1 评论 5980 浏览 39 收藏 22 分钟


How to nail a great chatbot experience

Even though it felt like the entire world was building a next generation experience using chat bots in 2017, the reality is that we’re at the beginning of a slow-burn revolution that’s going to take decades.


Chat-bots are here to stay, but they aren’t the overnight paradigm shift some thought they would be for one reason: they’re hard to pull off. Chat-bots are revolutionary because they feel like a more human way to interact with our devices, but that’s what makes it so easy to get wrong.


Not only are there massive technical challenges — such as understanding user intent from free-form text — it’s a whole new paradigm for design: what do you do when there’s very little interface?

For designers working on chat, text itself is now one of the only canvases they have, making it the most powerful tool in the modern design kit.



Over the last year I’ve worked directly on a handful of chat-first interfaces with big brands personally, and wanted to look at what makes a great chat experience, from beginning to end.


It’s incredibly easy to build a bot but not something that people actually want to willingly use — it all comes down to the way the user experiences it and whether or not it’s getting in the way of actually getting the job done.



One of the first big brands in the world to wholeheartedly embrace chatbots was KLM, the national Dutch airline, which is often hailed for being an early adopter to new technology.

The company has one of the best chatbots available, and it has a good reason for caring so much about it: the company employs more than 230 dedicated agents to reply on social media.


With more than 100,000 mentions publicly every week, the sheer impact of being able to quickly solve simple questions with the use of artificial intelligence and chatbots is clear.


KLM has invested heavily in both chatbots and A.I tools to solve messages as quickly and precisely as possible, but has spent a lot on developing marketing tools as well — to the point that you can book almost your entire flight via Facebook Messenger!

KLM在聊天机器人和人工智能上投入了大量的资源,它希望通过这些技术能够快速且精准的解决用户的问题。与此同时,它也在投入了很多在具体工具的开发上。现在你就可以直接在Facebook Messenger上预定机票。

Not only is the KLM chatbot a fantastic thing to use, it actually seems easier than booking via the website, which can often be clumsy and confusing as you’re trying to figure out which button will do what you want it to.


Here’s what makes KLM’s bot so good, and how other brands could learn.


Don’t just assume a single intent


A common mistake I’ve seen from other companies that use chatbots is assuming that users who land on their bot will understand it — or have the same intentions.


This often leads to high failure rates as people just argue with the bot, which doesn’t understand their request, or they close the conversation immediately.

KLM’s bot understands this risk, so immediately offers the user a choice of where to go; is the query about support, booking a flight or something else?

Even if the other options end up with a human, this is a fantastic way to figure out where to route the user internally without any humans involved.




If you choose Book Your Flight, which is what this bot is made for, KLM lets you type where you’d like to go.

This is basically every bot developer’s worst nightmare, because users could say anything right now, and the bot is left to interpret it based on a very limited understanding of what could happen next.

Even if you get the user to write something you’re expecting into the text box, most people tend to type something vaguer than you’d hope at this point — leaving it with you to figure out the specifics of their answer.



Even being vague doesn’t break KLM’s bot


I ended up naturally typing New Zealand without the actual city I was planning to visit — and I expected the worst but found myself surprised: they’d thought of this scenario.

A good bot development project — particularly from the UX writing side — will consider all of the different weirdness that could eventuate here, and KLM did this right.

Not only did KLM ask for more specifics politely, they nailed combining the two separate data points to figure out what I meant, rather than forcing me to enter the full destination myself all over again.


KLM的确考虑到了这种情况,而一个好的聊天机器人开发项目,就应当考虑到各种类似于这样的异常情况,尤其在用户体验文案设计的阶段(UX Writing 是设计人和软件交互时所见话术的一种实践,它关乎设计产品和用户之间的对话— 知乎)。



When you’re building a chatbot, your words are everything. They’re the beginning and end of your user’s experience with you, so you can’t afford any misinterpretations, dead ends or confusing phrasing.


I’ve written the UX copy for a number of chatbots, and your use of language should be the principle consideration before writing a single line of code. I noted a number of places that KLM uses great copy to guide the user, so let’s walk through them.


(1)KLM sets expectations immediately by making it clear it’s a bot through the use of an emoji and in a friendly tone explaining its own limitations.

By doing this, the user already feels comfortable, but understands something might go wrong, so is far more willing to be patient because they know it’s not perfect yet.



(2)KLM uses a smart, subtle trick to win points from users: repeating what the bot understands to be the correct query back to them before continuing.

Once you’ve figured out dates and destination, for example, KLM spells the search out, offering an opportunity to correct any mistakes. This may seem tedious, but there’s a great trick behind this.

Think of the times you’ve used Siri and how frustrating it is when she gets it wrong; if a computer is trying to be human and makes a mistake, the illusion is ruined immediately. By leveraging subtle language cues, KLM able to avoid the computer giving the wrong answer before it happens, and maintain the illusion that we’re getting everything right, even if it isn’t perfect.




(3)KLM does a great job of helping you along the way with the wording it uses. When you’re given the chance to respond in free form, the chatbot guides you on how it expects you to respond.

These types of cues avoid frustration on the user’s part and make it easier on the developer’s side: predictable input is the best input, and trying to figure out if 11/04/2018 is the 11th of April 2018, or 4th of November 2018 is impossible if you’ve got customers around the world.


Dates are particularly hard, because there’s so many formats humans can respond in




A common area these bots fall over in is a lack of awareness of the user beyond that first interaction.

Often chatbots don’t understand who you actually are because they are unable to access data from existing backends.

KLM thought of this, and their bot is able to be useful beyond day one: you can choose to receive travel updates in one place and get your boarding pass without leaving it.

While it’s still fairly limited, this a great example of extending a conversational interface beyond just that first chat, and keeping users engaged long-term.



它将会话式界面(conversational interface)扩展到第一次交谈之外,并保持用户在较长的时间内依旧原意使用。


When Facebook launched its chatbot platform, there was a deluge of different bots to try, but many of them were a frustrating experience. As it turned out, many brands jumped on the hype train without really considering the nuances involved in building a great experience.


KLM is a rare example of a chatbot done well. While it’s not perfect, it’s a fantastic way to search for flights that doesn’t feel more cumbersome to use than its app or website — which is the entire point in the first place.


If you’re considering building a chatbot, sweat the details and more than anything else, focus on the words you use. Your phrasing is the beginning and end of a great chatbot story, and it’s key to whether or not it succeeds.


You can try KLM’s chatbot here.


原文作者:Owen Williams


本文由 @leglars 翻译发布于人人都是产品经理。未经许可,禁止转载


  1. 目前还没评论,等你发挥!