that.
There are specific verticals and industries. And then if it does matter for you, that’s when you have to figure out, how do we inform Google about this. Right?
And then the other question that’s really hard and that we’re going to have to play a role in as well is the maternity ward says, “Yeah, we’re full.” But really, how full? Are you 25% more full than unusual? And if you’ve reached capacity, then does that actually mean you want a bid higher or do you want to bid less?
That becomes a strategy decision, and it becomes a matter of you probably having to plug your data into your own machine learning system — and if you don’t have enough data, you can still do some statistical analysis on that. But now you have data points. You can say, “OK for a full moon we’re going to increase our bids by 20%.” And so increasing a bid in the old days used to mean “boost your CPC 20%,” now it could mean “increase your CPA target by so much or lower your ROAS by another number” because you know that on the back end your conversion rate is going to spike. And even though you’re supposedly willing to pay more for an acquisition, the added conversion rate’s going to make up for that. And again, that goes back to fundamentals. If you don’t understand how conversion rate and CPA targets interact like that, you’re just not going to be able to make the right decisions.
I am thinking about, you know, the newborn photographer … who says, I know I’m going to be really busy at this time. What are some of the resources and best ways to start making sure that you know what you’re talking about and know how to plan?
Read the book. [Laughs] There are three sections in the book. The introductory section is about what is this technology. And I think understanding the technology that’s driving it helps you position yourself, but also helps you think about, Oh, how could I leverage that similar technology for my own benefits?
And one concept that I introduced in my talk at SMX Advanced was about automation layering, and it is huge. So you have these in-engine automations like smart bidding, and they do certain things really quite well, but they don’t do it perfectly. So how do you layer your own system on top of that to either change the targets or take action when you see that there’s some outside factor that influences it? And I think just by understanding the core technology, the core machine learning, you will be in a better position to build your own solutions for it.
And I loved your analogies for the doctor, the pilot and the teacher roles. You know, this is one of the things we talk about — is what our role is going to look like and will we have jobs? Can you talk about the doctor and the pilot and the teacher roles and how those apply?
I really just wanted to simplify it and give an analogy that people can easily understand. And so when you think about a doctor, you go to a doctor with an ailment, and a doctor knows there’s 17 different drugs that might work for this, and some of these drugs are more aggressive have more serious potential side effects. And so they kind of take a look at the patient, and they say, “Well how serious is this issue?” And then they prescribe what they think is the right level of drugs. And they know about interactions. They know about other conditions that you have. So what’s the interplay? Bringing that back to PPC, it’s like, well, if we’re going to automate bidding, that’s not just a button that you push. It’s like now you’ve got to choose between, I don’t know at this point, like the nine types of automated bidding that Google offers. And if you choose one specific one like target ROAS, now they’re introducing seasonality bid adjustments and introducing conversion value rules, and there are ways to track it. Understanding all these interplays and what’s the right solution, that’s kind of the doctor.
And the secondary part of the doctor as the bedside manner. You’re not always going to hit your targets you’re gonna have a bad quarter and understanding why that happened and what you will do about it and talking to your boss about that. That’s important because the machine learning system doesn’t explain why it missed its targets.
The pilot actually breaks down into two kinds of pilots. There’s the commercial pilot whose job it is to get everyone there safely, you know, usually not taking her jets out along the way. And so it’s an oversight role. It’s making sure that the systems that are monitoring the plane are working correctly and that those decisions are being made correctly by the auto-pilots. The average pilot flies the plane like seven minutes out of a flight — one crazy little stat that’s out there. And then there is a fighter pilot. So as you see competitors maybe using automations and you figure out what the shortcomings are — so maybe that psychic or maternity ward is not looking at the lunar cycle, and you do, well that’s a competitive advantage. So figuring out, “Oh, they’re using that technology platform, and that one doesn’t take them into account.” So how do we pick a better tool or put in more information to achieve that?
So I think those are the two most common roles: the doctor and the pilot that PPC people today will start to play. The third one is a little bit more aspirational, I suppose, for most people, but it’s how do you teach a machine. Because the machine learning system doesn’t just magically exist. Someone has to build it. And that’s pretty valuable work. If you know how to build that system just a little bit better. When I was at Google, quality score, we make one percent improvements to it every now and then. And that’s one percent, who cares, right? But if you look at the billions of dollars they’re making, do one percent on top of that, yeah, that’s serious money. So there’s a lot of money in this space if you can figure out machine learning system that ekes out another percent. That’s a big deal.
So one of the things is we’ve talked about this over the last day or so here at SMX Advanced is machine learning wasn’t that great when it started. And PPC marketers who’ve been in this for any number of years have been trained to be really in the weeds and control, control, control. There’s always this tension between automation and control.
