7 AI Tools for Brilliant Automated Facebook Ad Campaigns
Designing the perfect Facebook Ad campaign is no longer an art. It’s a science. You need to be embracing AI technology to design, optimise and automate your Facebook Ad campaigns, and stop wasting resources.
The Demise of Facebook’s Organic Posts
Over the last few years, there’s been a huge drop in organic reach on Facebook.
This began way back in 2012, when Facebook began to manage ad content, making the News Feed space much more competitive for marketers looking to get a big reach from high-frequency, organic content.
But in 2012, Facebook itself was reported as saying that marketers needed to assume that a day would come when organic reach is zero.
From 2013 – 2015, over the space of a few short months, organic post reach fell from 12.05% to 6.15%.
Put simply, there’s too much content being created on Facebook, and not enough space or inventory.
So what’s behind the big change?
The Big Algorithm Shift
The Facebook algorithm is pretty complex. It uses machine learning to adapt to individual user interests.
It identifies the content that drives the most engagements and shares, especially native and live videos – and gives this a nice little boost.
It also looks at whether or not users click on a post before liking it, and can identify “socially valuable” trigger words such as “congratulations!”.
The algorithm was recently updated to be more about “connecting with people, and less about consuming media in isolation”.
But what does this mean for marketers?
Essentially, Facebook’s goal isn’t simply to match News Feed content with what an individual will find the most interesting.
It’s also about displaying content that maximises the opportunity for social connections between Facebook users.
This means it’s harder than ever for marketers to get organic content in front of the right audience – because selling socks (for example) has zero social value in the eyes of Facebook’s algorithm.
If organic reach is dead, we need to be more heavily reliant on paid advertising on Facebook and Instagram.
Facebook is currently making around $4 billion in revenue from advertising.
So let’s say you’ve settled on an advertising budget, you’ve decided on the product you want to promote and you’re all ready to go.
What do you do next?
Do you stick together some copy and an image from your archives and hit ‘publish’?
Facebook ads don’t work with a “scattergun” effect. Your audience needs to be highly defined, your creative carefully tailored and your copy perfectly enticing.
But coming up with the right combination takes a long time.
That’s why you need to get some help from the latest generation of AI tools.
We took a look at some of the Facebook tools that use AI prediction models and Natural Language Processing!
Imagine you could find out which images and text are more appealing to your lookalike audience for a Facebook ad campaign, or a marketing email, without having to guess at random.
Pomegranate generates an “image appealingness” score to help you to select the best images. The free-access version allows you to build HubSpot and MailChimp integrations too!
Pomegranate uses machine learning trained from images previously seen by humans in different contexts.
This helps the model to identify which tags are positively correlated, and which are negatively correlated with human engagement.
DataSine is also fantastic for designing Facebook Ads, because you can use it to optimise and improve both images and copy.
The pro-version (coming soon) will also allow users to pick up images and text that are a better fit for different personality types.
Just think of the capabilities of this! For example, if you know that your target audience are predominantly introvert, you might be able to use AI to tailor every aspect of your ad creative to that type of person.
And rather than guessing that pictures of people on their own will work better for your introverted audience, you’ll be able to back this up with data.
Pomegranate in action
Datasine ran an experiment to see if AI could interpret beauty and predict engagement levels by pushing six images through the Pomegranate platform.
The AI then broke down each of the images into tags before assessing the tag’s positive or negative appeal.
Images of warmer seasons were ranked higher, whereas images with wood and more concrete features were seen as less appealing.
After these images had been analysed, the AI ranked them from most to least appealing and a Facebook test was run to see how accurate the predictions were.
The results? The two images that were ranked the highest by Pomegranate’s AI also performed the best!
forAdext is a tool that uses artificial intelligence to repeatedly define and optimise your target audience.
It’s machine learning models run thousands of simulations to find the best, most responsive audience for your ads.
It automatically updates your budget too, so you can increase conversions without wasting money on learning what doesn’t work.
Currently available for use on both GoogleAdWords and for Facebook Ad campaigns, Adext manages to achieve an average conversion increase of +500% compared with what humans were doing with the campaigns first.
Adext’s CEO, Daniel Molano, notes that ‘AI’ is being used as a marketing buzzword at the moment, and that a lot of the tools and products being marketed aren’t actually AI-driven. They still require too much human input.
Adext theoretically only requires a quick connection to Google AdWords or Facebook Ads and a quick 5-minute set up – and the rest is done by the AI, which uses the Bayesian Inference and Gaussian Process.
This is an awesome service for optimising your ad creative.
Based on the idea that machine learning allows Facebook to optimise and deliver ads to the correct audience, but only with enough data to make those decisions in the first place, Creadits helps you to work with designers to build an inventory of creative, then uses AI to analyse and rank your creative with up to 87% accuracy, so you can start using the best-performing creative straight away!
