On this episode of the Paid Search Magic podcast, we’re talking to Will Critchlow about some really fascinating advancements in the field of machine learning, and how paid search marketers can benefit from understanding these changes.

Will is CEO of Distilled – a company he founded in 2005 with Duncan Morris. Distilled provides online marketing services from offices in London, New York and Seattle, hosts the SearchLove conference series in the US and UK and produces the popular online training platform DistilledU. Will has spoken at conferences around the world, consulted with the largest websites out there, and is widely regarded as a leading expert on all things search.

[highlight]Get $150 off the upcoming SearchLove Conference in Boston (May 3 & 4, 2016) [/highlight]

Take-Aways and Insights

• Machine learning is the process by which computers can set out to understand the world around them, and Google is at the forefront.
• Machine learning lets Google use implicit signals to better understand what a searcher is looking for.
• Google is fundamentally a search advertising company, and needs to give marketers a way to control their bids even as Google relies more heavily on implied intent.
• The term “mobile” is problematic because it conflates device and context.
• Companies such as BuzzFeed are on the cutting edge of understanding user behavior.
• SearchLove conferences offer a single-track, shared experience with expert speakers from in and around the industry. ## ticket.
• Distilled’s new SEO split testing tool solves some of the main challenges in testing and deploying changes.

 

Detailed Show Notes

ELI5: What is Machine Learning?

Amy: Let’s get started with a definition. Explain like I’m 5: What is machine learning?

Will: Machine learning is the process by which computers can set out to understand the world around them. It draws on the way we understand that humans learn. Most computer programs – we write very explicitly, we tell a computer exactly what to do. With machine learning, we write programs that are designed to take in observations and things of the world around them, and turn that into decisions.

It’s a branch of artificial intelligence, which is the discipline that looks at trying to create computers that can think. We may well be some way off from what’s called General Artificial Intelligence, which is the idea of a computer that can think like a person can, or better. But there’s certainly applications already of machine learning, from simple applications to the breakthroughs we’ve seen in popular culture:

• Improvements in the ability to understand what’s in pictures, if you do an Image Search or Google photos – “my photos of mountains”. It’s machine learning behind the scenes that enables the computer to figure out which of your pictures have mountains. That’s come along in leaps and bounds in recent years.
• Things like voice recognition.
• The recent victory of Alphago – Google’s DeepMind acquisition – playing and beating some of the best players  in the world at the board game “Go” which was thought to be a very hard problem for computers to solve. 

It’s this whole branch of computer science, essentially, but it applies to marketing as well.

Why Understanding Machine Learning Matters

Amy: How would you say machine learning applies to marketing, and how could marketers leverage their knowledge of machine learning to be in a stronger position?

Will: I tend to approach this from a couple of different directions.

When it comes to search, Google is at the forefront of machine learning. A lot of the academic breakthroughs, and certainly  the breakthroughs in putting it into practice, are coming out of Google.

I think it applies to search marketers in particular – we need to understand this on a couple of levels.

Probably the simplest thing we need to understand is simply that it’s happening. Previously we had a mental model for how search worked, and we understood that the engineers at Google were creating these very deterministic algorithms, where they got to tweak every ranking factor and every element of the auction when it comes to paid search. What we should realize is that’s becoming less true, or at least, that in addition to those things, they’re using machine learning algorithms  to enhance the user experience.

[quote style=”boxed”]Probably the simplest thing we need to understand is that it’s happening.[/quote]

As a marketer, even before we try to understand anything about how machine learning works, it’s important to realize that they’re using these very powerful techniques – techniques that let them understand what’s in a photo, or understand spoken word – they’re using these same techniques to analyze what we type in as a search query, and to understand the websites and even the adverts that are returned. Step 1 is understanding, that is a thing. A lot of the edge cases or things we’ve become used to as best practices – we should realize that those aren’t necessarily sacred. If the computer figures out a better way to make users happy, those things will dominate.

At a zoomed out strategic level, that’s probably the level that marketers need to understand. I find it useful to try to understand the nuts and bolts. I’ve always found when it comes to technical marketing – which covers everything we do  – it’s useful to understand these systems are actually working behind the scenes. The better mental model is, the better you can market against that and reach the audience you’re trying to reach.

