Marketing Automation on Google and Facebook Explained i.e. Machine Learning

Nowadays, Google's and Facebook's machine learning algorithms will help you get your results faster, faster, and faster. All you need to have is a little bit of patience, a team of experts, a stack of creatives, correct set-up for your machines, and of course "right strategy".

The rest is history.

You will hopefully get what I mean from the right strategy when you finish reading this article.

Marketing automation embodies lots of other meanings. From creative automation to bidding automation, from budget scaling to targeting automation, from machine learning to marketing orchestration, and many more we expect to experience in the upcoming months.

One thing is crystal clear. Machines have become better each day in delivering results, especially in Self-Attributed Networks.

That's going to be our topic in this article.

So how can you prepare yourself as the decision-maker?

The answer is simple. By adopting the technology in your favor, instead of resisting it.

Where are we heading?

Self-Attributed networks are implementing features that help any advertiser to use algorithmic campaign management to optimize their campaigns. Both platforms keep pushing advertisers to depend more on its machine learning, and event-driven, algorithmic campaign management are going to control Facebook and Google advertising more and more in the following years.

Less is more now, and this trend will continue.

How does it work?

Google and Facebook’s algorithms usually work on Bayes’ Theorem. It allows machine learning to use some knowledge or belief that it already has to calculate the probability of a related campaign result. If you feed your machine learning, it simply calculates better results for the upcoming results. They usually use the Bayesian inference in which Bayes' theorem is used to update the probability for a hypothesis as more evidence or information becomes available. 

Considering the bidding automation in these platforms based on your KPIs, machine learning algorithms are no different from the marketing automation. It simply automates your optimization efforts and helps you to get there in a more sustained fashion. It’s not surprising to expect more and more new machine-learning driven features are going to roll out soon.

Automation will level the playing field in the competition and allow all advertisers to be even more equal. More competitors come in to play in the bidding play, more investors start to flow into the SANs which in turn more advertisers start to come into play under the same conditions.

How do you make a difference when all things are equal here?

I will cover that further in this article.

The Difference Between Machine Learning and Artificial Intelligence

Before jumping into Google's and Facebook's machine learning structures, I would like to clarify the difference between machine learning and AI because we, industry people, usually think that both of these concepts are the same thing. It's not!

Artificial intelligence is not something new. It has been around since 1950s. The goal was back then to enable computers to mimic human behaviors. Machine learning, on the other hand, has been in our lives since 1980s, which is simply a subset of AI.

I would like to tell readers to get ready for the concept of Artificial Intelligence Marketing which includes both machine learning and deep learning for the next 5 to 10 years, but we have been not there yet!

Nowadays, we are usually focused on machine learning which is the ability to give machines access to data and let them learn for themselves through the historical data with human interference. Computers figure things out from the data and deliver.

Artificial Intelligence, on the other hand, is the broader concept of machines being able to carry out tasks in a way that we don’t need to interfere in the decision making process. When artificial intelligence with the power of deep learning kicks in the following years, we would focus more on the business strategies, partnerships, and creating the ideal space for the AI.

#1 Google’s Machine Learning i.e. Smart Bidding

For marketers, Google’s machine learning is no exception from marketing automation for a while. Advertisers give up their controls to Google's machine learning, then automation kicks in to take control of what previously had to be done manually. Nowadays, we can give more attention to trying new things out, creative production, and focus on more to the new strategies to make a difference in the competition.

It's important to underline the fact before getting more details that what I believe Google search and display campaigns still have account-based learning, instead of campaign-based learning. So, structure your accounts based on a single goal thus single conversion. Create a single conversion accounts for better results.

On the other hand, things get quite a bit different for ACe campaigns. It just works for selective conversion campaign targeting. Thus, machine learning can learn from your campaigns based on selected conversion.

There are over one million bidding combinations available to you on just one keyword in that example. It becomes pretty difficult to scale that level of bidding precision when you’re managing accounts with thousands of keywords.

We created our Smart Bidding to solve that problem, and the only way to utilize Smart Bidding is through Google Ads. Smart Bidding takes even more factors into consideration when adjusting bids than are available in the interface with standard bid adjustments. Not only that, those bidding decisions are made uniquely for each auction that you enter. It’s highly specific bids, made efficiently and at scale.

-- Google

What is done It helps advertisers to get into the advertising auction with the automated bidding. It looks at your account history, the relevancy of your ad copy, the time, and so on. It’s all about giving you the best possible options to show you ad to the best match. 

#1a Search Advertising

In Google Ads search advertising, the machine learning evaluates how you perform in every 8 hours. If you have SA360, the auction-time bidding technology analyzes different contextual signals in real-time for every auction.

