Artificial Intelligence, Machine Learning, and Paid Search…Oh My!
Podcast: Play in new window | Download
Really, AI has been around as long as computers have. But over time, AI and machine learning have been able to take over certain functions in more concise, error-proof, and productive ways than humans can.
We have benefitted from this in ways that encompass Netflix being able to recommend the kind of shows we like based on our viewing history, all the way to bidding on PPC ads on our behalf based on certain data it’s picked up from the various platforms.
Rich: He’s an online advertising expert and owner and CEO of FMB Media, a PPC advertising agency. He’s worked for Apple computer and Ziff Davis Publishing. His articles have been published in MediaPost, Search Engine Watch, Search Engine Land, and MarketingSherpa, among many others. He’s the author of Customers Now, on AdWords display advertising, and PPC Advertising In One Hour a Day. He’s been judged one of the top advertising experts on Twitter, so Agents of Change, please welcome David Szetela.
David: Thank you Rich, nice to be here.
Rich: I’m glad you’re here, I’ve got some questions for you.
David: I’ve got answers.
Rich: What drew you to paid search in the first place?
David: Well I had had a long career that had one common element, which was direct response to advertising. I had a couple jobs in the magazine industry where we lived and died by direct mail advertising. And the keys to direct mail advertising success are testing a list, testing a group of people that might be receptive to your offer, and also testing the actual mailing piece making changes to the coupon or the price of whatever you were selling. That always fascinated me, I actually have an educational background in Science. The combination of a background in Art and Science – the Art being the creative part, and the Science being analyzing data that came back to replicate the same circumstances in which you’re going to test – was fascinating to me.
Fast forward to about 2004 and I became aware of Pay-Per-Click advertising – which is online on the search results page mostly, but also expresses itself to display advertising – it had the same elements and characteristics as direct mail advertising in that the key to creating compelling ad copy similar to creating a compelling mailing piece was testing both the premise that if you offer advertisement to a group of people, that group will be receptive. And also testing the different ways you can word the advertising in the display ad and how it would affect the response rate. To sum it up, it was an early fascination with something that was not at all digital to the digital world.
Rich: It is fascinating how much we can measure our marketing and advertising activity once we take it online.
David: Oh yeah. Quickly and with great precision.
Rich: Today I know that we want to talk about AI and machine learning and how it impacts PPC, but let’s start with the real basics. David, how do you define “AI” and “machine learning”?
David: Well I also have a background in software development so this makes it easy for me to kind of grasp these concepts, but let me simplify them. Artificial intelligence is really software that performs tasks that normally require human intelligence. A perfect example is speech recognition. We’ve all had sometimes maddening experience with telephone voice response trees. Fortunately the speech recognition is getting better and better. So that’s an example of software that basically mimics what a human can do, and does its job more efficiently and sometimes more accurately.
Machine learning is actually a type of artificial intelligence where the software not only performs a task that would normally be done by a human, but modifies itself over time based on the experience it’s having with interacting with humans. A good example of this is what Netflix calls the “recommendation engine”, where they build up a profile of a customer and it watches the customer choose videos, rate them, and then over time it becomes better at really knowing a customer and knowing types of videos they should suggest a customer to watch next.
Rich: So I feel like there’s this Venn diagram of AI and machine learning with a lot of overlap here. If I understand what you’re saying, AI is more about having a computer do tasks that normally would take a person to do. But like speech recognition, it tends to get batter over time. Machine learning is more specifically about machines getting smarter based on our previous activity. Pretty close?
Rich: Ok. So there’s definitely overlap here and one feeds into the other.
David: That’s correct.
Rich: Ok. This is a stupid question. I know there are no stupid questions. Is artificial intelligence or machine learning, when we talk about it in terms of PPC, is it something that takes place on our end or is it taking place on the platform’s end?
David: It’s hard to distinguish between the two. When you say “our end”, I assume you mean the advertiser’s side.
David: It’s really the advertiser. Picture it as a big machine that the advertiser is controlling & turning knobs, and the advertiser sets up the machine to which they think will produce the best results, then pushes a button and the machine just chugs away and does its thing. The thing in this case is applying artificial machine learning to the tasks that need to take place to display the advertisements, charge the advertisers the right amount of money, and learn over time better and better ways to get more customers for that advertiser at a lower cost.
Rich: So this sounds pretty cutting edge, how new is all this?
David: Well think of it this way – broadening it out a little bit – computers have been around for going on 100 years now, and they have been growing in power in a couple of different ways. One is the actual core of the computer, the CPU, has become faster and the amount of memory the computer has to work and move it’s information around is getting bigger. So as that is happening, the complexity of the data that can be acted on and the speed the data can be shuffled around has increased exponentially a couple of times.