It’s pretty expensive, but worth it if you’re running a lot of ads and you’re short on design resources.
AiTarget was the winner of Facebook’s Innovation Spotlight award in 2016.
Essentially, it’s a platform that uses AI to optimise and scale campaigns, based on hyper-personalisation. Here’s an overview of what it does:
- Automatic Target Audience Splitting
- Dynamic Creative Optimisation – Delivering different versions of image, title and text to find the perfect combination
- Automated Ad management – starting and stopping ads, duplicating, changing budgets and changing bids based on results.
- Formula builder – Make a formula to calculate indicators, show them in the dashboard, create automation rules and use them for optimisation
Now this one is a bit more technical. Wit.ai proclaims to be a Natural Language Processor for developers.
By using it, you can create text or voice-based bots that users can interact with and chat with on their preferred platforms – including Facebook Messenger.
Using the ‘understanding’ tab, you can train your app to understand sentences.
For example, we used the sentence “What is the weather”, and set the “intent” to “weather”.
We tested it with “How’s the weather”, but it didn’t recognise the intent, so we manually marked it again.
By the third test of “What’s up with the weather?”, our app was able to recognise the intent!
While Facebook Messenger doesn’t allow you to use a completely custom NLP, you can use wit.ai to customise the existing one!
MonkeyLearn takes an input of raw text and maps it to specific topics and pre-selected sentiment scores.
The models have been trained on historical data labelled by humans and now can be used to automate similar tasks.
You can upload CSV/Excel files or connect with your apps via direct integrations, Zapier or Facebook API. Text may come in various formats such as emails, reviews, NPS Feedback, Surveys, Social Media posts and comments.
This has a number of different uses, but it’s very useful as a Facebook comments sentiment analysis tool. Use it to analyse the positive/negative sentiments left by your users and fans, and you’ll quickly be able to identify where you need to improve.
From the University of Cambridge comes this tool that predicts your psycho-demographic profile from digital footprints of your behaviour. You can link it to your Twitter profile or upload your Facebook Data.
This is what happened when we ran it on our Digital Marketing Specialist’s Twitter Profile:
According to her: “this is incredibly accurate, although I’m not thrilled that it over-estimated my age”.
Running this tool on your own profile will give you some interesting insights into the data footprints that you leave behind – and it will give you some interesting insights into the types of hyper personalised Facebook targeting that you can experiment with.
Give it a go, and let us know what you think about the results!
So, we’ve talked a lot about using Facebook tools to set up your ad campaigns, but there are even more free tools out there that use AI and social data to do pretty interesting stuff. This is one of our favourites.
We’re wrapping this up with a bonus add-on tool. While not strictly applicable to Facebook Tooling, Simplifai is a great tool for getting to grips with the capabilities of prediction models.
A prediction problem is basically a problem of filling missing information.
For example, let’s say you have a Google Sheet with 5 columns representing 5 different variables you collected data in the past. You could now train a model to predict one of these columns when we feed the algorithm with info about (any) of the other 4 columns. That’s what Simplifai does!
It will allow you to predict your expenses, your income, fraud detection, your logistic delay and costs. The use-cases are endless!
So, on to the technical stuff…
How does AI identify high-converting marketing images?
Being able to predict which images will convert customers the best is one of the most useful AI features for marketers using Facebook Ads.
As humans, we typically can’t tell what images our audiences will like – we’re subject to bias.
We might like specific colours, patterns or designs – but these biased options may not perform the best when actually placed in front of an audience.
Computer vision is a scientific field that enables computers to interpret and understand images in a similar way to how human vision does.
Predictive algorithms can use computer vision to decide which image elements are most likely to be appealing to specific audiences, segments and individuals, and therefore select the best options for advertising campaigns – the options that are most likely to convert.
Bayesian Inference is a method of statistical inference that tools such as Adext use to optimise your target audiences. Put simply, it updates the probability for a hypothesis as more information becomes available or is updated. It’s a bit like educated guessing – in the same way, we learn new information and update our predictions about an outcome, Bayesian Inference allows an algorithm to do the same to your Facebook Ads.
For example, let’s say that we decided to race two swimmers, Megan and Emma, four times. 3 out of 4 times, Megan won.
You’d predict that Megan would win the fifth race.
But what if, the one time that Emma won, the pool was cooler than usual. In the upcoming fifth race, you know that the pool will be cool again. What would you predict this time?
This is the type of problem that Bayesian Inference attempts to solve, and it’s the core of many AI prediction models. You can read about it in more detail here.
So, there you have it! A complete AI toolkit to get you started with intelligence Facebook advertising.
Go forth and optimise those campaigns!