Well, money talks.

It might be a good idea to move on to SA360 if you are very advanced advertisers because theoretically speaking, it takes more things into account comparing to Google Ads. It also gives you the flexibility to use more than one conversion optimization with its weighted conversion bidding technology. You can also set up automated rules on SA360.

Put simply, both leverage the best signals and have to help set your bid right at the time the auction happens based on your given target.

They recommend you to add your first and third-party data audiences to your search campaigns as observational lists to feed your account's machine learning for a given conversion.

WARNING: If you want to acquire new customers for your display campaigns, your first-party data learning might bring you your customers because that's what you teach with your first-party data to the system. Don't blame the machines for your incorrect structure. :)

How does machine learning decide then?

It looks at the history of conversions. More conversions you have in your account means more learning it is going to have. More variation you have in your creatives more data it does have to learn from.

It’s all about that in the simplest form.

You have to give the most you can give to the machine learning to learn better and better each day.

Nowadays, machine learning uses even what is called “ghost advertising” to learn backend even before you start spending some in your campaigns. 

It's getting smarter and smarter. You should be happy with this :).

WARNING: you have to teach your machine in the right way. Like setting up the right attribution window and conversions. Otherwise, you might not get what you targeted.

#1b How about Mobile and Display advertising work for Google’s machine learning?

Let's start with mobile advertising.

You all know that Google ruled out the Admob ads 3 years ago. Advertisers now just give in what they want for their app campaigns, then let Google decides where theirs are going to be showing up. With ACs, you are forced to give up most of the control over creative, targeting, and optimization. It doesn’t end up here, what you can see now is very shallow compared to the times where advertisers can see where their ads showed up.

This trend will continue, while Google still works on perfecting its ACe campaigns.

Your salvation for mobile advertising here is:

• To create as much as creatives possible to resonate with your target audience,
• Set up your attribution correctly (try to limit tracking as much as possible to get better incrementality from your campaigns)
• Integrate Firebase as soon as possible to get the advantage of similar audience features that work backend for your AC campaigns.
• Target the right audience in your ACe campaigns and picking up the right conversion (select the earlier step of conversion if you don't receive more than 10 in a single day) for optimization.
• Design the right campaign structure for your AC campaigns (Probably the best scenario is to create one single campaign for each model and have the bulk of ad groups inside of these campaigns for your creatives) (You have more flexibility for ACe campaigns)

From my perspective, Smart display campaigns with discovery ads are the future of display ads in Google Advertising. We shouldn’t be surprised to see display campaigns turn into Google’s App Campaigns type of campaigns soon. Advertisers might get less control in display campaigns mean machine learning can decide more for you. 

Well lucky to you, you still have some time to play around. However, can you really get results for the segmented campaign structure like before?

Well, the answer is no. You have to leave your reins to the machine learning algorithm again from some perspective.

The salvation here is simple. Pick the right conversion (a single conversion count for each account), select the shortest conversion window as much as possible for you, use responsive display ads (you don't even get the impression from your classical 320x480 banners, my advice is to stop producing those and start working on responsive specs) and use similar/optimized/mass targeting options in your display campaigns. Maybe you can give a shot to custom-intent and life-event targeting.

If you want to get fast and consistent results, and your business addresses masses, I don't recommend you to narrow down your targeting on Google display ads.

Trust the machines! In machine learning veritas!

#1c How about the zero-sum game?

It’s still a controversial concept to understand whether it’s going to be something good for us or not. You have to understand that you get into the same basket with all other advertisers if you lay your back to the hands of Google’s machine learning. This simply means if you want more volume, you have to pay more for one more extra conversion.

Three things you can differentiate after you make everything you can for machine learning. Your account setups, your products/services itself, and your creatives.

#1d Attribution dilemma

Attribution is just a philosophy. Moving towards a data-driven attribution model from last-click attribution helps machine learning to learn even better. Move towards that approach whenever it's available in your account after receiving a certain amount of conversion.

#1e Final words for Google's machine learning

More than likely Google will continue to take your control away from other campaign types over time, and you can expect Google’s entire advertising platform to look like Universal App Campaigns at some point shortly. This means that machines will continue to take control over advertisers. 

Less human inference means more depending on machine learning. Machine learning decides in the likelihood that a click will result in any of these outcomes and helps adjust bids accordingly for your campaigns. So you have to give machines access to data and let them learn for themselves.

From targeting like similar audiences to smart bidding, from creatives like responsive ads to account-budget scaling, machines are on our side. Favor them on our side! You design the fate of your campaigns.

You can move your search operations to SA360, or Kenshoo Search to take full advantage of marketing automation on Google.

