We all know that automation is radically changing our approach to CPD. From configuration to campaign management and optimization, our roles evolve. Frederick Vallaeys, former Googler and co-founder of the campaign optimization platform Optmyzr, explains in his new book "Digital Marketing in an AI World" why the understanding of CPP's core principles is more important than ever, what machine learning can (and can not) and the skills and abilities marketers will need
I met Vallaeys at Seattle's SMX Advanced last week to discuss We discussed his time as a member of the AdWords team and as one of his major advertisers, and why he is optimistic about our various roles as digital marketers in the era of the automation Listen or read the interview below.
This interview with Frederick Vallaeys, CEO of Optmyzr, was légèrem Edited and condensed.
Let's talk about it at a very basic level: what are we talking about when we talk about AI, when we talk about machine learning and when we talk about automation and what marketers really need to understand the terms means that we talk a lot about it.
At a very high level you need to understand where technology is going and how quickly it is evolving. Basically, it is a very real change that will cost a lot, but it is not enough that many people think
So, if you look at traditional movies or media, AI is like those humanoids that do pretty much everything, and we are not where we are. . In PPC, this is mainly machine learning, targeting very specific problems such as: "How to define the right offer if we have a ROAS goal?", Because we have to make an offer to the CPC, because that is how that the Google auction is working properly?
So, automation is very specific to this. But also understand, I think, at a certain level the history of it. Artificial intelligence is therefore not new in Google Ads. So the level of quality is like we were over ten years old. This was the original machine learning system: determining the expected clickthrough rate on a combination of keywords and ads given for a given query.
What you helped to create.
Yes, I was in this team. So it was a fun part of the book too - share this idea and tell stories of, you know, play hockey with the founders and cross Sergey [Bring, Google co-founder] while having the chance not to get fired that day.
Tell a little about your history at Google and some of its strengths.
First of all, when I joined Google, I had put a fairly short list of companies I wanted to work in, and Google was setting up itself, but it was not a big deal yet. That's why I could play hockey with Sergey, and I ... checked him because I did not really know who he was at the time. Then he became really famous.
The acquisition of Urchin. Go to San Diego and meet this small team, which eventually became Google Analytics. And that's one of the most fundamental aspects of online marketing nowadays, right? What would we do if we did not have these numbers?
Some other highlights: So I started advertising while I was working at Google because I saw the light. I was like, "Oh my God, people are spending $ 30,000 a month - and that made you a Level 1 advertiser, it was like the biggest fatter.
You were able to be an advertiser, who explained how the products were developed.
Because there is such a gap between the people who build these tools and those who use them tools, and you have helped make those connections.
Well, and you see it even here. I'm at SMX, and there's a Google speech. This is an excellent vision, but there is a gap between this vision and how to achieve it. And that was my role in Google. I started to advertise and I became a pretty big advertiser and affiliate advertiser. I do not know what keywords are converted. And guess what? Conversion tracking did not exist at the time.
We all had the following concept: yes, the PPC works very well, but in reality we still did not know what really worked. We were just spending money. And so, I built my own little conversion tracking system. Then the team took note and said, "It's pretty cool. Maybe all advertisers should have that. Can you go talk to the product team? "It was one of my first products.
So you say in the book, thinking of the skills we must have, that many of us do not need to understand some technical details, but you also need to understand what "smart" systems do, which in my opinion was really important. What are the factors that PPC marketers, who are not technical, need to understand the factors and ways in which these tools and systems work to make better decisions and ensure that machines do what they are
And this is one of the great premises and, I think. One of the most important tasks of PPC marketers is to help the machine learn by providing more data about your business. And sometimes it's scary, right? You talk to many companies who say, "We do not want to give more data to Google because you're going to do something wrong with that." I worked at Google. This is not a bad business. I do not think that has changed since I left in this regard. And finally, one of the guarantees is that it's still an auction, right? So, you know, you give additional information, and all that Google actually does, that is, it tries to increase your bids when it thinks that the conversion chances are greater, and vice versa .
But now, I think. the fundamentals or fundamentals of PPC are always so important to understand. I talk to people, and they use thoes auctions. I ask them, "How to move from a target ROAS to a real CPC based on your conversion rate and your values?". And they can not do the math. And if you can not ... I do not expect you to do it on the spot, but you should be able to think about it and explain to yourself somehow, and then say, "Well if you use the historical conversion value per conversion or per click, what kind of data do we have that could possibly inform some of the better or weaker audiences, and what can we do about it? Should we supplement Google's knowledge by creating a new audience that looks at something we care about? "
Because it's one of the fundamentals of machine learning, it uses historical data to predict the future. to think your data in a new way.And one of the things you are talking about is to inform these algorithms with more data, and that there are gaps that Google has. when teams talk to each other, OK, we want to use smart bidding. Maybe that's not even Google's smart offer, it's a smart offer, it's a third-party tool, and you realize, OK, these are the data we have, and that's our goal, can that help us achieve our goal?