A lot of what I’ve talked about over the years has been in organic search, but actually I was originally fascinated most by paid search, in large part because of the auction model. My academic background was in Maths and Statistics and I studied Auction Theory, the very, very specialized little bit of Operations Research and Statistics that I ended up specializing in. The AdWords auction is probably the biggest, most pervasive auction in the world.

[quote style=”boxed”]The AdWords auction is probably the biggest, most pervasive auction in the world.[/quote]

 I personally find it useful to understand some of those nuts and bolts behind the mechanics of it. When it comes to machine learning, just trying to understand what computers are good at, and what computers are not yet good at, gives us some indication of where this technology’s going.

How will Google handle implied intent for PPC?

Amy: Let’s talk about the implications of machine learning. Google’s moving more towards implicit signals and implied intent. They’re inferring different things about us that we’re not typing into the search box, and using that information to determine the content we see. As that happens, people are going to be less precise with their search queries.

As a paid search marketer, it’s a lot easier to bid on precise queries. If someone types into Google “I want to buy tickets for The Beatles Love show in Las Vegas,” and I’m Cirque du Soleil,* that’s a much safer keyword to bid on than someone who typed “Love Beatles.” Maybe Google knows what “Love Beatles” means, if someone just booked flights to Vegas and they’re staying at the Mirage. But that’s really too broad for me to know, and I don’t have access to those signals.

[typography font=”Cantarell” size=”14″ size_format=”px” color=”222222″]*note: I was using YouTube’s automatic closed captioning service to assist in this transcription, and it actually correctly interpreted the spoken phrase “Cirque du Soleil.” This is especially noteworthy because it’s one of the only words it got right.[/typography]

Do you have any thoughts on how paid search marketers can compete, or are we just losing some of our ability to compete when the searches get really imprecise and marginally relevant?

Will: It’s going to be very important that Google figures out how to solve this. For all of their power in other areas, they’re fundamentally a search advertising company, that’s where their revenue comes from. This is as big a threat and/or opportunity to Google’s business model as mobile, and we’ve seen how aggressively they’ve moved on that.

[quote style=”boxed”]For all of Google’s power in other areas, they’re fundamentally a search advertising company.[/quote]

I would expect them to move aggressively on this as soon as it becomes a real problem for their biggest advertisers. I’m kind of surprised we haven’t seen more of these signals available to bid against already.

The implicit side of what you can bid on is still pretty limited. There is some of it there. It’s not just pure keyword bidding, we can now do time of day, location, there are some of those signals. It is surprising that they don’t expose more of these things that they use on the organic side as factors that we can use in our bidding and auctions.

You’re absolutely right, those vague queries  – that Google understands because of all the context they have – generally make bad PPC bidding terms.

Thinking more about the organic side, I caught myself doing a search for “breakfast” in Boston one day. I was jetlagged, I literally typed in “breakfast” before I really thought about it. Google figured out exactly what I meant and gave me a local pack and highly reviewed, great breakfast places in the area.

A couple of years ago, I would have got the wikipedia page defining breakfast, maybe a page about “Breakfast at Tiffanys.” It wouldn’t have been a good user experience, and these days that would still be a poor PPC experience, I suspect. They need to figure out a way to solve this.

There are two possible ways they can do that, that I can think of.

1. They expose more of this in the console and let you bid explicitly on implicit factors.

You could say “only bid on these keywords for people whose search history includes other kinds of things,” or “modify my bid if the user has previously traveled to a certain place…” all the factors they’re going to factor in.

That may be a dead end. It probably gets super complicated really fast and destroys the auction model. You’ve got very sparse bidding. How would you even manage that? It sounds pretty complicated. 

2. Possibly more likely is we may get access to some of the post-processed queries.

If Google’s going to figure out that when you do a 1 or 2 word query, it’s equivalent to a much longer, more specific query, and they can bundle that in with implicit factors, then maybe the thing we get to bid against as paid search marketers is the more specific query, in other words the query after they’ve processed it.

The problem with that is it might lack transparency. You would then be saying “I only want to bid on ‘Beatles Love’.” We might be bidding on this very specific keyword, but it would be like extended broad match. Google would be showing our ad against much more broad keywords in cases where Google was confident that’s what the user was really searching for.

The problem there, of course, is as advertisers, that’s back to us putting our complete faith and trust in Google to figure out what’s in our best interest, even though it’s in their best interest to take our money. It’s a hard problem. I don’t know which direction they’ll go in to resolve this.