Again it's not AI, it's machine learning! So machines still need your interference in accessing data and let them learn for themselves.

#1f Does it outperform human workflows?

Again it’s all about how you design your machine algorithm. 

#2 Facebook’s Machine Learning Algorithm

Facebook relies on machine learning for a long long time. It works on showing the most related content to the people who care the most about in the given time. Facebook’s advertising is no exception. It works around giving you the best results with its machine learning. 

#2a Facebook's targeting is like collecting apples from trees

Facebook’s advertising simply tells you based on the results you get is broader you get better results you might get for your ads. More you feed the machine learning better it works you.

When you first start to advertise your campaign Facebook will start the learning phase to get what you want for your interest. Facebook will try to find which users within your target audience are most likely to take your desired action during the learning period. If you use a broader target audience, Facebook has a higher ability to catch high-converting users within it. Starting with small budgets for your ad sets would help you get lower CPA when the learning phase will be completed.

WARNING: If the learning phase doesn't complete in 7 days, you would guess that that campaign will never learn.

Let’s take the example of apples.

Imagine an ecosystem that your potential customers are the apples. Facebook’s machine learning, on the other hand, is the apple pickers and Facebook advertising is the very big farm that helps you to collect these apples. If you keep your targeting options broader, machine learning goes to the different apple trees and collect the lowest hanging ones. If you keep your targeting options narrower, it simply goes to the very few apple trees and starts collecting to the higher hanging ones. If it goes above the lowest hanging apples, it increases your cost per result. 

What do you do?

Narrower you get in your targeting decisions higher cost per result you are going to get. Broader you get in your targeting decisions lower cost per result you are going to get. It’s better to train your machine learning in a way that it always collects low hanging fruits.

#2b Change your creatives as much as possible to reduce ad fatigue

Relevance score, on the other hand, plays a different role and simply tells your machine learning to higher up ad’s CPM. Ads with low relevance scores come up with higher costs, and ads with high relevance scores come up with lower costs. Negative and positive feedback decide whether you get a higher relevance score or lower relevance score. Frequency also comes into play.

Once your machine learning hangs around a few apple trees, your target audience will be the same, thus get bored to see the same ad over and over again. This will lower your relevance score, thus you get a higher CPA.

To keep machine learning in your favor you simply have to have more ad variation. 

By selecting the auto-placement features and selecting different creative for each placement, you don’t need to worry about creating tons of different ad sets because of the creative dimensions anymore. From my perspective, dynamic creatives are going to play more crucial roles in the upcoming days. Showing the right creative to the right people at the right time will help to increase the certainty of machine learning. Using placement asset customization helps you to give different messages across different platforms. 

#2c Space is limited and competition gets more fierce each day. How can you adopt machine learning?

One thing is clear in Facebook's advertising. When you increase your spending, you get less return. Your salvation here is to adopt marketing automation to change your bids, budgets, targeting, and creatives automatically in real-time.

Hold on, hold on. Can Facebook ads handle make these automated decisions itself?

My answer to that is of course not! It still needs external marketing automation tools like Smartly, Bidalgo, Acquired, or Kenshoo to make these decisions automated. Like smart budget allocation, ad fatigue signals, automated creative changes, and ad level bidding. These tools are still an integral part of Facebook's advertising to adopt marketing automation in your favor.

If you want to learn my favorite one, please send me a message. :)

From Facebook's side, you can follow through these suggestions below:

Account simplification, auto advanced matching, campaign budget optimization (CBO), dynamic ads, and more automatic placements as they released under the name of Power5. Facebook moves towards creating an ecosystem for advertisers to make use of their machine-managed strategies even more. It might push advertisers like what Google did for UACs back in 2017 to adapt to certain limitations. 

Facebook's UAC which is called "AAA" is about to be released in the upcoming months. It is going to simplify the mobile acquisition and retention campaigns based on your goals. I like the fact that they call it "UAC Killer", we will see about that...

#2d How do you operate your campaigns on Facebook like a smooth operator?

Back in the time, many campaigns and ad sets account structure used to work best when machine learning is less extensive. Nowadays, It’s become crystal clear that machine learning will get ideal performance from less “campaigns and ad sets” in the account setup. Facebook’s machine learning learns deep into the campaign level. Fewer campaigns and ad sets you have better machine learning work for you in Facebook advertising.

The reason for me is a pretty simple algorithm can control your intraday account changes better than you can.

It doesn’t sleep, doesn’t get tired of, doesn’t drink tea, or never gets hungry.

Another reason is that it allows your campaign to get out of the learning phase quicker and get into the optimized phase longer.