So, what are the steps to think through this process and think about "What are the gaps?" How can anyone come back to the office? And di re: "It's our challenge. How can I do it myself or work with teams because I do not have access to this data at the moment?
That's right, and it's really one of the most important aspects of PPC experts. will play a role, and I do not think it's a simple answer, but it's a little bit why we will have jobs in the future. It's just that our jobs have been redefined because of this new issue. This poses several problems. One of them is Google, but it does not really tell you what data it has and what data it uses. So, an example that I gave was for the quality score. At one point, we said, "Hey, we should examine the lunar cycle and determine if it has an impact on the predicted click rate." And we found that the whole system was not - so we decided not to use it's a factor. Now, if you are a tarot card reader or a psychic, you know, maybe that really matters.
I know, when I read that I thought it was also a maternity ward. So many ... deliveries take place in full moon. Any maternity nurse will say that the full moon occurs when we are full. So I thought, oh, that's a great example of what lunar cycles may not have been, but there are things like that.
There are specific sectors and sectors. And then, if it matters to you, that's when you need to determine how we can inform Google about it.
And then the other question that is really difficult and one that we will have to play on is also that the motherhood says, "Yeah, we're full." But really, how full are you? % more complete than "unusual" And if you have reached your maximum capacity, does that actually mean that you want to raise the bid or bid less?
It becomes a strategic decision and you probably need to unplug your computer - the data in your own machine learning system - and if you do not have enough data you can still do a statistical analysis on it - but now you have points You can say, "OK for the full moon, we're going to increase our bids by 20%." So raising a bid in the past meant "increasing your CPC by 20%", but now "increasing your target CPA by many or lower your ROAS from another figure 'because you know that your conversion rate will reach a peak. And even if you're supposed to pay more for an acquisition, the added conversion rate will offset that loss. And again, it comes back to fundamentals. If you do not understand how the conversion rate and CPA goals interact in this way, you simply will not be able to make the right decisions.
I'm thinking of the newborn baby, you know. photographer ... who says, I know I'll be very busy right now. What are some of the resources and best ways to start making sure you know what you are talking about and know how to plan?
Read the book. [Laughs] The book has three sections. The introductory section discusses what is this technology. And I think that understanding the technology that guides it helps you position yourself, but also helps you think: "How can I harness this technology similar to my benefit?"
And a concept that I'm In SMX Advanced, there was talk of overlaying, and that's huge, so there are built-in automations in the engine, like smart bidding, and they do some things very well, but they do not do it So how do you overlay your own system to change goals or take action when you find that an outside factor influences it? And I think that just by understanding the basic technology, the Basic machine learning, you will be better able to develop your own solutions.
And I loved your analogies with the doctor, the pilot and the roles of teaching You know, that's one of the things we're talking about - what will our role be like and will we have jobs? Can you talk about the role of the doctor, the pilot and the teacher, and how they apply?
I really wanted to simply simplify things and give an analogy that people can easily understand. So when you think of a doctor, you consult a doctor who is suffering from an illness. A doctor knows that 17 different medications may be effective, and that some of these more aggressive medications have more serious side effects. And so, they look a little at the patient and say, "Well, how serious is this problem?" Then they prescribe what they think is the right level of medication. And they know the interactions. They know of other conditions that you have. So, what is the interaction? To return to PPC, it's as if, if we were going to automate bidding, it was not just a button you press. It's like now that you have to choose between, I do not know at the moment, like the nine types of automated auction offered by Google. And if you choose a specific one, such as the target ROAS, they now introduce seasonality bid adjustments and conversion value rules, and there are ways to track them. Understanding all these interactions and what is the right solution, it's a little doctor.
And the secondary part of the doctor as bedside. You will not always reach your targets, you will have a bad quarter and understand why it happened and what you will do about it and talk to your boss about it. This is important because the machine learning system does not explain why it missed its goals.
The pilot breaks down into two types of pilots. There is the professional pilot whose job is to get everyone to their destination safely, you know, usually not with their jets. And so it's a supervisory role. It ensures that the systems that monitor the aircraft are working properly and that these decisions are made correctly by the autopilots. The average pilot pilots the plane seven minutes away from a flight, which is a little crazy statistics. And then there is a fighter pilot. So when you see competitors using automatisms and you identify gaps, the psychiatric service or maternity may not be looking at the lunar cycle, and you do it, it's a competitive advantage. So, how "Oh, they use this technology platform, and that one does not take them into account." So, how can we choose a better tool or put more information to achieve it?
So, I think these are the two most common roles: the doctor and the pilot that the CPP people will start playing. The third is a bit more ambitious, I suppose, for most people, but that's how you teach a machine. Because the machine learning system does not exist as if by magic. Someone has to build it. And it is a very valuable job. If you know how to build this system a little better. When I worked at Google, we made 1% improvements from time to time. And that's one percent, who cares about it, right? But if you look at the billions of dollars they make, make one percent more, yes, it's serious money. So there is a lot of money in this space if you can understand a machine learning system that generates another percentage. It's a big problem.
