top of page

Machine Learning, Facebook and getting it right.

Machine Learning may quite well be the future of digital marketing, a view shared by 97% of marketing influencers (Park, 2018), and with continued investment in 2019 we may be finally on the cusp of the continued evolution of the digital transformation cycle that began with the very first Google Ads.

Facebook logo in amongst mathematical formulae

What precisely is Machine Learning? Well it’s certainly evolved from its’ modest roots of the past where computers were programmed to complete very specific and somewhat basic tasks within a specific period of time and then require human intervention (Andrieu et al, 2003). Nowadays machine learning is around the use of artificial intelligence and an emphasis on the ‘Learning’ part of the name – software programming will now learn from previous successful formula’s, completions and results to produce accurate and most importantly repeatable results and decisions.

The biggest benefit of Machine Learning is certainly its’ ability to not require that aforementioned human intervention, being able to instead produce and predict outcomes with no explicit programming. The use of complex algorithms, pattern recognition and statistical analysis form the primary framework of the technology and allow said complex work to be processed in nano-seconds or less* (Amazon, 2017).

One crucial fundamental to understand when approaching the idea of Machine Learning is that AI (Artificial Intelligence) is thrown around repeatedly, we must understand that AI is ‘the broad science of mimicking human abilities’ (SAS, 2018) where Machine Learning is a subset which is around the learning aspect and then growing out-with original human capabilities, abilities and restrictions.

So where does AI and Machine Learning fit within the marketing landscape and why is it now so important? According to The Economist (2017) ‘the world’s most valuable resource is no longer oil, but data’. Between Alphabet (Google), Facebook, Microsoft, Apple and Amazon there is an increasingly unstoppable trend towards the value of data with Amazon capturing not only half of every dollar spent in North America but also the data that goes with it ranging from your demographic and personal information to your lifestyle habits, addresses with location data and most likely your workplace – add in some smart ‘machine learning’ and they could also likely predict your Household Income (Amazon, 2018).

This is now a world of segmentation on steroids, allowing marketers to now target specific personality traits, job profiles and exact individuals; think of a Sniper rifle rather than a spray can. The new problem that’s created is how do you manage this amount of data, use it or even make sense of it? (DMI, 2018) In an ever increasing cost sensitive business environment with looming Brexit, more savvy consumers and start-up competition there is a necessity for computer assistance which arrives in the form of Machine Learning, now being a necessity rather than what was once seen by companies as a possible ‘nice to have’.

ML Explained

Let’s look at one example of AI and Machine Learning in practice on major retail giant Amazon’s entire website and digital marketing strategy using the clustering algorithm K-means clustering of which Hartigan (1975) explains takes key data variables (Clusters) of which can then pair these clusters together based on similarities and ultimately then allow you to provide a recommendation based on past behaviour or current data set.

Here’s a brief look at K-Means Clustering based on a fictional limited customer data set;

Table representing customer purchases against product items

In summary: Customer A purchases a red hat and a red bag, whilst Customer B purchases a red hat, yellow coat and a red bag and Customer C purchases a yellow coat and red bag.

We can now deduce from table 1 that if based on the purchasing behaviour of Customer B who purchased the Red hat and bag and also the yellow coat along with Customer C who purchased the yellow coat in addition to the red bag that we could recommend the Yellow coat to A based on their purchase of the hat and more importantly the Bag. Likewise we could recommend to C the purchase of the red hat.

In a clustering diagram we would see it looks like as below in Diagram 1 showing the key relationship between each vector and products, ultimately forming a small cluster. If we were to add additional data-sets to this then it can grow exponentially.

A diagram showing K means clustering

Diagram 2 from Coffey (2016) shows how the multi-dimensional Clusters would begin to be transformed with larger data-sets which would allow for much more actionable insights.

A diagram showing a clustering dimensional model

One further useful way of looking at K Means clustering is diagram 3 from Towards Data Science (2018) where we can clearly see the clusters and patterns forming based on specific previous behaviour – sound familiar? It should be, as this is precisely what marketers have been doing for decades in the form of segmentation based on demographics; it’s why Tesco launched their Clubcard and the reality behind most major ‘store cards’ (Clements, 2018).

A diagram showing K means Clustering

This is of course just one use at a very primary level of which is called Descriptive Modelling and can be used as in this instance for customer product recommendations which could then be featured across remarketing channels for purchase acquisition (Amazon, 2018) in particular with passive buyers – allowing advertising and marketing to be much more relevant to their tastes and interests.

Such data management or ‘mining’ is crucial as data-sets grow and variable vectors increase exponentially ultimately forming Big Data with the understanding that will then allow every marketing pound (or dollar) to be measurable and provide a clearer return on investment.

Phil Gerb who is Chief Digital Officer at Vengreso wrote that [Machine Learning] will ‘lead to mass personalization and super-specific paid advertising that knows our intentions even better than we do’ (Park, 2018).

Machine Learning and Facebook

Facebook has continually been seen as at forefront of algorithm-powered content management, which is to be expected when last year the company made $39.4 Billion Dollars on ‘sponsored content’ (Swan, 2018).

A tweet from Mark Zuckerberg

Relevance is the thing that we must understand and we should really approach Facebook digital marketing strategy the same way we do with Google and Bing search engine optimisation as at a fundamental level such as what we explored with the Amazon ML concept above, relevancy is at the core with authenticity and high quality content always rising to the top.

