How to respect customers in the new digital era: A case study in UX writing and machine learning

Teri Hason
Jan 20, 2020 · 8 min read

Written by Christian Reni and Teri Hason

In 2012, when a high school student received a flyer from Target for maternity clothes to her house, her father fumed. Why was Target encouraging her to get pregnant at such a young age? It turned out, Target knew her secret before her father did: she was expecting.

Identifying consumers around the second trimester would prove to be a goldmine for the company. This newfound ability set off a culture of panic and mistrust when we learned how consumer data was actually being used.

Years later, our fears about how our data is used still exists, but to some extent and depending on the industry, we’ve come to accept this reality. In our day-to-day lives, we make use of Netflix recommendations, share our locations freely, rely on complex algorithms for choosing a restaurant and use in-app products that help automate, personalise and understand our needs — sometimes before we even do.

And despite the perceived chaos that seems to be happening, from the inside a different story is unfolding. Those who produce the tech that we consume are also consumers themselves. We’re on a mission for the right and responsible way to use data that’ll help a greater good and propel our lives forward. While we learn to navigate the opportunities we’re discovering, we’re also a generation of socially mindful citizens who are acutely aware of the changes the tech world will lead us into. These realities are what shapes the questions we ask ourselves in our day-to-day work and forms our north star of accountability.

How privacy affects UX Writing

From Google searches to travel habits, UX writers are faced with a new challenge in understanding the best way to communicate with users. How do you explain the seemingly unexplainable? Do you need to explain everything or nothing at all? Is it better to be vague or explicit, and is our approach reassuring?

As UX writers at Booking.com, our approach is guided by our writing principles, which are first and foremost about prioritising the traveller. So even though we write for teams that create highly personalised experiences, the introduction of machine learning-derived products and using customer data meant we had to radically reevaluate how we speak to our customers.

A case study in writing for personalised models

When we first set out to write for recommendation models that used customers’ preferences and historical data, we assumed that the concerns much of the public had about data privacy also applied to Booking.com. This meant that our writing took a softer, less direct approach when we were making recommendations.

Is this really recommended for me?

We quickly learnt that going down the pathway of ambiguity was the wrong approach.

In the example to the left, the intention was to highlight why a particular property might be a good choice based on the fact that it matched their past preferences. So our assumption was that it was enough to simply let customers know that the property was recommended for them based on their previously used filters. This is when we got our first clue that something was amiss in our strategy.

When we went back to ask our customers about their thoughts about this product, the responses we got was that it sounded like just another generic marketing message. “Recommended for you” didn’t feel personal at all, but rather just the opposite, and the messaging didn’t provide enough of an explanation as to why we were recommending the property.

From these learnings, we rethought our strategy and became more direct about what we knew about our customers by highlighting what we thought matched their preferences. This, we hoped, would result in more interaction, which would translate to booked properties.

In the example to the left, we highlighted which feature of the property we determined was important to each customer (cleanliness, location, etc) with the corresponding guest review score.

But still, this failed to resonate with customers. Why? Because the model we used to determine what feature was important to each customer was so complex, it was beyond a one-line explanation. We started to understand that it wasn’t enough to just say what we knew about customers, but we also needed to give the messaging adequate real estate to say how we knew it.

Lack of real estate along with overly simplifying a complex topic leads to confusion and vagueness.

These failures lead us to deep dive into what we were doing wrong, which ultimately lead to a major shift in our thinking and an understanding of what’s actually considered to be overreaching. This insight would go on to guide the rest of our writing and could be applicable to UX writers across industries — even for those not working on products derived from models.

Creepy is contextual

  1. Travel is personal and emotional, perhaps more so than other industries. But the benefits of leveraging data for personalisation are so great that customers told us they welcome a more personalised experience that would provide them with better recommendations.
  2. Transparency builds understanding and trust, and messages must be transparent if they’re going to be believed. The importance of this is two-fold. First, we need to provide the right amount of real estate for explanatory messaging; one-liners simply don’t work for complex models providing recommendations. And two, with greater transparency comes more understanding, which ultimately helps build trust.

How to do it right: creating transparent and direct messaging that works

Image for post
Image for post

What we refer to as our “match score” is a feature we developed that indicates how closely a property matches a user’s preference and history. It was developed from some of the learnings discussed in this article and from the realisation that we needed to be more direct and honest about the fact that we’re using customer information to make recommendations. But perhaps most important to the customer is that this feature helps them make informed decisions on the different properties they’re considering.

