I’m currently renovating my house, and recently needed some items for the kitchen. After searching on Google, I found the items at a retailer that has a branch in my neighbourhood. I clicked through to their site, arriving straight onto a landing page featuring my item (a novelty in itself, even today), and was able to check the stock in my local store, and then click and collect.
This fairly mundane example describes a basic experience expected by today’s consumers. If we don’t click through to the items we’re expecting from the advert, we leave the site. If we can’t check stock for the item in a local store, or take advantage of click and collect, we also flee the retailer’s online presence.
Whatever we’re buying, customers demand personalised experiences connecting across channels and devices, guiding us along seamless journeys toward a purchase—and long-term loyalty. However, there is still a major disconnect between what consumers want and what businesses offer. Although nearly two thirds of consumers feel loyal toward brands that tailor experiences to their preferences and needs, fewer than 10 percent of organisations feel they are effectively personalising their messaging.
This gap has several interconnected causes. Most businesses recognise that personalisation can increase their relevancy with consumers, and improve their primary key performance indicators (KPIs), but many simply don’t know where to start. It can be hard to develop a roadmap and also prove success along the way. Sometimes it’s even hard to get buy-in from the business’s leaders, who may lack faith that machines and algorithms can effectively handle “human” jobs at scale.
On a more practical level, it’s challenging to develop a strategy for deploying artificial intelligence (AI) in accordance with a brand’s unique needs and goals, as well as those of its customers. Even where that strategy exists, many businesses remain too siloed to execute it. In most organisations, incomplete profiles of identical customers are fragmented across various departments and marketing channels. For example, it’s frustrating when your telephone provider offers support over Twitter, but can’t connect the conversation that you’re having with the call centre team, which happened to me last year. As the 2017 Adobe Digital Insights Summit Survey reports, close to 40 percent of organisations work with three or more analytics platforms, three or more attribution platforms, or three or more data management platforms. With each tool focused on a specific element of the data, customer profiles are bound to be fragmented.
The good news is that automation technology progresses every year. Tools such as Adobe Target now provide user-friendly solutions to many of these problems. By leveraging a spectrum of automated personalisation, ranging from rules-based personalisation all the way to fully automated recommendations, businesses are handing the repetitive heavy-lifting over to machines, freeing their marketing experts to focus on the innovative thinking that truly differentiates their brand.
Here are four reasons why your organisation needs to embrace automation to achieve deeper, faster insights, and stronger return on investment (ROI).
Resources are assigned to the activities that drive conversions exactly when they’re needed.
Blind spots pose a major problem for any business lagging behind on automation. Auto-allocation identifies conversion-driving activities and seamlessly routes traffic to statistically significant winners.
The latest generation of tools can not only A/B test multiple messages, they can also shift traffic to the best-performing experience, in real time, as they learn. This “multi-arm bandit” testing approach runs tests on 20 percent of your traffic, while delivering optimised experiences to the other 80 percent. Thus conversions and ROI steadily increase while you’re actually in the process of running the automation.
Say you’re running a conventional A/B test on an audience of 300,000 people. The test allocates Experience A to 100,000 people, Experience B to another 100,000, and Experience C to the final 100,000. Experience C turns out to be the best performer, making you £700,000, while Experience A makes £600,000, and Experience B makes only £550,000. By running this standard A/B test, you made £50,000 more than if you’d just served Experience A to every customer.
Watch what happens with an auto-allocated multi-armed bandit A/B test on the same audience of 300,000, with the same differences in profitability among the three experiences. After two weeks of testing, the algorithm learns that Experience C is the most profitable, so it immediately begins directing the bulk of the traffic to that experience. By the end of the two weeks, 30,000 people have been served Experience A, another 30,000 have seen experience B, but 240,000 people have interacted with Experience C, which netted £1,680,000 as a result. The multi-armed bandit test made £225,000 more than using Experience A alone.
Swisscom recently used auto-allocation to determine which phone model to picture as the default image on its product details page. They found that a silver iPhone delivered a 37 percent lift in conversions over the grey model they’d been using. This might not sound radical, but it highlights how something simple can have an impact on engagement and sales. Most companies are sitting on this kind of potential without necessarily realising it.
Recommendations become more targeted and easier to scale.
We’ve all witnessed the power of personalised recommendations on sites such as Amazon and Netflix. No matter what industry you’re in, you likely have a large collection of digital assets that need to be recommended at scale. The more precisely you tailor these recommendations, the more you’ll be able to serve customers the products, videos, content, and other assets they’ll want to interact with.
One of the most powerful tools for achieving this is an algorithm known as “item-based collaborative filtering.” By ranking your digital assets by popularity, recency, and frequency—as well as in relation to many attributes within each customer’s profile and purchase history—this algorithm serves recommendations with the highest likelihood of driving clicks, sales or whatever your particular goals are. You can even add manual rules, based on your own expert insights, and the algorithm automatically factors those rules into its decisions, too.
The more this algorithm learns from customer interactions, the more it begins recommending combinations and permutations of assets that never occurred to you before. These combinations might make no sense from a traditional viewpoint, but because they’re inspired by your customers’ viewing and purchase patterns, they’ll drive undeniable results.
Swisscom tested Adobe’s item-based collaborative filtering in its smartphone app, where the algorithm ranked the self-service tiles on its help page. Even though the tiles’ names and content remained the same, simply reordering them produced an immediate 2.94 percent lift in clicks. Swisscom is now testing the algorithm on its news stories page, and on various product recommendation pages.
Significant offers achieve the greatest impact when driven by automated customer profiling.
Although it’s crucial to leverage machine learning to optimise delivery of large libraries of content, sometimes the reverse is true: You’ve got several pieces of cornerstone content that need to be delivered; when working with a small number of offers, a technique known as “full-factorial machine learning” ensures that each offer is matched to the right visitor at the right moment.
This technique uses a range of complementary algorithms (e.g., residual variance, random forest, and lifetime value), along with data from visitors’ profiles. The model also learns from changing population behaviour in real time, enabling it to rebuild itself and weight each of those algorithms differently in response to specific behavioural triggers.
For example, look at Swisscom’s homepage. You’ll see it’s more minimal, in terms of content, than the homepages of many of its competitors. This is because Swisscom learns from its customers, serving only the most relevant content to new visitors. You’ll see only the offers that are statistically most likely to catch your interest, which makes you more likely to click.
With automated targeting, the right experience always reaches the right customer.
Effective personalisation is also about delivering holistic experiences tailored to each customer. To maximise your clicks and conversions, you need to be able to A/B test not only pages or products, but also large-scale variations in content, navigation, layout, timing, and many other interrelated attributes.
That’s why multi-armed bandit testing aims not only for individual pieces of content, but for whole customer experiences, testing and targeting of entire app or website layouts, as well as functionality such as personalised navigation and featured offers—all from one convenient control panel.
All these techniques boil down to the same central principles: Speak to your customers the way they want to be spoken to. Start simple, by automating A/B testing to achieve greater insights and ROI. Keep scaling automation beyond simple personalisation of assets to deliver holistic experiences targeted around each customer’s needs.