Skip to content

Using Machine Learning to Make the Most of Your Facebook Campaigns

--Advertisements --

If there is one thing that the last few months have taught everyone, it is that Facebook has a large amount of data about its users.

Naturally, recent scandals have cast a negative light on this revelation – some users feel uncomfortable with sharing their private data. A handful of major brands such as Mozilla and Commerzbank have taken a stand against the platform for security reasons.

But amidst all the pessimists, we, the marketers, should not lose sight of the fact that data, when used correctly and safely, is a gift. This allows us to run campaigns that are both highly effective for advertisers and relevant to users, minimizing the number of annoying or unnecessary ads they face every day.

But this wealth of data is not solely used for targeting purposes. When you run advertising campaigns on Facebook, you have a plethora of automated optimization options that, when used properly, can help you run more effective and efficient advertising campaigns.

Let's look at some of the ways you can use Facebook's highly sophisticated machine learning to improve the performance of your campaigns, with proven examples of its success.

Objectives of the campaign

Having worked in other channels before the social, one of the first things I noticed and liked about the Facebook Ads Manager, is that it was requires you to choose a goal for your campaign.

-- Advertisements --

I have seen countless examples of search or display campaigns that attempt to be a game of chance, and they inevitably end up mastering no one. "I need this campaign to reach far reaching, a lot of clicks, and of course a positive return on advertising spend (ROAS)."

Sound familiar?

Too often, advertisers can forget that if you want to achieve multiple goals, you need multiple campaigns. But Facebook does not let you forget:

When each new campaign is created, think carefully about what you are trying to accomplish and select appropriately from the 13 options available to you. If you are trying to achieve two of these goals, create two campaigns.

This campaign goal is not just an attribute of vanity. It tells the Facebook algorithm to whom it should put your ads in front of. If you choose Video Views, the algorithm will place your ads with users in your target audience who, in his opinion, are more likely to complete the video views.

See also  How to Build a Stronger and More Effective PPC Team

At Merkle (my employer), we have seen dramatic results in ensuring that the goals of the campaign accurately reflect the needs of a client. Once, we took over the activity of Facebook for a new customer of the utility sector and noticed that while it was clear that their main goal was account changes, all campaigns were using Purpose of traffic.

-- Advertisements --

It is not possible to change a campaign goal after the launch. So we decided to rebuild from scratch and switch to Conversions. By making this simple change, the team was able to increase account changes by 40% and stay well below the target cost per action (CPA).

Tip: Facebook users believe that "clicky" tends to be very expensive, and you often find that campaigns with a Traffic goal can have an incredibly high cost-per-click (CPC).

As an alternative, consider using a Conversions lens, but select a custom conversion higher up in the funnel than the final purchase or registration – for example, a page view or a quote.

This way, you will not get into the fray with all the other advertisers vying for these clickers, but the algorithm still has enough data to find people likely to come to your site .


Gone are the days when Facebook was only Facebook. They now own four of the top 10 most downloaded apps in the world: Facebook, Messenger, Instagram and WhatsApp.

And even though they have not yet monetized WhatsApp as the platform of ads, Messenger and Instagram are available as placements in your Facebook campaigns.

-- Advertisements --

They sit alongside the Facebook (FAN) and In – Stream audience network as an alternative stock they can take advantage of.

It is clear that Facebook has expanded the sites this way: The more ads there are in more places, the more money it can make. But there is also a big advantage for advertisers. By selecting all the default locations, placement optimization can take place.

See also  AI marketing and travel through the strange valley

It's here that the Facebook algorithm can decide where to show your ad based on where it can reach the lowest cost per thousand impressions (CPM). For a large entertainment client, we found that using site optimization allowed CPM to be 20% lower than manual Facebook feed selection. and the right column.

On the other hand, getting out of the news Facebook has some risks.

FAN, Instant Articles and In-Stream carry all brand security risks. While Facebook has provided tools to mitigate this risk, such as blocking lists and category exclusions, brand security may be a problem if a slightly higher CPM is a small price to appear on the line. d & # 39; news.

-- Advertisements --

We also observe different degrees of success with different placements. When the FAN can realize CPM reductions with some customers, for others, we saw it work badly and decided to delete it. I would recommend starting any new campaign with all the locations selected, then regularly evaluating the performance and adjusting your strategy as needed.

Dissemination of advertising

In addition to using campaign goals to help Facebook decide who to show your ads to, you can go further by selecting the "Optimize for Ads" method.

The options available to you will depend on the purpose of the campaign you have selected and will allow you to accurately target the right users to achieve your goals.

Recently, the value has been added to this drop-down menu where you select a Conversions goal. So, instead of just optimizing your ads to people who can convert, Facebook can show ads to people who can make valuable purchases, increasing the ROAS of your campaign.

-- Advertisements --

We have recently tested this and have seen a great success. Among a selection of customers testing value against conversions, we found that, on average, optimization toward value improved ROAS by 110%!

However, there was one exception: a large retail customer saw that his ROAS was 40% lower than value-for-money, and we decided to go back to Conversions.

See also  Quantcast, Smaato Publishes Tools in Support of the IAB Framework for the GDPR

But all we are taught is that there are few strict rules when it comes to launching Facebook campaigns. Testing is absolutely essential, and the Facebook Sharing Test feature allows you to accurately test one optimization option against another to determine which one is best for your campaign.

Final Thoughts

Machine learning and automation are hot topics in the industry, but as for many current topics, it can be hard to see s & # 39; There is substance behind the hype. Is it really useful, or just a buzzword?

The combination of innovative advertising products and rich Facebook user data means that it's worth it to trust them to make decisions. They know a lot about the people you are trying to target, more than you could ever hope to learn with all the market research in the world.

So, test as many of these options as possible, making sure you remember that what works for one campaign may not work for another and you should constantly evaluate the performance.

-- Advertisements --

The opinions expressed in this article are those of the guest author and not necessarily Marketing Land. The authors of the staff are listed here.

About the author

Laura Collins is Payroll Director of the UK-based pay media agency, Merkle | Periscopix. During her five years of digital marketing, she gained extensive knowledge of Facebook, Twitter, AdWords and several other platforms. She has managed accounts in various sectors with a specialization in finance and retail. Laura is a regular Marketing Land Associate and recently spoke at PPC Hero Conf Philadelphia and SMX London.