Fundamentally, it’s got to go in one of those 2 ways, either we get to control it and explicitly bid against implicit signals, or the other direction: we continue bidding the way we do now, and Google lays an intelligence layer on top of that: “When someone’s short query is equivalent to one of these long queries you’ve decided to bid on, I’m going to show the ad there as well.”

Amy: To some extent they do that, with enhanced CPC and conversion optimization. They don’t tell us the factors, they just say “trust in us.” As long as we see it working – I can get better results from doing that than trying to figure it out on my own, because I don’t have access to those signals – if it works best for me, I will do it even if I’d rather have more transparency. What are my other options?

Will: I have to suspect that’s the way they’re going for a couple of reasons. One is, there’s already some evident of them doing that. The other is, they seem very okay with pushing their opinionated view of the world. For example, the enhanced case. Rather than encouraging us to explicitly bid on different devices,  they say “we’ll figure this out.”

Amy: Then they’ll happily tell us that tablet and desktop perform the same, even though we can all see that’s not the case.

Will: Exactly, they’re kind of shaking the web around them.

Loss of the Right Rail: Standardizing the Search Experience

Amy: You mentioned Google has to find a way to continue to make money with advertising. Do you think moving to the 4 ads on the top of the SERPS is one step towards that? They can continue to control that space and have more ad revenue.

Will: I think it will probably have that effect. I’m not super cynical on this front. I think that change is largely driven by the need to operate “Mobile First.” Everything I’ve seen publicly and privately about the way Google is working right now, that’s a very real thing. They need to think first about the mobile experience and roll that out to the desktop. Standardizing that kind of experience is pretty important to them. I think that change is driven by needing to improve the mobile experience, and possibly to improve revenue on mobile, and standardizing between mobile and desktop.

The changes are so user-centric; they’ll have tested user happiness and user satisfaction. To the benefit of paid search marketers, for a lot of those commercial queries, adverts are a pretty good user experience. This often gets missed in the mix.

There are certain things that mean I as a user really like getting a paid search ad. When it means someone’s more likely to have something in stock, or they most likely serve my area. Things that are much more on paid search marketer’s radar than organic. With organic, each individual click doesn’t cost you incremental money, you don’t mind ranking for slightly less relevant things. Whereas with paid search, you want to be hyper targeted and hyper relevant.

[quote style=”boxed”]There are certain things that mean I as a user really like getting a paid search ad.[/quote]

I suspect this is something they tested. It’s good for users and good for revenue. I think it’s driven by mobile. They are definitely going to come under increasing pressure. They got away with it by blending their YouTube results with paid search in their financial reporting. They got away with obfuscating some of that. With the move to Google as a subsidiary of  Alphabet, we’ll probably see more transparency. Over the coming years, at some point, that’s going to put pressure on this. They’ll have to face up to the fact that the market expects them to continue to monetize mobile in the way they’ve monetized desktop in the past. 

Maybe that was a slightly hedged answer – I’m not being tin foil hat about it, I suspect it does increase revenue. The worst is yet to come. Prediction time: I think we’re going to see paid search ads mixed in with organic search on mobile.

[quote style=”boxed”]Prediction time: I think we’re going to see paid search ads mixed in with organic search on mobile.[/quote]

Amy: Mixed in – do you mean not having a tag that says “ad” on it?

Will: They’ll still have to label it, no I mean not just at the top and bottom. You’d get some organic, some paid. Some organic, some paid. Treating each individual result more like a card, and then just ordering them in whatever order makes most sense. I think they’ll at least test that.

On the labeling part, they’re treading a fine line. They’ve been told explicitly by regulators in the US, and I think in Europe as well, that it’s really important they label adverts clearly. The FTC is really big on this. As a user, I feel like that trend is in the other direction. It feels like ads are less clearly labeled than they have been in the past.

My suspicion is the average user does not know when they’re clicking on an ad and when they’re clicking on an organic result.

Amy: I have to agree. You said that people like ads. I think people think they don’t like ads. We have this tendency to say “I hate advertising. I never click on ads.” Obviously, the data suggests something else. The data suggests that it does provide a positive experience for people. We just like to think that advertising doesn’t work on us. Even if it’s in the form of a paid ad on a search engine, we’re above advertising somehow. It is interesting to see – if it is the best result – it almost becomes native in a way. It looks and feels like any other listing, it looks and feels like an answer to their question or a solution to their problem. If it could be mixed in there to give the best results, and people aren’t noticing it’s an ad, how will it perform? Will it be better for both the user and Google’s bottom line?