Netflix offers a similar score. But unlike Netflix, which is made up of just what you’ve watched, our score is made up of more than just where you’ve stayed.

Image for post
Image for post

This feature is made up of a hybrid of both explicit and implicit data that our users share with us. While the method in which the model is communicated (with a percentage) may look simple, its makeup is anything but.

The first time we introduced the score, we intentionally didn’t offer a lot of explanation. But where we did, we wanted to be clear and as honest as possible.

The hover message (as shown below) was where we first explained the match score. We were focused on creating a transparent message that was easy for customers to understand and therefore (we thought) engage with. But it wasn’t enough. Our customers demanded more information and we moved quickly to provide it. Once again, we quickly understood that one-liners weren’t enough.

Image for post
Image for post

The score is complex and consists of reviews, past stays, indicated preferences, and so on. But we can’t say all of that — and even if we could, it wouldn’t translate into added value for the user. But we needed a way to indicate and explain that this score is unique to them.

Assist, don’t direct.

A key aspect of the match score is that it assists rather than directs customers what to purchase, which goes towards gaining trust. Match score is offered as a way to help reduce cognitive overload. It does not direct them or tell them what to book, e.g. “we recommend” is merely another tool to help them compare properties because, ultimately, we’re there to guide them — not to decide for them.

We knew that it was up to us as UX writers to communicate more transparency about the score.

To do that, the team created an explanatory card — appearing when customers clicked on a property’s score — in order to provide justification for each score.

An important part of this was to provide the review scores of highlighted pillars, such as value for money, location, and traveler type. For these pillars, maintaining the same language and rating system, as displayed on our site, was critical. This added a level of familiarity and lowered the cognitive load for first-time users.

The concept of recommending isn’t unique, but match score is, in the sense that it’s interactive and each time the parameters change (such as dates or the number of guests), the score is recalculated. Currently, while all users see the same value calculations (such as value for money, location, and traveller type) along with featured facilities, the score is truly personalised and every user potentially sees a different score.

Another key part of the messaging strategy was that the card could be both quickly scanned and read in context. User endorsements, which are personalised based on previous stays, are worded in such a way that they can be understood on their own or if needing more context as part of the descriptive text above. Read alone, “easy & convenient” and “scenic” convey features to guests that we hope will provide insight, with a closer reading revealing the source which they’re derived from.

One of the strongest ways that messaging can foster trust is allowing users to explicitly influence the score. Making space for users to provide feedback and gain information was important to us for future iteration to increase engagement and serves as an opportunity to build even more trust through transparency. We chose to offer two places for feedback: the classical thumbs-up/down presented in the centre of the card and within the information (i) popup, which serve to fulfill different necessities. The information not only simply displays a legality message (which we have to do), but provides an additional explanation for how we calculate the score. This helps answer any immediate questions. The thumbs up/down provide an entry point for further explicit user feedback. If a user isn’t satisfied, a popup will prompt them for additional feedback.

Image for post
Image for post

This feedback allows us to further communicate openly and transparently with users, constantly prompting us to rethink components based on user feedback.

Crafting engaging, human and simple messaging in the area of machine learning brings new challenges. Concerns about privacy and the use of data are real, and yet conversely we also welcome it into our lives in many ways.

Getting it right then may seem like a tightrope walk, but because travel can be an emotional purchase (and not necessarily a private one) we’re in a unique situation to assist in pivotal decision-making moments.

Regardless of the industry or platform, the focus of UX writers should be about creating transparency so that customers can easily understand what we’re communicating — including complex models.

Thank you to Steven Baguley and Sarah Wilson for helping edit this story.

Booking.writes

Stories from the writers at Booking.com

Medium is an open platform where 170 million readers come to find insightful and dynamic thinking. Here, expert and undiscovered voices alike dive into the heart of any topic and bring new ideas to the surface. Learn more

Follow the writers, publications, and topics that matter to you, and you’ll see them on your homepage and in your inbox. Explore

If you have a story to tell, knowledge to share, or a perspective to offer — welcome home. It’s easy and free to post your thinking on any topic. Write on Medium

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store