Artificial intelligence has been around as long as there has been computer software. In the PPC world it’s always been present in the platform in some kind of small way, but it’s only recently been applied in much more complicated ways aided by the capabilities of the personal computer and the computing power behind the platforms.
Rich: Alright, so the artificial intelligence and machine learning are just able to do a lot more because of all this speed. When I think about PPC I think about AdWords, Bing Ads, I think about Facebook ads, Twitter ads, Pinterest. Which platforms out there are using this level of machine learning in their ad platforms?
David: Definitely AdWords, Microsoft Bing ads, Facebook. To a lesser extent, LinkedIn, although LinkedIn was purchased by Microsoft and watched that space for some similar complicated AI and machine learning in the future. So it’s the big platforms, AdWords, Bing ads, Facebook, LinkedIn, Twitter, and Instagram and Pinterest kind of bringing up the end there.
Rich: So obviously you kind of referenced this before, there’s different elements when it comes to doing PPC. There’s the creative, there’s copy, there’s images, there’s bid management. What elements of these platforms are using AI?
David: Well think of it this way, the aspects that you just mentioned vary in the amount of time it takes a human to do their job. Let’s take bid management for example, one of the key tasks that an advertiser needs to undertake in advertising – let’s say AdWords – is to tell AdWords this is how much I want to spend on my ad. And without getting into too much detail, the advertiser needs to specify how much they want to spend on a keyword by keyword basis. Meaning, within the AdWords account there are 100’s, could be 1,000’s, could be tens of thousands, of keywords that the advertising is saying to Google show my ad when a search query contains these keywords.
One of the tasks that is time consuming and error-prone is setting the bid prices, telling Google how much it makes sense to pay for a click depending on the return on investment that the advertiser is after. So it turns out that that is one of the things that computers can do much better and much faster and much more accurately than humans can. It’s naturally been one of the functions that machine learning and AI has taken over to the benefit of the advertiser.
Rich: So it’s certainly a time-saver, it sounds like. It might even be money saving. I can’t say that I’m usually the guy in our company that goes in and does all the PPC, but from the times I’ve been in there, there seems to be you can bid manually or you can turn on automated bids. Should we start automated bids right away? It just sounds like, why wouldn’t I right away turn on that autopilot, so to speak.
David: By “right away”, I assume you mean immediately upon creating a new ad account or a new ad campaign.
David: My advice is that the automated management algorithms work best when they have some data to work on. In other words, you can’t really automate a process without having a certain number of conversions actually happen to base the future predictions about whether someone doing a search, for example, is more or less likely to convert.
So my advice is to wait until approximately 15-30 conversions have happened within a 30 day period before turning on the autopilot and saying Google go ahead and control my bids from now on.
Rich: Ok, so if we’re not turning on automated bids and we’re basically just doing this ourselves, or we’re setting up a wider net & then saying from that wider net Google I now want you to get a little bit smarter than me about how I’m managing these bids?
David: Right, yup.
Rich: I’ve also heard some people – and this is more on the Facebook side of things – I don’t know you agree with, a lot of times people say that in the beginning of a Facebook ad campaign, do not be looking for conversions just be looking for traffic. Because Facebook in its infinite wisdom will often find the best people for you, but again, they need a larger sample size. So that’s why some people start an ad campaign based on website traffic. And after they get a certain number of conversions – meaning just traffic to the website – then they switch it over to something more like conversions.
David: Well, what you just said, everything you just said actually applies more to AdWords than to Facebook. And that is because AdWords allows the user to kind of manually control the bidding machine, manually set the bid process per click. With Facebook the only option really for regulating bidding is to pay Facebook on a cost per impression basis. Meaning that for every impression Facebook might charge $.50-$3.00, whereas their second possible bidding option is to turn the automated bidding to Facebook.
So I guess I would agree with you with one caveat, and that I sit’s not manually bidding the same way that Google puts it. It’s really manual bidding based on the number of impressions and the number of clicks to the site.
Rich: That makes sense. Absolutely. Now Dave, you’re obviously a very smart guy, there’s no two ways about this. And I know that you love this AI machine learning side of things. But how much of this should we know so that we can do our job better?
David: Well I thinks it’s best to understand the options. For example, AdWords provides about 7 different flavors of automated bid management. It isn’t difficult at all to understand what each flavor does. It’s really totally unnecessary to understand in the background of how the sausage is made. So I would encourage AdWords advertisers – and this goes for Facebook as well – to become familiar with what the options are and what the effects of those options are, but not get bogged down in the miscellaneous details behind the algorithms.
