Marketing lies at the intersection of many business processes such as advertising, sales, ROI, customer acquisition and customer churn.
The marketing world is all around us and has grown into the world’s biggest digital sectors accounting for over $500 billion spending in 2019.
Marketing is yet another sector that benefits massively from machine learning (ML). Working seamlessly and intelligently with an abundance of data, ML algorithms can capitalise on data that already exists to drive new value and income streams, or induct entirely new datasets to provide new levels of marketing action and insight.
In this article, we’ll be assessing how ML interacts with the world of marketing alongside some key use cases.
ML in the Marketing Industry
Marketing is a data-rich sector. It fuses many disciplines together ranging from graphic design to data science and content creation.
At the intersection of many of these disciplines, you can use data to describe marketing channels, their performance and their effects. ML algorithms are exceptionally good at working with complex datasets and can provide accurate insights into campaigns to ensure businesses aren’t just shooting in the dark.
ML is not only useful for providing advanced and sophisticated measurements and insights for analysis by marketing specialists, it can also assist in ideation and the creation of new product concepts.
The internet is phenomenally wide-ranging and ML algorithms are capable of mining unstructured data for sentiment analysis.
Knowing what people talking about and why is absolutely essential for marketers and ML algorithms can go where no marketer has gone before – behind the scenes of billions of webpages – to mine insightful information on trends and sentiments.
The potential is massive.
The Benefits of Machine Learning in the Marketing Industry
ML is well-suited to an industry that works with large but complex datasets including many human demographic variables. Machine learning is the perfect partner for the big data revolution and allows SMEs and enterprises to make effective use of the vast quantities of data they are collecting today.
1. Reduced costs
In marketing, content creation and ad creation is only a fraction of a campaign’s lifecycle. With many platforms such as Google Ads, Facebook Ads & Bing Ads, it is possible to build a machine learning model that optimises cost across multiple ad platforms.
2. Reduced Labour
Parallel with cost reduction is the reduction of labour-intensive tasks. ML can automate certain time-consuming aspects of running a business, such as paperwork, accounting, cost analysis & data input. Marketers are already working with large volumes of customer and website data where machine learning can automatically yield insights or provide decisions.
Also, Chatbots are yet another key example of how ML algorithms reduce labour and associated costs.
3. Big Data
Big data – the term has been around for a while now and many will be familiar with it already. Big data is the action, or culture, of businesses ramping up their data strategies to collect more data on almost every business variable imaginable. The more data, the greater potential for efficiency gains and refinement, or at least that’s the idea.
The problem with big data, however, is that the more data you have, the tougher time you’ll have analysing it with traditional methods. ML changes this – it can work effectively with big data in a way that isn’t labour-intensive and for marketers, this means actually making effective use of the stacks of customer and site data a business might be collecting already.
4. Predictive Power
Marketing is about execution speed as well as accuracy. Beating the competition requires businesses to remain ahead of the curve. Predictive modelling allows marketers to predict what customers will do and when, as well as what products society is most interested in and whether there are either supply or demand surpluses across certain product ranges.
5. Value Creation
All of these actions and benefits create value for marketers. ML can run behind the scenes and automatically generate data insights whilst human marketing teams focus on creating the campaigns. Big data sets can be worked with efficiently and without unnecessary labour inputs from marketing specialists. Quantic Marketing found that 97% of marketers agreed that machine learning will enhance marketing performance and ROI over the coming years.
Machine Learning in the Marketing Industry: Uses Cases
Detecting The Main Content Of A Web Page
There are a range of python packages allowing you to extract the main content from web pages including:
These packages allow you to leverage DOM tree parsing and machine learning to extract on the most relevant pieces of text. This can be very useful when you want to use the data for search engine marketing or for document similarity techniques.
Automatically Assessing Content Quality
It is also possible to automatically assess content quality with BERT. This technique allows marketers to find un-optimised blog posts at scale. It can also be complimented with readability score python packages which include:
Text can be classified and labelled with machine learning, this allows large content publishers to automatically tag new text documents based upon the text classification of previously tagged documents.
It can also be used for simple binary classification tasks such as:
- Is the content evergreen or not?
- Is the content engaging or not?
- Does the content contain offensive language or not?
