On Friday 11 October on the Personalisation Stage at Festival of Marketing, we explored just that.

  • Why personalisation seems hard
  • How we can simplify our thinking
  • How we can help the unknown user
  • How machine learning and AI fits in

The personalisation struggle

When it comes to personalisation, many brands suffer from the same problems:

  • Not enough bandwidth to do personalisation properly
  • Not enough connected information about customers
  • Not enough content to serve customers individually
  • The wrong technology or systems in place

Brands are suffering from personalisation paralysis – they know the why of personalisation, but can get stuck on the what, and the how.

Hand-made vs. Automation

As human beings, traditionally we consider everything made by hand as something special; as something individual and personalised.

Conversely, we think everything made by machine is generic.

But in the age of the Internet, where the same experience is pretty much seen by everyone, and personalisation rules are set up manually by someone sat within the marketing department, we need to turn this way of thinking on its head.

In the age of the Internet, the things designed by human hand are generic. It’s the things made by machine that are personal.

Today, customers are crying out for tailored products, services, and experiences.

Brands need to move to a model for personalisation that hits the sweet spot of less rules, with more people seeing a uniquely different experience.

That’s where machine learning comes in.

Manual rules 
Brands need to use machine learning to have less rules and more customers seeing unique experiences

AI to the rescue

There are five ways in which machine learning can work:

  1. Recommendation – providing customers with ideas of thing they will enjoy, based on previous interactions, other similar users’ interests, or related content
  2. Prediction – using historical data for suggestion, to influence immediate purchase or action
  3. Classification – suggesting the right kinds of answers your customers might need to their search or problem
  4. Clustering – grouping similar products or services together in the background to give customers better recommendations
  5. Generation – creating a new feature, product, or service, based on existing data

What does this all mean for personalisation?

To deliver more unique experiences for customers could mean more hand-written rules. However, most of us are not Spotify, we’re not Amazon, and we’re not Netflix. We don’t have the size of company with the bandwidth and the staff numbers to enable this. Our data is in silos, we don’t have any data scientists, and 95% of our web traffic is unknown.

Scientists call this the Cold Start Problem – where you cannot draw any useful inferences about someone before gathering some information about them.

AI can help.

Personalisation via content-based filtering

Microsoft Azure Cognitive Services has a great tool called Text Analytics.

This machine learning tool allows you to detect sentiment, key words and key phrases, named entities, and language from the text on your website and within your documents.

Content based filtering 
Content-based filtering can be used for better recommendations

This allows you to recommend similar articles to website users more accurately than just via a rule that says if someone reads this article about ‘German politics’ recommend them this article on ‘Angela Merkel’.

Machine learning in the form of Microsoft Azure Cognitive Services Text Analytics might, for example, recommend an article on ‘Brexit’ that would be more relevant to the user based on the content of the article they were reading.

A human being just wouldn’t be able to think through all the permutations of possible recommendations.

Personalisation through session-based missions

Every website session has a potential mission or a goal in mind. There’s a reason that person has come to your website today – you just don’t know what it is.

You could base your personalisation rules on historical path analysis, to say that someone who lands on this page normally does ‘x’ or goes to page ‘y’, and set up a rule accordingly.

However, in reality, there are too many missions, and too many signals, to be able to set up all the manual rules you need.

Sitecore path analyzer 
Sitecore’s Path Analyzer shows you all the paths users take on your website

Imagine you’re shopping online for a shirt, and the first shirt you look at is a lovely white dress shirt.

The marketing team at the retailer might have set up a rule to say – If someone views this product, recommend a matching suit, trousers, and socks. Fair enough.

But you then take a look at a more casual white shirt. It’s still a shirt. It’s the same colour. But it’s a different style.

The marketing team have set up a rule to say – If someone views this product, show them a matching pair of trousers, shoes, and a nice jumper.

But you’re looking for a shirt. You don’t want to buy anything else.

Clothing personalisation 
Brands need to move from “next time” personalisation to “real time” personalisation

Surely it would be better if the website recommended similar kinds of shirts in real-time, based on what you’re looking at, rather than based on some rules made up by the marketing department?

Machine learning can help.

Brands need to move from “next time” personalisation to “real time” personalisation.

Want to learn more?

We’ve been working with the leading technologies in the personalisation and machine learning space for some time, namely Sitecore and Microsoft.

If you’d like to know more about these technologies and what they can do for your business, or you’re interested in a proof of concept for AI / machine learning to get things up and running in your organisation, then we’d love to hear from you.

To talk personalisation and machine learning, just get in touch.