Rich: Know how to drive the car, not necessarily how the motor works.
David: Exactly. Yes.
Rich: So one of the things I’ve heard about, and I’m not sure if this fits in, but I‘ve heard about “dynamic creative”. Does that fit into this category, and if so, can you explain that to us?
David: Yeah, I would classify Facebook’s Dynamic Creative as artificial intelligence. Basically you tell Facebook I want you to take the following three images, the following two headlines, the following three description lines, and I want you to mix them all up and rotate every possible combination of these in ads that will be shown to my target audience. And at the end of this process once the data has accumulated, run the ad that is the best combination of elements that I’ve fed into this machine.
So it’s a very powerful capability that ensures that the best possible ad will run and does what could be manual procedures automatically. There’s one kind of “gotcha” in the whole thing though, and that is Facebook doesn’t really show you the winning ad. They’ll run the right combination, but for some reason they don’t show what elements actually made up the winning ad.
Rich: When I was a kid I used to have this toy which had different faces, different torsos, and then different legs, and you’d rotate them around and all that sorts of stuff. It’s almost on some sort of level like that on Facebook. I still have to create all the heads, torsos, and legs, but then Facebook is going to find the best, most engaging, most clickable combination. Is that correct?
David: Yeah, that’s a great analogy.
Rich: Ok. This is interesting stuff. If you had a crystal ball, where do you think all of this AI/machine learning is going to take us in the next couple years?
David: Google has not been shy in describing where they want it to go. Basically they want advertisers to walk in the front door of Google AdWords and say, “here’s my product or service, here’s some dynamics like the profit margin I want to make, Google you just do what you need to do to make sure that my messaging reaches the right people and that the amount that I spent brings in a return on that investment that’s acceptable to me.”
That portends things like automatic ad copywriting, which Google does to some extent right now. It portends things like automatically finding the right audience for a product or service. And again, AdWords does have primitive capabilities that are kind of pointing in that direction right now.
So again it boils down to Google doesn’t want – and this goes for Microsoft and Facebook as well – they don’t want advertisers or site owners to become computer scientists. They don’t want site owners to become even technical advertisers. They want them to just know their business, know what they’re selling, and to some extent know who their audience is and how to find it, and then turn over all of the decisions of when and how to articulate the unique selling propositions to the platform.
Rich: If you were listening to this interview and you maybe had played around with some of these ad platforms but didn’t really understand the AI and machine learning behind it, what would you say the big takeaway should be for that listener after today’s chat?
David: Well unfortunately the platforms today are pretty complex and not very easy to learn. I would suggest going to the educational resources that are available from the platforms – as well as third parties – and kind of just diving in and learning what Google will describe as the fundamentals. A fairly literate and computer savvy person can actually get through the fundamentals within a day. And Google provides free testing to make sure that the lessons have been learned. So that’s really the place that I would start.
And then I’d get my hands dirty in actually conducting campaigns. That’s actually how I got started. I had a lot of people – who are now experts in the field – that started by advertising their own products and services, or affiliate products, just pushing the levers and turning the knobs and getting better and better at steering the ship.
Rich: That’s awesome. I know that you’ve written a lot of articles, I know that people are going to want to try and understand more about it. If they’re looking to learn more from you, where can we send them?
David: I did write a couple of books on the topic, one of them I would be happy to offer to your listeners as a free PDF copy. It’s a little bit outdated in that it describes user interfaces that have changed over time, but the fundamental lessons about ad copy writing, landing page design, how to determine how much to bid, those kinds of things are rock solid and evergreen. So they can get that free book by sending me an email at firstname.lastname@example.org. Just put the word “book” in the title and I’ll send you off a free copy.
Other than that I would suggest that there are a couple of other really good books on the market that kind of compliment mine. One of them is written by Andrew Goodman, another one is written by a brilliant guy named Brad Geddes. And really just, I’m afraid I wish it were easier, but just plowing through the material and learning the craft is the best way to succeed in this.
Rich: David this has been great, thank you very much for your time and your expertise today.
David: My pleasure, Rich.
David Szetela is an expert in PPC advertising, with a particular passion for AI and machine learning. Shoot him a quick email (email@example.com) and he’ll send you a free copy of his book on the subject.
Rich Brooks is the President of flyte new media, a web design & digital marketing agency in Portland, Maine. He knows a thing or two about helping businesses grow by reaching their ideal customers, and to prove that, he puts on a yearly conference to inspire small businesses to achieve big success. You can also head on over to Twitter to check him out, and he just added “author” to his resume with his brand new book!