Documents can be clustered into topical groups using K-means clustering. Additionally, you can find the closet 10 or further 10 documents using word embeddings and Cosine similarity.
Trends and Sentiment Analysis
Marketing is heavily associated with human behaviour including our emotional responses to products and services. We won’t involve ourselves in a product or service that we don’t feel compelled by.
Question marketers must try to answer are:
- What are people purchasing and why?
- What topics are people interacting with and why?
Sentiment analysis can play a key role in answering these questions. Traditionally, marketers might use focus groups or targeted surveys to research the desires the public have for a product and this does still have its place, but ML has allowed marketers to tap into a much greater resource: the internet.
The internet is full of textual information and this contains sentiments. ML algorithms with the service of data miners can trawl the internet for these sentiments and provide marketers with an unparalleled level of depth and detail on their subject matter.
Sentiment analysis does not only reveal potentially trending areas for product creation, licensing and investment, but it can also reveal how people feel about a product or brand.
Unilever and Dessert for Breakfast
One excellent example here is how Unilever used ML to discover sentiments that indicated the public wanted desserts for breakfast.
Unilever developed an ML algorithm that discovered the sentiment “icecream for breakfast” was contained in over 50 songs. They also looked closely at how more consumers in the US were heading to donut shops in the morning and consuming more sweet foods for breakfast.
Resultantly, one of Unilever’s brands, Ben & Jerrys, released a brand new ice cream breakfast product that you keep in the freezer.
ML here revealed a sentiment that marketers may have never gotten a handle on themselves. “Two years down the road and our competitors are now doing the same,” says Sthanunathan.
Want to improve your data skills?
See the best data engineering & data science books
GitHub Curated List of Sentiment Analysis Tools
This curated list of Github projects for sentiment analysis is written primarily in Python but also in Java, R and Ruby. ML algorithms for sentiment analysis use deep learning and natural language processing (NLP). This allows them to A) locate relevant information and B) make sense of its tone and sentiments.
Key targets for sentiment analysis are:
- Social media, primarily Facebook and Twitter. Twitter is ideal because it is highly structured and charged with readily-analysable sentiments.
- High-tier publications such as The Guardian.
Personalisation and Recommendations
The way we interact with businesses is changing and the pressure is on for businesses to become “living” in that they know us personally and can provide us with a personalised experience. On the flip side of the coin, businesses that are impersonal may lose the trust of their customers. A report by Accenture Strategy suggested that poor personalisation and drops in consumer trust cost US businesses $756 billion last year.
Personalisation is an effective way of building rapport and creating a more humanistic ‘living’ business. Experian revealed that personalised promo mail has 29% open rates and 41% click-through rates compared to normal. Marketers can harness the power of ML to tune their content to different audiences whilst tailoring website UI options, ads and other campaigns to particular demographic profiles.
ML algorithms learn from our online shopping tendencies and habits and these have long been incorporated to platforms such as Amazon or eBay. Amazon collects and utilises huge amounts of data in their customer personalisation AI framework DSSTNE and these have been released in a GitHub project.
Github Project #2 – DSSTNE
Pronounced destiny, this GitHub deep learning spare tensor network project provides the tools needed to create product recommendation models. These work with sparse inputs and can scale up to enterprise-level if needed.
Collaborative filtering is one such method used to filter offerings to different groups of users. Clustering is used to group customers that have performed similar actions, for example, a jogger and a cyclist may have both bought health and fitness products and rated them similarly, and therefore when one of them makes a further purchase, this can be recommended to the other party also.
Recommendations using Sequence Analysis
Recommendation systems such as that on Booking.com use time-sequence analysis to recommend extensions to trips, places to stay and attractions to visit. Based on the time and date of the trip, their ML software can work out what specific recommendations to make e.g. seasonal attractions or special events.
ML algorithms can classify customer types to tailor recommendations. For example, there’s no good in recommending a business traveller stays in a campsite but this may be exactly what a young couple is looking for. Site data gathered that can help predict business travel, e.g. travelling solo, staying in a hotel for a short stay in a city, using business class, etc, can be used to train models that tailor recommendations for the different traveller types.