WARNING: If you don't get results from your campaigns in a few days, it simply means that you are never going to get performance from that campaign. Simple delete it and open a new one with a new thesis.

What I believe is that the fate of an ad on Facebook depends on the first 1000 impressions or something close to that within a couple of hours after the launch. Is your target audience going to click, share, like, and comment on your favor? Or, controversially they click on "I don't want to see this ad", or comment negatively.

If things seem extremely bad to you after a couple of hours, I would recommend you to now wait to stop a campaign on Facebook.

The algorithm depends on these signals above. Please watch this video if you want to learn more.

#2e Trickiest part of Facebook advertising is the budget scaling

What I have seen in the past was getting a lot of diminishing returns when I tried to increase the budget for a given campaign. Later on, I decided to split the given budget across different ad sets evenly and started all of them together at the same time. Let's say that you decided to give a thousand Turkish liras for a given campaign. The smartest thing was to distribute this budget across 5 different ad sets evenly and watch out for the best performing outliers. If one of them goes well, simply duplicate it and shut down the bad performing ones.

Well, things have changed when Facebook introduced CBO.

Campaign budget optimization (CBO) has become mandatory for some advertisers this year. It is going to block giving different budget strategies for different ad sets. Theoretically speaking, it helps you to maximize campaign results and spend less on underperforming ads over time. To speed up the learning phase, it might be a good idea to batch different audience segments and placements under one ad set if it’s possible. If you have more than one edits to make, changing them all at once makes sure that learning is only reset one time. After the learning process will be completed, marketing automation comes into play when you don’t have to play with budgets of different ad sets anymore that often.

While these claims are theoretical, I still play the 5 different ad sets rule with different creatives even though I run my campaigns with CBO. The key part is to vary your creatives as much as possible with your mass target audiences.

If you are a high spender and target the mass audiences on Facebook, what I recommend you to open up a couple of advertising accounts to distribute your budget across these accounts. You should do that for your web conversion-based campaigns. Beware of the overlaps! Theoretically speaking it's not an advisable thing to see. Practically speaking, it hasn't become any big deal for our business.

In the end, Facebook wants everyone to compete in the given space. It's not like Google's display network. If you try to dominate Facebook with your ads, it won't simply let you do that with your desired return. You might experience diminishing returns. The smartest thing to do become your own competitor. Do things that your competitors haven't done before.

We are currently working on generating our own dynamic feed to use the advantage of catalog sales. We are also in the edge of dealing up with one of the biggest e-commerce brands here in Turkey to try out Cpas and will become one of the very first financial institutions in the world to take advantage of that.

WARNING: As an advertiser find out your best practice. It's all about test and learn. It might work for one of your clients, and might not work for the next one.

Facebook will check total value of ads [user value + advertiser value (bid * estimated action rate)] which try to enter auction and will let only the one with the highest value into auction. In Ads Manager, auction overlap % is shared with advertisers to help them understand the size of this overlap. -- Facebook

Marketing automation tools like I mentioned above will help you to play your budget automation game on Facebook. It will give your better budget scaling based on your KPI and overall goal. You can set up a budget for the selected campaigns and let your automation to distribute it. Some of them also give you forecasting.

You definitely need an automation tool here to do better budget automation on Facebook!

#2f Final words for Facebook's machine learning

Auto advanced matching will work on improving ad performance by more accurately ascribing where conversions come from, while dynamic ads will work on promoting similar or complementary products across Facebook’s platform. 

Facebook’s targeting expansion helps you to reach a broader set of people to meet your goals. Theoretically speaking, If you enable these features, you would be able to deliver more than you expected.

It's expected to see more automated advancements on Facebook pretty soon. However, I still recommend you to try social media automation tools until Facebook gets there.

Changing HR Ecosystem with Marketing Automation

Machine learning dependency lowers down the need for expertise in the channel ecosystem, and increase the need for other expertise like data analysts, art directors, and design growth professionals. Marketers have to transform themselves in the hiring process for their teams to keep up with the trends. 

We as marketers always look for more data-driven people the ones who like to deal with a huge number of metrics, sales funnels, and so forth, that we tend to hire very analytical people, and then we expect them to be creative. Yin and yang don’t work in this scenario because these two things tend to be rare to find in one person.

What we have to look for a creative manager in a mixed role unifying creativity and analytics. We can expect them to interpret hard data into actionable insights, produce massive amounts of creatives, and automate the testing process for analytical people. Again, channel expertise will be less prime, and the ability to create a high pace of new creatives to feed into algorithms are key elements for the next success. 

Coming end to my words, if you want to learn more about marketing automation, I recommend you reading the marketing automation ebook that I created with Adjust.