Will: If they do that, it will be a really big deal. It looks like paid inclusion. It hasn’t been a thing for quite some time. It was more common on other search engines, and Google came out pretty strongly and said “that shouldn’t be a thing, that’s not how search works, we’re never going to do that.” I feel like all the rules are re-written on mobile, and it could be legitimately something that is valid. If they were to do that, it would be a pretty big deal.

Mobile conflates device and context

Amy: Do you think marketers should be running different ads for different devices?

Will: Should there be different ads on different devices? Possibly. Is the best way to do that to have the advertiser explicitly choose? Probably not.

It’s so market-specific. In some markets, user behavior is very similar across all devices. There are other cases where the use case is clearly different.

I wish they’d give us more control over this. The trend is they’re going to say “let us handle that for you.” Again, I think we’re going to see more of a trend towards things that look like product listing ads and google shopping works at the moment, where you don’t even write the advert, you just pay for inclusion, essentially. I think we’re going to see more of that in regular paid search as well. Google already crawls our websites, already is capable of ranking them with titles and snippets and URLs in the organic search. That could be an advert, right? And in some cases, why not?

Especially at the smaller end of the PPC market – SMBs – I think it’s going to look more and more like that, where Google is the one writing the advert.

On the organic side, increasingly the title that shows up in the search results isn’t necessarily the title in the page. Google is doing a lot of smart stuff to rewrite those things already. It’s a relatively small step from there to actually just writing the advert for you.

Amy: That sounds a bit like the dynamic search ads that they’re pushing towards. “Give us your website and we’ll grab that, make the ad for you, take you to the right landing page…”

Will: Which in our experience, doesn’t work nearly as well. It’s a great way to spend a lot of money relatively ineffectively. Clearly a trend, definitely happening in organic search, every time we’ve seen hints of this in paid, it’s kind of been a bad thing. We kind of want to turn it all off and get manual control back. We tend to see that we can beat the algorithms.

Amy: Absolutely. I’d say there are some times I can put a campaign on Conversion Optimizer and it would do better because Google does have access to those implicit signals I don’t have. For the most part, especially when it comes to ad writing and knowing what’s going to convert – getting specific about the keywords – that’s something that does make sense to retain control over.

I’m interested  if I can get more specific with you about targeting devices – where we do have the opportunity to make ad extensions for mobile, make our ad for mobile, and adjust our bid for mobile. Where do you feel you don’t have as much control as you want there?

Will: We think of mobile as just one thing. Mobile / Tablet / Desktop. I think there’s so much more to it than that. It’s not just device. Sometimes it’s context and location. There are times where I’m “mobile” in the sense that I’m using my phone, but actually I’m at my desk or my couch at home. Conversely there’s times where I’m apparently at a desktop – as in a laptop – but really I’m tethered and I’m on a train. So what does mobile mean?

[quote style=”boxed”]There are times where I’m “mobile” in the sense that I’m using my phone, but actually I’m at my desk or my couch at home. Conversely there’s times where I’m on a laptop, but really I’m tethered and I’m on a train.[/quote]

I agree with you on the device side, certainly in the ad extension and the ability to do direct calling. That should clearly be tailored to the device. As a marketer, I tend to think “mobile” isn’t the most helpful definition, because it conflates device and context.

Talking about devices, even phones aren’t really that uniform. There’s a great diagram that shows the range of different screen sizes on the market. All these things overlap, that’s the crazy thing, and that’s only going to get more prevalent.

I’ve got the Nexus 6P which is the same size as the iPhone 6 Plus, and I only just realized it has a higher screen resolution than my tablet. I think all these things blur and overlap. Yes, we want different experiences when someone is mobile, or on a small screen versus a big screen, but all of those things have different gradations or granularity, and they all overlap.

Amy: There is a lot of overlap. It can be hard to try to break that out. I had a client recently – they wanted to spend a certain amount of money on mobile, as if that were a strategy. No, that’s layered into everything you’re doing, it’s really going to be hard to say “this money is dedicated for mobile” as opposed to all the other campaigns and stages of progression along the funnel. It’s hard to think about mobile sometimes, I think we’re not used to that.

