AI-driven Personalised Marketing Part 2: Examples and Use-Cases
In my last post, I introduced one of the biggest trends in AI-driven marketing: personalisation. I recommend you check out the full post, but in a nutshell:
marketers have always tried to segment customer types and determine what messaging they would prefer, but in the past, the segmentation was coarse and the personalisation options were only available through simple, hand-written rules. But today’s consumers expect personalised messaging with every brand interaction, and luckily, thanks to masses of data, big data techniques, and AI, marketers can deliver it.
In this post, I’ll show you just how the marketing playbook is being rewritten, using real-world examples from both tech giants and innovative startups*.
*I’ll focus on Austria, since it’s where I work and has a buzzing AI scene!
Insider tip: interpreting an AI solution
All of these examples involve machine learning algorithms, which learn to predict a target variable based on patterns in some input data. If you’re not so familiar with the workings of machine learning, you can roughly infer how each example works by asking: what hidden variable do I need to solve my problem, and what kind of information would help me predict it?
Note that most of these solutions still need a marketer’s know-how, and may apply automation rules on top. Let’s consider an example:
Problem: create personalised loyalty incentives for individual customers
Target variable: customer lifetime value (propensity model)
Input information (training data): purchase history and demographic information per customer, for all customers
Inference (test) case: purchase history and demographic information of a single, unseen customer
Automation rule: if predicted target variable > €X, offer incentive Y
So, with that in mind, let’s jump into some examples!
Examples and Use-Cases
Optimised Emails and Personalised Messaging
Using machine learning plus automation rules, marketers can predict engagement of different subscriber types for different conditions and content, and then automatically create dynamic emails based on factors like previous email and website interactions, products or content viewed, time spent, most popular current content, and behaviour of similar visitors.
Push notifications are also becoming more popular. They create the positive feeling of receiving a personalised text, so are more engaging — which is great for marketers, since even engaged consumers use promotions and junk inboxes to filter their emails.
Timing can be optimised to the moment customers usually check their inboxes or phones, for convenience and better engagement. Or, using geo-location services, a store could trigger alerts when a registered customer visits them.
Tripadvisor uses customer data on previous trips and site browsing to detect and/or predict attractive locations and send helpful, curated emails about them.
Starbucks offers a loyalty card to record purchases, and an app through which customers can place orders. Combining these, they use predictive analytics to determine effective, personalised messages to send through the app, such as greetings and cross-sells, when customers approach a store.
Consumers expect customised shopping experiences, and want the benefits of offline shopping, online. To deliver this, recommendation engines use customer, product and site traffic data to learn relations between types of customers, products and content.
Content-based filtering recommends ‘similar’ items based on their characteristics, like having the same brand as a just-viewed item. It’s useful for first-time visitors (that’s why when you visit an online retailer for the first time, you immediately get relevant recommendations, though the store knows nothing about you). Collaborative filtering, on the other hand, recommends ‘similar’ items based on how groups of users have interacted with them. This is useful for returning visitors with a history.
At Amazon, collaborative filtering matches each user to similar users, and content-based filtering matches their purchases to similar items for recommendation. This gives them authority, and increases margins when the additional value exceeds the additional shipping costs.
YouTube gains more revenue from ads than subscriptions, so they work to show you just you want. Asos’ Fit Assistant’s sizing recommendations increase satisfaction and reduce returns.
Presize.ai (Munich, Germany) uses computer vision to 3D-model a user’s body, from which it can determine a shopper’s body measurements (the user has to take a video of themselves standing and turning in a slow circle). Based on this, it can provide sizing recommendations.
This is better than asking users personal questions, and it requires less product data from the retailer, since the recommender systems learn not only from existing product data but also from data on gathered measurements, purchases, and — most importantly — returns. It can even recommend a different size to your usual one, if it learns that a specific brand is larger/smaller than usual.
So-called ‘conversational commerce’ is becoming more and more common, driven by increased adoption of smart devices, and by advances in natural language processing. It’s predicted that the future of search is assistive: a search engine or avatar will guide you through the search process, rather than leaving you alone with a search bar and a whole internet’s worth of results
The North Face uses IBM’s ‘Watson’ to enable online shoppers to discover their perfect jacket by asking the customer questions about their intended use. Watson makes recommendations based on product data and additional sources such as typical local weather. Macy’s also used Watson to create a ‘personal shopper’, an AI-based shopping assistant.
Sentient Technologie’s (Vancouver, Canada) visual filter lets users click images of products they like, instead of having to search. And with Pinterest’s Chrome extension, users can select an item in a photograph on any website and preview similar images from Pinterest, thus enabling them to see something, want it, and find it.
Many websites already use some ML to customise content, learning from non-customer-specific data like trending topics or common searches for search prompts, or using location to localise the experience.
Getting more sophisticated, AI can use live behavioural signals to decide what to serve, and it can adapt this continually, rather than waiting for results as was required by manual A/B testing in the past.
For example, based on the predicted likelihood that a browser will click, add to a wish-list or make a purchase, you can dynamically serve personalised content or offers.
Or, collaborative filtering can discover different visitor types and provide them with an evolving, tailored experience.
YouTube, using Google Brain’s deep learning technology, has two neural networks. The first uses information on videos, playlists etc to learn to select hundreds of candidate videos; the second takes user feedback such as plays or thumbs ups and learns to rank the recommendations. This is even done according to your current mood, which is why the recommendations change with every video you watch. Such ‘stickiness’ is achieved by Netflix and Spotify, too.
506.ai (Linz, Austria), aim to deliver ‘unparalleled personalisation’ and thereby a better customer experience. Their services include dynamic ads, built by designers based on analytical insights and deployed using automation. For example, if you sell weatherproof clothing and it’s raining where a user who lands on your website is located, a banner for raincoats can be displayed automatically.
Personalised Products and Services
For products, it’s no longer true that ‘one size fits all’. For example, sentiment analysis, which uses Natural Language Processing to understand text, can be applied to reviews, social media comments and service emails to discover attitudes and trends and tailor products and marketing to reflect it.
Deep Opinion (Innsbruck, Austria) can apply sentiment analysis to reviews, complaints, and surveys, enabling customer service teams (and marketers) to find areas for improvement.
Many insurance providers use machine learning to learn what specific factors influence risk. This is used to create individual, more accurate policies, which saves all customers money.
In the ‘Nike maker’s studio’, customers could design and purchase their own unique speakers. In addition to sales, Nike scored customer preference data which could be used for future designs.
For services, AI solutions can reduce costs and provide a better experience. For example, chatbots are more capable and available 24/7 for customers, at a low cost to the company.
Harley Davidson’s ‘Albert’ flags potential high-value, lower-funnel customers to be contacted by a customer service person for assistance right when they need it.
Onlim (Vienna, Austria) enables allow marketers to “connect all relevant components of an integrated e-commerce solution for product search, recommendations, offers, promotions, and more”, in order to build conversational chatbots capable of so much more than FAQ answering.
AI-driven personalisation is coming: it’s demanded by consumers, desired by marketers, and enabled by Big Data and ML. From a business perspective, it will enable marketers to engage more effectively to improve business metrics. And from the customer’s view — that is, you and I — it will enable businesses to engage more meaningfully, and provide better retail experiences.