Campaign Analysis and Optimisation
Devising marketing campaigns is one half of the battle. What’s the point if you can’t figure out how well they’ve worked? Of course, there are quick and dirty ways to analyse marketing campaigns like measuring conversion rates from social media ads, promotional emails or other forms of marketing, but ML really allows you to get under the skin of your campaigns and automatically tweak them based on high-level or real-time data.
Automated Custom Campaigns
ML can help structure campaigns aimed at different demographics and use data from old campaigns to optimise future campaigns. ML can also customise campaigns on the fly, changing aspects of their design and colour to target different sets of users based on their engagement with existing adverts.
For example, Under Armour compares health and fitness information from users of their fitness products to compare these to similar profiles, thus segmenting them and allowing for targeted campaigns tailored to their fitness profile. These create value for both Under Armour and the customer, who will receive recommendations based on what works for them.
Fashion and lifestyle brand Sephora used ML to heavily customise their vast and complex high-end product offerings to their customers. Their data strategy used ML algorithms to pool and centralise data from all marketing streams, both online and in-store, to deploy sophisticated personalised campaigns and promotional offers for frequent in-store customers.
Targeted ads are also vital for targeting customer churn. By analysing when a customer’s interest in a business wains and targeting them with offers and other targeted promotions, businesses can effectively recapture customers.
A/B testing is central to creating optimised marketing and PPC advertising campaigns. Whilst traditional A/B testing compared a limited set of variables, ML can compare many interlinked factors to provide an altogether more accurate impression of how different ads and campaigns appeal to different segments of users.
GitHub Project #3 – Machine Learning A/B Testing
This repository provides the code required to perform advanced A/B testing in Python. It can measure and compare multi-step variables. It includes tools for collecting and cleaning data before performing regression and/or decision tree modelling.
GitHub Project #4 – Multi-arm Bandit Testing
An alternative to A/B testing is multi-arm bandit testing, aimed to explore what options capture the best traffic before funnelling users increasingly towards those site features and/or products. This is useful when training data sufficient for A/B testing is not present. It allows marketing campaigns or website UIs to change as new data on customer behaviour is inducted into the algorithm. This GitHub project contains 4 algorithms used in multi-arm bandit testing: Epsilon-Greedy, UCB1 and Softmax.
Identifying High-Value Prospects and Influencers
Targeting influencers is an excellent means to increase a brand or product’s marketing profile. ML can assist brands in locating and targeting influencers that will be most likely to accept product endorsements offers. Mazda used IBM Watson to locate and target influencers who would be most likely to endorse their new Mazda CX-5 ahead of the SXSW 2017 festival in Austin, Texas. T.
The ML algorithm scanned social media posts to discover which influencers could have the strongest brand affinity with Mazda and allowed 4 of them to drive the car and post their experiences to social media ahead of the festival.
Machine learning helps marketers solve many problems and has huge potential to automate and optimise marketing streams and funnels to newfound levels of performance. ML works well with big data but it can also collect, clean and use new data, including unstructured data, to provide marketing insights.
ML has also become increasingly able to deal with natural language and is able to creatively analyse sentiments and other text, this effectively allows marketers to tap into the internet as a rich resource of customer information, trends, thoughts and behaviours. From the examples here, you can also see how ML has increasingly creative and out-of-the-box uses.
Marketing With Machine Learning FAQ
How is AI Used in Marketing?
AI can be used to collect and analyse vast swathes of complex data including data on customers, their purchases, the products, services or content they interact with and numerous other variables. This provides marketers with an unparalleled resource from which they can craft personalised content, customised campaigns and optimised ads that can also function dynamically based on new data. This helps drive ROI and creates new income streams.
How Do You Use AI In Sales?
AI can be used to drive conversions with A/B testing that works out what products sell the best to certain audiences, or what page layouts convert customers with sales and subscriptions. AI using ML algorithms is able to use customer data to provide optimised ads and marketing campaigns that cater specifically to certain types of customers.
Is Machine Learning Used in Marketing?
Yes, machine learning is already used in many areas of digital marketing. Machine learning is exceptionally effective at working with big datasets and can analyse customers alongside relevant data to provide accurate insights into what products they’re interested in, their buying habits, etc. This allows marketers to target products and marketing campaigns at those most likely to interact with them.