In Facebook’s own statement ‘The goal of the News Feed is to show you the stories that matter most to you’, so to approach a socially charged Facebook strategy then we must understand that our marketing must matter to people.

A diagram showing an inspirational quote

Facebook’s NewsFeed system which displays posts, ads and content which is personalised based on what the users wants to see (Carrie, 2018). For instance, when scrolling if you stop at look, engage or interact with a particular friend’s post then Facebook will actually show you more from that friend and the same goes with business pages and so the necessity to have relevant high quality content that users engage with to ‘pass the algorithm’.

Passing this algorithm falls under 4 vector ‘variables’ just like as K-Means Clustering:

1. Inventory – The posts available to display

2. Signals – What the post is: Graphical? Video? Text?

3. Prediction – Facebook predicted ‘reaction’ based on previous behaviour of similar ‘signals’

4. Score – Based on the above 3 vectors

Let’s start with inventory as new businesses and also some established brands have fallen foul of this after the big Facebook algorithm changes in 2017. How much do you post a day? You should think quality rather than quantity – it can be easy to be drawn into a never-ending circle of posts and ‘chasing engagement’, but actually in reality this can be damaging to your ‘inventory’ portion of the score – remember to be relevant.

Avoid click-baiting of ‘Like XYZ, share XYZ’ as Facebook’s algorithm actually will penalise you as it isn’t relevant and could be seen as a social media form of spam, instead you should begin a discussion and conversion with potential customers and your audience.

A diagram showing a negative comment on an Ad

What is your brand and how best is it to portray the message you want to put across as social media is conventionally a type short form media consumption rather than essays or Hollywood Blockbusters. There is something quite powerful about a short ad rather than long form and research by media agency PHD (Facebook Business, 2018) shown that 6-second video ads performed better than their 30-second counterpart with an ‘estimated 11% increase in ad recall’ and for your inner business acumen, an 12% increase in return on investment spend.

What is the best practice? Well it actually depends, however what is commonly agreed in the sector is that content that naturally performs well organically will always do well and additional ad-spend to ‘boost’ and accelerate reach opportunity.

This is crucial not only to your online and Facebook success, but as a wider business after Sprout Social (2018) discovered that relevance is key to reception with brands who are authentic online and actively take a stand on issues will result in a higher level of brand loyalty, recommendations and higher weight on the purchasing decision.

You are not only fighting for business from customers now but the extensive use of Machine Learning/ AI by major players like Facebook of can actually be detrimental to your business if you don’t also begin to make crucial strategic decisions based on their use. At least, in Facebook’s case the only way they can judge the relevancy and quality level of content is through the aforementioned algorithm – not all is lost though as it’s increasingly ensuring a level playing field and the ‘value of the platform’ with meaningful interaction rather than passive ad experiences.


Park, A. (2018) Is Machine Learning the Future of Marketing? Experts weigh in. QuanticMind (Online) Available: (Accessed 15 November 2018)

Andrieu, C., De Freitas, N., Doucet, A. and Jordan, M.I. (2003) An introduction to MCMC for machine learning. Machine learning, 50(1-2), pp.5-43.

Amazon (2017) Overview: Replacing Tape with Cloud in Backup Workflows: A comparison and Three Reference Architectures. Amazon ML Labs (Online) Available: (Accessed 15 November 2018)

SAS (2018) Machine Learning – What it is and why it matters. SAS The Power to Know. (Online) Available: (Accessed 15 November 2018)

The Economist (2017) The world’s most valuable resource is no longer oil, but data. The Economist (Online) Available: (Accessed 15 November 2018)

Amazon (2018) Privacy Notice. Amazon UK (Online) Available: (Accessed 15 November 2018)

DMI (2018) How to Apply Machine Learning to your Digital Marketing Strategy (Online) Available: (Accessed 3 November 2018)

Hartigan, J. A. & M. A. Wong (1979) Algorithm AS 136: A K-Means Clustering Algorithm. Journal of the Royal Statistical Society. Vol 28 (1). Pp.100-108.

Coffey, K. (2016) K Means clustering for customer segmentation. (Online) Available: (Accessed 15 November 2018)

Towards Data Science (2018) K Means Clustering: Identifying FRIENDS in the world of strangers (Online) Available: (Accessed 15 November 2018)

Clements, L. (2018) Tesco is tracking your Clubcard data to see how healthy your shopping basket is. Wales Online. (Online) Available: (Accessed 15 November 2018)

Amazon (2018) Big Data Analytics Options on AWS. Amazon. (Online) Available: (Accessed 3 November 2018)

Swan, G. (2018) The Facebook Alogorithm. CPC Strategy. (online) Available: (Accessed 15 November 2018)

Carrie, J. (2018) Facebook overhauls News Feed in favour of ‘meaningful social interactions’. The Guardian. (Online) Available: (Accessed 15 November 2018)

Facebook Business (2018) In a Mobile-First World, Shorter Video Ads Drive Results. Facebook. (Online) Available: (Accessed 15 November 2018)

Sprout Social (2018) Championing Change in the Age of Social Media. Sprout Social. (Online) Available: (Accessed 15 November 2018)


bottom of page