So you’ve got people that have been trained to be really hands-on, and you have people that have been seeing poor results. They’ve been burned. They feel like there’s this guinea pig syndrome where they’re taking the brunt to help Google and Microsoft Advertising algorithms learn. At the same time, and your book makes it very clear and I completely buy into this, there’s no going back on this. There’s only going to be more automation. Resistance is futile. So how do you convince people that things are fundamentally changing and if they don’t change the way that they’re working every day that they’re going to get left behind? How do you make this argument?
Well I mean there are so many levels to that question, but I think kind of the core of it is thinking about the acceleration of machine learning. That’s one part of it. So every roughly 18 months, Moore’s Law says that computing power doubles, and artificial intelligence has been around for more than 50 years. So why is it a big deal today as opposed to when PPC started? And it’s because we’re actually now in that phase where the technology is doubled about 27 times and every time it doubles it’s like it’s just massively better than what it was in the last cycle.
So if it’s not good enough today, you bet that is going to be a hell of a lot better very soon. Is it always going to be better than humans? No, and that’s one of the arguments in the book is that there are certain things that humans can really help out with.
And so very specifically, if you’re at an agency and you work in a specific industry, you probably have insights about what factors you should look at — things that maybe the machine learning systems shouldn’t be experimenting with — and so responsive search ads are a really good example. It’s a machine learning system that figures out how to put together the ad components for each and every query. But it’s still up to us humans to say these are the headlines that you get to test. The machine is not writing those. And so you really have to make sure you’re giving a machine good information good opportunity to learn. And I think you made the example recently about flying with a toddler and it being a horrible experience in the beginning and never wanting to fly again. But then you realize, well hey, if I teach the toddler how to be a good flier, it’s going to be awesome, and maybe they’re actually going to carry my bags and be helpful. Machine learning is the same thing. Learning is part of the term. So it’s not going to be perfect on day one. But it’s certainly not going to get stupider.
Right. Although, talk about some of the mistakes that people make. This whole idea of training and learning — when you log in, you see the “learning” message, or you see something that’s not working. In the previous iteration of PPC world, you would think, I need to make an adjustment and go in, and you’d make two or three adjustments. Does that work anymore?
Yeah, I mean you just have to take a step back at that point. The point in the book is that automation researchers have found that humans, when they see the machine not doing what it’s expected to do, they are very quick to say, “OK it’s not working. We’re not going to use it anymore.” Whereas if you hired a new employee into your agency and that employee made a mistake, you’d probably sit them down and explained why that was a mistake and try to teach them better and give them more insight, more inputs, and trust that they would get better next time. And so I think we have to do the same thing with the machine. If we see that the machine is not learning fast enough, we have to step back and say, “Do we have enough conversions that we’re tracking? Are we measuring conversions correctly?” Garbage in garbage out basically. A machine learning system cannot do a good job if you’re not giving it the right data to make its decisions on.
Attribution models kind of play into this as well. If you’re doing last click attribution, in the old days, you’d look at your keyword list, and you’d see, Oh this keyword has very few conversions on last click. An automated system would say, Kill that keyword, bid it down, get rid of it. Whereas a human you would say, Oh that’s actually a pretty relevant keyword. Maybe it does matter high up in the funnel. You’d keep it, but the machine learning system doesn’t have that context. So if you’re not putting a better attribution model in place at the end, well then yeah, your automation that does the bids is going to be pretty horrible and actually kill your campaign. But it’s not because the machine learning is bad, it’s because you’ve given it bad information.
Well, thank you so much for this. Parting words for the PPC marketers out there who are anxious about what their roles, their careers, their businesses are going to look like?
I think this is an exciting time. And if you join PPC, I mean it’s never been a slow-moving field, there’s constant change here, and this is just another one of these. I think it’s actually nice because now nobody likes doing tedious, repetitive work, and that’s really what these automations are taking care of.
And we get to think more about strategy. We get to actually become marketers again. Think about audiences. Think about messaging. Think about the fun stuff of marketing. And the number crunching, which a lot of people think is actually fun as well, but more of that is going to be done by the machine.
And the analysis is what we can focus on.
Exactly, drawing the insights and how do you apply those insights to your next campaign.
But another point that I think sometimes Google misses in building machine learning is they tend to think that a thousand dollar experiment, that’s going to lead to a great result, but that’s a thousand dollars. To Google, that’s nothing. To a small business that’s like a thousand dollars during which they’re on the fence: Is this machine going to learn it or is it not going to work? That’s real money for them. And so if you are more of an expert in this arena and you know the strategies, you know the right place, the right starting point to give you the highest chance of success, that’s a leg up. And that’s something valuable that you can sell. So I think that this is full of opportunities for PPC, but it’s going to be different from what we’ve been doing for the past ten years.
I think that’s the thing. It’s a really fundamental shift and mindset shift with lots of upside and opportunity for first-movers.
I think so.
Opinions expressed in this article are those of the guest author and not necessarily Search Engine Land. Staff authors are listed here.