Product
Solutions
Resources

Machine Learning Marketing Uses: How to Enhance Analytics and Win Jeopardy

machine learning marketing

The best techniques and technology in the world won’t help you if you don’t know how to use them. Annoyingly, with it being so new, it’s hard to tell what machine learning marketing use cases there are to base your efforts on.

In other fields, the breakthrough tech is being used in medical diagnoses, spam filtering, fraud detection, translation and more. All you need is some inspiration to get started with your own uses.

That’s why this post exists.

I’ll cover, in detail, the key ways machine learning is affecting marketing, and how it can be used to handle some of your business process automation, provide actionable data insights and predictions, and even win $1 million on Jeopardy.

Before that though, if you don’t know what machine learning is (or want a refresher), check out our other posts on the topic below:

Let’s get started.

Machine learning marketing uses

Getting content suggestions for news and new ideas

feedly machine learning

RSS feeds are vital for keeping on top of the latest industry news and learning new tips and techniques in your field. It’s also a great way to fill in your dead time (e.g., on your bus ride home) and let you read more, which is a great way to generate blog ideas for your own content.

Feedly is an awesome example, with their mobile app letting you subscribe to various blogs, websites, and content aggregators to make reading the best articles around a case of scrolling through a single feed.

The main problem comes when you’re trying to find the websites to subscribe to. It’s hard enough to keep on top of recent articles without having to spend time vetting new sites for valuable content.

That’s where Feedly’s machine learning element comes into play.

Instead of recommending sites based on vague categories, Feedly assesses the user-assigned tags of every website and uses them to assign categories and relevant topics. Other factors like follower count and engagement amount are then used to dictate which websites are recommended with a higher priority, with the entire model being assessed and automatically updated based on current user tags and interactions.

The result is a recommendation system which detects the best relevant websites to offer to make getting the best content easier than ever before. Don’t sit around – get inspired.

Enhancing existing tools

machine learning example salesforce einstein

Marketers rely on analytics and predictions on who to target, what they will interact with, where the best place to find them is and so on. This is usually done using a formula, looking at recent trends, as part of an overarching technique (eg, targeting relevant keywords) or via the highest paid person’s opinion (HIPPO).

Trouble is, customer behavior is changing constantly, meaning any plans or formulas currently being used by your marketing department will become inaccurate and need updating sooner or later.

That’s where machine learning can help through services like Salesforce Einstein.

Using machine learning to crunch data and automatically update predictions for everything from customer reactions to conversion rates, Einstein lets Salesforce users take advantage of the power of machine learning without the need to create their own tool.

The result is a powerhouse suite of tools which lets teams do what they do best instead of getting mired in the tasks of data organization and analysis.

Automatically answering questions

Imagine a program which could automatically interpret what someone is saying and respond in kind. This would mean your team wouldn’t have to worry about answering basic questions and could focus on more valuable tasks like creating content or guest posting.

It’s called Natural Language Processing and, although there’s not a perfect solution available, machine learning is getting us close to achieving it.

Human communication is frustratingly vague at times; we all use colloquialisms, abbreviations, and don’t often bother to correct misspellings. These inconsistencies make computer analysis of natural language difficult at best, but in the last decade NLP and machine learning has progressed immeasurably.” – Seth Redmore, Machine Learning vs. Natural Language Processing

Watson is a system developed by IBM which does precisely this. Originally developed to answer questions on Jeopardy, companies like Caffe Nero are already using it to gather customer information, answer simple questions when customers sign up, and so on.

Watson also serves to eliminate some of the standard marketing automation tasks such as reminding customers when their subscription card balance is low or when seasonal offers occur.

Outperforming A/B tests with multi-armed bandit testing

Multi-armed bandit testing” might sound like a predatory gimmick but it’s actually a powerful machine learning marketing use case. It’s also got nothing to do with the gambling machines it’s named after (hence why I don’t feel sleazy recommending it).

Regular A/B testing displays two options to audiences and gives equal weight to both, measuring the results. While this gets the job done, it wastes a lot of time and traffic on the less-effective method and requires a lot of manual management.

Bandit testing doesn’t need the same level of human intervention to get quality results.

bandit testing machine learning uses

Instead of using two elements, bandit testing typically arranges three (or more) and adjusts the weight it gives each option according to its performance. After the initial performance data is received, the testing model is automatically adjusted to send more traffic to the better performing results, letting you benefit from them immediately with next to no downtime.

Here’s where knowing precisely what classes as “machine learning” is important.

If all bandit testing did was send more traffic to better performing results it wouldn’t technically be a machine learning marketing technique. Instead, the required self-improvement comes through the program asking itself “what is the probability of option X being the best?”, working out that probability for each arm within a certain data set, then repeating that every time new data is added.

With every new piece of data, the pages will be assessed and the traffic formula automatically adjusted to represent that, making the technique an example of machine learning.

Bandit testing naturally allows for more elements to be tested at once, thus improving your services faster and spending less downtime setting up the test for another run. Instead of ending the experiment and resetting the options, all you have to do is swap out the worst performing one with a new option and keep the test running.

It’s simple but effective – exactly what a marketing team wants to hear when it comes to continuous improvement.

Making SEO easier and more relevant

rankbrain machine learning technology

I have a huge amount of respect for anyone who can fluently speak a second language – the furthest I ever got was learning how to say “to take away please” to my local pizzeria when living in Florence.

Now imagine having to understand over 100 languages well enough to accurately predict what someone means when they ask you a question. That’s exactly what Google’s search engine has to do every time it’s used.

Manually programming every potential response simply wouldn’t have been possible, and so Google introduced RankBrain to handle unknown factors for them.