Will: It’s counter-intuitive. We have a tendency sometimes to talk about mobile users. But nobody’s a mobile user. People use all kinds of devices. The same user will use different devices at different times, and certainly in different contexts. Google makes a bid deal of this – how common it is to research stuff on mobile devices before making a purchase, and honestly it’s in their interest to talk about this, but it’s a real thing.

[quote style=”boxed”]Nobody’s a mobile user.[/quote]

Amy: One thing I’ve seen a lot of marketers do when they’re not sure what to do with mobile, is they’ll try to infer their own intent. Like you were saying, mobile can mean a range of things – it’s not that there’s a mobile user. I’ve heard a lot of people advocating assumptions of intent. If someone’s on a desktop searching for a locksmith, that means they want to change their locks, but if someone’s on their mobile phone, it means they’re locked out of their house.

Is that a safe assumption? I don’t know. When it comes to choosing an ad to show someone, it’s really leading someone down a particular path they may not be meaning to go down. I could be on my phone trying to do pizza delivery or dine-in, and if an advertiser’s going to choose for me, and worse, drive me to a dedicated landing page that doesn’t give me the option to choose – that limits my choice because it’s trying to limit distraction- that becomes a  very poor user experience when the whole purpose was to make it a good user experience.

I do feel like it does become really hard as we try to infer intent for people on mobile devices.

BuzzFeed is leading the way in understanding user behavior

Will: I think that’s only going to get harder. We’re still kind of in that space now, we may be at a tipping point. Over the last couple of years, we’ve been in that space where the vast majority of our target market has a smartphone, but also people spend a lot of time on a computer, in different contexts. They spend their working day on the computer and are on the phone while they’re commuting, and you have these mental models of how that works, but that’s already broken down.

You look at some of the leaders in understanding this kind of user behavior. For whatever you think of the listicles and the cat articles, BuzzFeed is right there at the cutting edge of understanding how users are using all their devices these days. Incidentally, if anybody hasn’t tried the BuzzFeed news app yet, I think you should install that as a marketer, just to understand what is possible in some of these spaces. They’re doing some really smart stuff around customized push notifications. It will change your view of BuzzFeed – if you haven’t already discovered their news reporting – to realize what a serious news organization they are and how scary and dangerous they are to the established media players.

[quote style=”boxed”]For whatever you think of the listicles and the cat articles, BuzzFeed is right there at the cutting edge of understanding how users are using all their devices these days.[/quote]

Jonah Peretti, the founder of BuzzFeed, has laid out a bunch of his strategy – very clearly, very explicitly – and he’s talked about where they’re publishing content everywhere, not just on their own sites. They’re publishing native content on Facebook, native content on Snapchat, etc.  They are seeing these trends. A few years ago, BuzzFeed would see the biggest spike in mobile use in the evenings, people filling up their “couch time.” They’re seeing it much flatter through the whole week now, because people are on their devices at work as well. Increasingly, there’s all kinds of bits of work that can be done better on mobile devices.

You look at some of the presentations that have come from Andreeson Horovitz who’ve been doing research on mobile user behavior and they talk about Slack – the chats and collaboration tool.  They talk about how slack is actually a challenger not just email but also to things like spreadsheets and presentations.

It’s hard to grasp because you think when you’re never gonna make a slide deck in Slack. Aside from conference presentations, that the vast majority of presentations are for internal use. Most people’s job isn’t make a presentation, the job is convince somebody of something… A lot of those tools are better done by methods that aren’t PowerPoint slides, and Slack and things like that are disruptive because they don’t necessarily create a better presentation, they get to the job really was. If your job is actually communicating status, and you’re currently emailing rounds of PowerPoints, maybe your entire job should be replaced by Slackbot.

It’s interesting in particular because it means suddenly a lot of that stuff can happen on mobile. Slack is the place where mobile work happens. It’s an interesting mindset. I’ve even noticed this changing my user behavior. As my work becomes more like these discussions and meetings, more of it can happen on mobile. And as soon as it can, it will.

Everybody’s going to be using mobile all the time. I still think we’ll use proper keyboards and actual screens, but I don’t think it will be predictable which things we use for what.

References & Resources

• SearchLove Conference – Boston May 3 & 4 2016 (Save $150 with this link)
• Distilled.net
• Optimization Delivery Network 
• Follow Will on Twitter @willcritchlow
• BuzzFeed News App

Podcast: Play in new window | Download (Duration: 40:46 — 28.0MB) | Embed

Subscribe: RSS