How we’re building a chatbot that does more than just talk

Serial skim-readers, user research, and a robot that just wants to help

The Booking Assistant: a travel companion that talks to people so you don’t have to.

Ever organised a trip with five of your most disorganized friends? I have, and I continue to put myself through the same ordeal every summer. Oh wait, better make that seven friends this year because Dan decided to tag along, and he’s bringing his boyfriend. Looks like I need to check in with my chaotic travel companions one last time.

How many extra beds do we need? Does everyone want breakfast? When does the bar close? Fast forward 100ish group chat messages and we’ve pretty much decided what we need. Then all I need to do is email the hotel to arrange it all. (Somehow we all agreed on the same place…)

I always thought it would be cool if I could just chat to someone or something who could figure these kinds of things out for me, or at least do most of the legwork. Especially if it meant I didn’t have wait around for emails or, worse, pick up the phone.

As it turned out, Booking.com was thinking the same thing.

Meet the Booking Assistant

Whether you need a late check-out, breakfast or extra slippers, the Booking Assistant is always there to help you arrange things or answer any questions you might have about a booking. It even knows whether your pet shih tzu can stay too! It’s a neat combination of design, machine learning, complex backend code and a whole lot of copy. It also involves three major pillars of the Booking.com ecosystem — guests, accommodation partners and customer service agents.

The accommodation provider and our in-house customer service agents communicate with guests via the Booking Assistant.

Over the last year and a half, I’ve watched our capable machine help everyone — from business travellers to messy groups of friends like mine — to organise their bookings. To do this successfully, the Booking Assistant needs to convince real-life people to trust a talking robot with something as personal and as expensive as their holiday.

As a UX writer, deciding what our talking robot would say to help it gain that trust has been an incredible learning curve.

Nobody reads, especially not the fine print

Surfacing the right information in the right order is important, especially when there’s a lot of it for the user to process. Reading through lots of text on a small screen is never fun and a lot of us deploy a popular coping mechanism to deal with it: skim reading. The only problem is, skimming makes it far too easy to miss something and/or totally lose track of what’s going on.

Imagine the chaos in my holiday group chat back when we were choosing the hotel.

To make sure you don’t miss anything important, one of our main priorities is to make sure you actually notice what you’re reading when talking to the Booking Assistant. It displays everything you need to read at the beginning of each message to either fix your problem right there, or tell you what the next step is.

When you have more than one booking to talk about, prioritising information is even more relevant; not least because it reduces the anxiety you might feel about a potential problem. Like, say, whether or not an extra bed can be added to a room when another friend shows up for that group holiday. Especially when my super pricey booking is non-refundable…

The most important thing we could tell you here is that breakfast is included. We also mention the room, in case you booked more than one with different breakfast policies. That way, you can start work on your buffet strategy before the hanger sets in…

“Helpful” is a personality type

Sometimes the best thing the Booking Assistant can possibly do is get out of the way.

It turns out there’s a time and place for entertaining, engaging speech bubbles. And it definitely isn’t when someone is trying to get something done. So to make sure it facilitates that, the bot only offers useful contributions to the conversation when it’s appropriate. These are either a suggestion, instruction or a fully-formed solution.

Once you’re done and there’s something to celebrate, the bot is more than willing to show you how happy it is for you with an emoji or a silly pet-related pun.

Timing really is everything.

Mr Sniffles will be pleased.

Honesty is the best policy

A lot of user testing and a healthy dose of common sense has taught us that most humans feel better when they understand what’s going on. They’re also more likely to cooperate.

Whenever the Booking Assistant needs you to do something, it tells you why. If it needs a helping hand from a Customer Service agent or accommodation provider, it tells you that too, as well as letting you know how long it could take to get a reply.

The Booking Assistant also does its best to make you feel reassured. It’s what our army of human agents do on the phone and via email every day, so it makes sense for their automated colleague to work the same way.

Fear of the unknown is a powerful thing, so we do our best to remove the unknown and, hopefully, a lot of the anxiety it can cause.

Let the users decide

More than once, I’ve been found lying down in front of a whiteboard after trying to make sense of more complex topics. Oddly enough, those exasperated moments led to what is possibly the most valuable realisation I’ve had at Booking.com so far: on this occasion, and many others, the users need to decide what happens next.

At the time, it was hard to step back and understand that the best people for the job were those who would actually use the final product. Nowadays, user research is a reflex.

It’s how we worked out that being helpful beats a silly joke at the wrong moment — if you’re looking for a parking space, you’re probably not in the mood for a knock-knock joke. It’s also what taught us that using the word “just” can come off as dismissive when we’re coaxing some kind of action out of a user. Who are we to say how much effort it takes to communicate with the bot?

That’s all up to them.

Information is only useful when it can help

User research is also how we finally came to understand that providing more information isn’t always the best thing to do.

During one particularly valuable testing session, a handful of users quickly pointed out that our hyper-transparent approach to a specific payment issue only piled on more stress without providing an obvious solution. We listened, and replaced it with a brief explanation of why something is happening and what can be done about it.

When it comes to something as serious as what’s happening with your hard-earned money, proactivity counts for a lot. Sure, we can show you a long, detailed explanation of how something like preauthorization* works, but unless we also tell you what to do about it, it’s useless.

Providing a clear next step helps to reduce anxiety when a user doesn’t know what’s going on.

It’s more than just words

There are so many different things that go into writing for the Booking Assistant. This means I get to do cool stuff like work on interaction design for days at a time. Or hang out with data scientists while we debate whether or not abandonment is technically a bad thing (sometimes it isn’t!). Or go sit on a real-life user’s couch and ask them all kinds of questions about how they use what we’re building. It’s so much more than just writing.

And the best part? It feels like we’re building something that works.


*Preauthorization is how an accommodation provider tests a payment card to make sure it’s valid, by holding an unspecified amount of money for an unspecified amount of time. See? So long and vague it got its own footnote.

We’re always on the hunt for new writing talent, wanna join us? Apply here.