On a daily basis, 15 percent of queries submitted — 500 million — have never been seen before by Google’s search engine…” – Dan Farber paraphrasing John Wiley (lead designer for Google Search), Google Search scratches its brain 500 million times a day

This example of machine learning uses Google’s vast data reserves to make accurate guesses for when a search term is interchangeable with another. In other words, it helps people find relevant content more easily, making it more important than ever to have a central optimized keyword in your content.

For example, let’s say you search for “how to use a guitar”. In this case, “use” could be replaced with “play” and it would mean the same thing. However, “play” couldn’t be substituted in if I searched “how to use a toaster”.

Images of trying to “play a toaster” aside, RankBrain helps people using Google to find useful content which is optimized for relevant keywords to their search. All you have to do is make quality, targeted content.

Getting actionable insights from analytics

analytics machine learning marketing

There’s a fair bit of overlap with the examples of machine learning marketing uses but that’s for good reason. By the very definition of “machine learning”, they have to be tasks which could be performed by people, have an evolving model/prediction, and all be based on large data samples.

In this case, it’s interpreting data from Google Analytics, with machine learning allowing insights to be gained more easily.

Take the language interpretation ability of RankBrain and the innate power of Google Analytics and you have Analytics Intelligence. This lets you ask questions to find custom variables and views of your analytics more easily than setting them up or searching for them yourself.

Say you want to find out how many users subscribed to your blog in the last week or what your desktop bounce rate in a specific country is. Problem being, you don’t want to find the result manually or create a formula to do so.

Instead, all you need to do is ask the question in natural language and Intelligence should be able to interpret and fulfill the request automatically.

Yet another case of machine learning making data analysis easier and more accessible to everyone. After all, who doesn’t love the ability to understand their audience more effectively and accurately?

Segmenting customers by interests (not useless facts)

affinio machine learning

Half of the battle with marketing is knowing your target audience – the other half is taking that knowledge and effectively catching their interest. This usually involves grouping your customers by factors such as their age, spending habits, and so on.

Affinio uses machine learning technology to group customers into interest groups using data such as social media activity. This lets you know more than the bare facts, showing what topics and emotional tones you could use to resonate better with your target audience.

If all you know is that your primary targets are 18-24 years old and don’t spend much, you have very little to work with in terms of marketing to catch their attention. If you know that most of them love cooking, holidaying in the sun, and watch the latest blockbusters, you suddenly have three massive topics which can be used to tie into your marketing.

Affinio’s graphs also make visualizing the size of audience interest groups incredibly easy to see. In the age of machine learning and social media, there’s next to no excuse for not knowing how to resonate with your target audience.

Predicting customer conversion

machine learning API salesforce edited

Predicting customer conversion is a massive deal. The more accurately you can predict whether someone is likely to convert, the more effectively your team can focus their efforts on the easier wins and produce better results.

You don’t want to waste time trying to convert a big lead if they didn’t have much chance of converting in the first place – effort would’ve been better spent closing 3-4 smaller deals which were a more surefire success.

As you might have guessed, machine learning can help greatly with this, as predicting conversion likelihood involves analyzing a huge amount of data and takes too much time to act quickly enough to follow up on the customer’s initial sign-up. That’s one reason why customer priority is often defined by their potential deal value.

All you need is a little developer knowledge to really get the ball rolling.

As Joseph Ferraro and Vincent Reeder (both of Mavens Consulting) demonstrate in this Salesforce Live show, a little API manipulation allows you to do things such as assess customer conversion likelihood in a fraction of the usual time.

Their demo shows Google’s Prediction API assessing new customers via the data they submit (such as the products they own) in Salesforce. This is compared to existing data sets which can be manually supplied by the duo, then produces a rating which is fed back to in Salesforce automatically. This lets the team know which customers are likely to convert (based on previous data) and should thus be high priority.

Being a machine learning technique, the predictions will only get more accurate as more customers sign up. There will always be outliers but the ultimate conversion rate will affect future predictions with each closed lead.

Analyzing customer sentiment

customer sentiment analysis

This next example is from the same Salesforce Live video used above, except this time looking at our old friend customer sentiment. This lets you automatically know the mood of your customer according to their messages.

Once you know how your customers are feeling, you can target the more negative messages (and thus upset customers) with your high-end support faster than working through every request manually. In other words, it lets you apply help where it’s most needed or where failure will be most damaging to your reputation with the customer.

Ferraro and Reeder trained their prediction model using random tweets containing various emoticons. These tweets were given an emotional value according to whether they were positive or not (a smiley face would be four points while a sad face would be zero).

This trained the machine learning model to recognize by association what language it could expect to see in positive and negative messages. Thus, when they apply it to a new message logged in Salesforce via a customer email, it’s able to automatically detect words such as “sad” and assign it the appropriate negative sentiment.

Support teams can then perform their duties more effectively by focusing their talents on the most negative sentiment messages first, potentially reducing customer churn in the process.

If you’d like to set up your own customer feedback analysis system or you’re interested in sentiment analysis, check out our related posts below:

Machine learning is like automation – it’s more powerful with creativity

As shown above, machine learning can’t be applied in only one way. You get the most out of it by getting creative with your uses and thinking about what’s possible, rather than what’s obvious.

Not to mention that new breakthroughs are being made with machine learning all the time. The future only knows what advanced AI, machine learning, and natural language processing will do for marketing and businesses as a whole.

So what are you waiting for? Identify tasks that you could use machine learning in and find out how you can save time, money, and resources.

What do you think of machine learning in marketing? I’d love to hear from you in the comments below.

Get our posts & product updates earlier by simply subscribing

Ben Mulholland

Ben Mulholland is an Editor at Process Street, and winds down with a casual article or two on Mulholland Writing. Find him on Twitter here.

Leave a Reply

Your email address will not be published. Required fields are marked *

Take control of your workflows today