One of the things is that we talked about it about a day ago at SMX Advanced: machine learning was not so good at first. And PPC marketers who have been in this field for several years have been trained in weed control and control, control, control. There is always this tension between automation and control.
So you have people who have been trained to be really pragmatic and others who have achieved poor results. They were burned. They feel that guinea pig syndrome is trying to help the Google and Microsoft Advertising algorithms to learn. At the same time, and your book says it very clearly and I totally agree with that, there is no turning back. There will only be more automation. Resistance is futile. So how do you convince people that things are fundamentally changing and that they are not changing the way they work every day, that they are going to be left behind? How do you make this argument?
Well, I mean, there are so many levels to this question, but I think that a little bit at the heart of this question is to think about the acceleration of machine learning. It's a part of it. So, every 18 months or so, Moore's law says that computing power doubles and that artificial intelligence has been around for more than 50 years. So why is this a big problem today versus the beginning of the CPP? And that's because we are currently in this phase where technology is doubled about 27 times and whenever it doubles, it's like it's just better than what it's was in the last cycle.
So, if it's not good enough today, you bet it'll be much better soon. Will it always be better than humans? No, and one of the arguments put forward in the book is that there are some things that humans can really help.
And precisely, if you are in an agency and you work in a specific sector. you probably have an idea of the factors to consider - things that machine learning systems might not have to experiment with - and responsive search ads are a good example. It's a machine learning system that explains how to assemble the advertising components for each query. But it's still up to us humans to say that these are the titles you have to test. The machine does not write them. And you have to make sure that you give a machine good information, a good opportunity to learn. And I think you recently gave the example of flying with a toddler and that it was a horrible experience at first and you never wanted to fly again. But then, you realize, well, if I taught the toddler to be a good thief, that would be great, and maybe he would carry my bags and be helpful. Machine learning is the same thing. Learning is part of the term. So, it will not be perfect the first day. But this is certainly not going to be stupid.
Good. Although, talk about some of the mistakes that people make. All this idea of training and learning - when you log in, you see the message "learn" or you see something that does not work. In the previous iteration of the PPC world, you might have thought that I needed to make an adjustment, then two or three adjustments. Does it work anymore?
Yeah, I mean you just have to step back at this point. The importance of this book is that researchers in automation have discovered that humans, when they see the machine not doing what it is supposed to do, are very quick to say, "OK, that does not work. We will not use it anymore. While if you hired a new employee in your agency and that employee made a mistake, you would probably explain them and explain to them why it was a mistake and try to teach them better and give more insight, more inputs and confidence that they would be better next time. And so I think we have to do the same thing with the machine. If we find that the machine is not learning fast enough, we need to step back and say, "Do we have enough conversions that we follow? Are we measuring conversions correctly? A machine learning system can not do a good job if you do not give it the right data to make decisions on.
Attribution models also play such a role. If you perform last-click attribution, you would review your keyword list, and you'll see that this keyword contains very few conversions at the last click. An automated system would say: Kill this keyword, bid, finish. While a human, you would say, this is a very relevant keyword. Maybe it's important in the funnel. You would keep it, but the machine learning system does not have that context. So, if you do not put a better attribution model in place at the end, well yes, your auction automation will be very horrible and will actually kill your campaign. But it's not because machine learning is bad, it's because you gave it bad information.
Well, thank you very much for that. A word for PPC marketers who are worried about their role, their career and their business?
I think we are living an exciting time. And if you join PPC, I mean the domain has never been so slow, there are constant changes here, and that's another example. I think it's really good because now nobody likes to do tedious and repetitive work, and that's really what these automatisms do.
And we think more about strategy. We can become traders again. Think of the public. Think about messaging. Think about the fun things of marketing. And the calculation of numbers, which many people think is actually fun, but the machine will do more. 19459003 19459002 And analysis is what we can focus on.
Exactly, draw the ideas and how do you apply these ideas to your next campaign.
But another point that, I think, sometimes misses Google in the construction of machine learning, is that they tend to think that a thousand dollar experience it will give a good result, but it's a thousand dollars. For Google, it's nothing. For a small business of about a thousand dollars during which she is on the fence: will this machine learn it or will it not work? It's real money for them. And so, if you are more expert in this area and you know the strategies, you know the right place, the right starting point to give you the best chance of success, it's an advantage. And it's something precious that you can sell. So, I think this is full of opportunities for PPC, but it will be different from what we have been doing for ten years.
I think that's the problem. It is truly a fundamental change and a change of mind with many benefits and opportunities for the former.
I think so.
About the Author
Ginny Marvin, editor-in-chief of Third Door Media, manages the daily editorial operations of all our publications. Ginny writes on paid online marketing topics, including paid search, paid social networks, targeted posting and retargeting for Search Engine Land, Marketing Land and MarTech Today. With over 15 years of marketing experience, she has held senior management positions in both in-house and agency management. It can be found on Twitter under the name @ginnymarvin.