Artificial intelligence (AI) and machine learning are woven into the fabric of our everyday lives today, often for huge benefits. (See understanding and using AI models).
Algorithms direct the flight patterns we depend on to keep us safe during air travel. Natural language processing (NLP) powers our interactions with Siri and Alexa. Machine learning drives the curation of our Netflix suggestions, sadly they deprive their users of the ability to upload their past library, and have a limited product range anyway, rendering the results as dire at best, unless you happen to fit a preconceived profile to start with.
On the other hand, whether the decisions behind the algorithms driving groupthink in our social media feeds and Google searches or the biased data reinforcing disparities in hiring, AI and machine learning aren’t always beneficial. But machine learning focusses on using the data that it has more than say neural networks which tends to more look for whims and tendencies based on it. (See The difference between deep learning and neural networks).
As consumers, citizens, and professionals, we should all have an understanding of the ways AI and machine learning are being put to use, how they affect us, and what benefits they provide. In this blog, we consider the role of AI and machine learning in personalisation. By grasping how machine learning is being put to use to drive personalisation, specifically within digital customer experiences, you’ll be better equipped to take advantage of this exciting technology for massive benefits in your organisation.
The personalisation imperative remains
Personalisation is here to stay. Once a luxury, personalisation has become a baseline service in today’s digital economy, one the vast majority of consumers appreciate. Sadly there insult an anyone policing determining what can and can’t be included. Thousands of companies that have made their millions developing segmenting and triggered solutions successfully for retailers would argue they deserve to be including, we would agree that personalisation is determined by a denominator of one, “you are not your brother’s keeper”.
If you don’t like even what your siblings do, how can you predicting product selections personal to each customer and hope to succeed? We all experience problems in our own lives where someone has offered you a product simply because you bought something almost the same last month – aarrrggghhh!. The result is easily apparant to the CMO through the abysmal returns.
Consumers’ love for personalisation makes sense, after all it’s what we all use everyday, such as when deciding whether we like someone or not. We look to see whether someone actually knows who we are, your likes and desires, interests and passions. If they’re merely full of bravado and don’t actually give a toss, it quickly becomes apparant very quickly.
Now apply that to you as the retailer, talking to your consumers, and it is reflected in what products they know you want to buy, and so offer them the right ones. Triggered solutions would now be saying “yes exactly, as illustrated as an abandoned cart” , but wouldn’t have a clue if consumers didn’t put products in their cart. Segmentation solutions, would still be showing people “red” ones. Machine learning, on the other, picks not only the product, but in the style, colour, cut and material preferred, obviously from their preferred brand, at the time and day they wanted.
We all embrace experiences that offer us value, and this means being treated like the individuals we are. All businesses today must look for opportunities to show customers they understand their interests, preferences, and intent by delivering relevant content and products to ensure they’re not wasting their customer’s time shopping with you. Even if this baulks in the face of practises, like those for shifting dead stock.
Unfortunately, getting a personalisation program up and running is no simple task. Enter machine learning, whose algorithms can support, automate, and accelerate the process. Be careful to investigate when solution say they use automation, as true ML solutions are now typically autonomous (without human involvement whatsoever) as opposed to those that still require extensive labour costs to exist.
AI and machine learning
Artificial intelligence (AI) refers to the broad arena of techniques used to get machines to perform tasks that appear intelligent. Machine learning is a subset of AI.
But neither machine learning nor the other AI methods can currently begin to compete with humans when it comes to improvising, formulating strategies, communicating empathetically, imagining novel situations, inventing new products, and the list goes on.
Machine learning and AI can now support many tasks and completely automate others, AI solutions like ChatGBT are now displaying creative intelligence too.
Machine learning techniques for personalisation
While machine learning can feel like magic, the truth is it’s simply statistical and probabilistic models put to work toward a (usually) defined end. Machine learning analyses large datasets to identify trends. From this it can extrapolate what’s most probable to happen or what type of experience is most likely to lead to a certain result.
Here are some of the most common machine learning methods used for personalisation and what they’re used for:
Regression analysis
Linear regression could help discover which pages are most likely to lead to a conversion. Logistical regression could be used to discover the best follow-up actions for an abandoned cart.
Association
From Netflix to Amazon, this method is a critical tool for building out recommendation engines. Based on your purchase of Dan and Chip Heath’s The Power of Moments, for example, Amazon’s machine learning recommends Seth Godin’s Permission Marketing. the nuances that distinguishes one offering to another increases with ever greater data collection and integration.
Clustering
Clustering algorithms are a great tool for grouping customers into segments. But not to be confused with email segmentation, as this is merely a data store from which greater division into personalised individual consumer preferences occur.
Markov chains
Can analyse a user’s real-time website behaviour and make navigation predictions based on it, which can be used to personalise their experience.
Deep learning
From the natural language processing (NLP) that powers Siri and Alexa to determining the value of possible direct marketing tactics to segmenting audiences for mobile advertising, deep learning is where much of the most exciting work in machine learning has been done in the last couple of decades.
Most machine-learning engines use a combination of these methods to analyse data and offer insight.
Getting started with machine learning in personalisation
It’s important to have a working knowledge of what’s under the hood, but you want to get the machine learning engine started for your personalisation program. The following are not linear steps to take. Your program will be unique depending on your market, size, and in-the-moment goals. But keeping these suggestions in mind as you begin imagining, designing, and creating your program will streamline the process significantly.
Keep it user-centered
The user is always the place to start. You know your business goals, and, hopefully, you’ve aligned them with your web goals. (If not, check out this article on customer engagement.) With these goals in mind, you can start looking for various ways to improve the user experience. What are the critical points of interaction? How can you remove friction or better direct a user toward a specific action?
Keeping your user’s needs front and center and letting empathy drive your use of machine learning and AI is a great way to ensure you’re offering value, versus just using the shiny new thing.
Know your rules
You can (and probably should) use personalisation across the entire web journey. This can take many forms, personalised search being one great example. There are, however, four broad categories of personalisation rule types.
Contextual
Contextual rules personalise experiences based on known facts about a user, such as Geo IP address or the channel of entry into a site. leading ML solution providers automatically include this along with each individual’s navigation, and idiosyncratic navigation “tells” that inspire awareness of imminent product preferences and immediacy
Explicit
Some visitors to your site will self-identify by filling in a form for a discount, giving you an email, etc. Explicit rules personalise experiences for these known visitors by, for example, using data from previous page views and conversions to assess the most likely content the visitor is looking for. To illustrate solutions such as SwiftERM deliver ROI at levels in excess of 5 figures have been published, verified by leading researchers including Forrester, Bain and McKinsey.
Implicit
When you don’t know who a user is, implicit rules can use pattern and persona matching to personalise experiences based on the actions anonymous users take on your site.
Custom
While requiring further development, custom rules can personalise using anything you want.
AI and machine learning can support all of these rules, but some solutions will require you to determine which rules you want to implement where and when.
Conclusion
Chances are my Netflix queue and your Netflix queue have at least one overlapping movie or TV show suggestion. But this movie or TV show probably looks different in each of our queues. While I love comedy, you’re (let’s pretend) a huge action fan, and Netflix uses this knowledge to tailor the image it places on the recommended movie or TV show. Now translate this into ecommerce retailing for product selection to put in front of your consumers, and you can see how and why the ROI goes through the roof.
Implementing solutions can be as easy as ensuring your stack is expanded to have these new and appropriate machine learning solutions on it. Pure ML SaaS require zero human input, which makes this the easiest marketing decision you’ve ever made.
