A Creative Director on chatbots - Part 1: This is great... but how do they make us feel?

Andrew Larking

Feb 21 2017

Led by Andrew Larking, Creative Director, this will be a multi-part series of short blogs that cover chatbot technology from a UX and design perspective.

Find out more about our chatbot services.

The web has been a stale and scratchy place for a few years. We - the people who build it - are a very inward facing bunch. Sites are looking very similar, very much on trend, and they follow UX rules which seem to have come from a random opinion solidified into stone.

"Ye must have the word 'Search' in your search bar."

You don't. Looking out the window this morning I think the high street fashion world must be similar to the web design world. I saw a line of four women wearing the exact same outfit, with the same hair, and the same walk. It was like being in a Netflix Original TV Drama about clones. Or like looking at the Awwwards site.

But last week Ronald threw a grenade into the room and then locked the door. Conversational Interfaces, of which the most common expression are currently chatbots, have entered our lives. And wow do they make things exciting again. New design challenges, no pre-existing user experience patterns, it’s like the Wild West and nobody has much of a clue what’s going to happen. All we do know is that whatever happens is going to start the next web revolution.

Before I begin the diarising of what we are up to I wanted to take the time to ponder what this technology could maybe one day do for us. I think it could allow us to reimagine the web, and this is why.


Site search is normally terrible. If you look after a site and you're reading this thinking "my site's search is great” then you're either Larry Page or you're lying to yourself. They are universally bad. Site admins and content editors build taxonomies into their systems which is what search hooks into, but unless those taxonomies are continuously evolving - which would be a full time job for someone - they quickly become out of date. Language changes and trends in language use changes, and as they do your taxonomies need to change. I'd put money on it that the average taxonomy change happens once per site rebuild. Chatbot technology could easily and (fairly) cheaply replace what most sites have with a system that can offer what all end users want, which is a search system that reads between the lines.

This is especially true where you own a collection, where the words and meta data used to describe collections tends to be extremely academic. Integrating conversational interfaces with collections could allow us to reveal fascinating stories to a wider range of people. I'm so confident of this in fact that I can guarantee it will feature in this diary series.


Ticket booking systems are, and I'll accept no argument on this, entirely bad. If you tried your hardest to design a system to decrease engagement with a brand you'd fail to do worse than ticketing systems. Conversational interfaces offer a set of potentially very exciting ways to address these problems. Buying a ticket should be like ordering a drink at a well staffed empty bar. Quick, efficient, pleasant, and conversational. And ideally, and this is where chatbots could offer something no standard web experience can: surprising. More on this later.


Actual brand engagement. Forgive me talking about branding for a second but just stick with it. Your brand is not your logo or your font. Your brand is your company's personality, it's what it wears and how it talks and the opinions it has. Your brand is the personification of what your company believes in and does. If you don't nail the tone of voice of your chat bot you will undermine your brand in seconds. As such chatbots will, more than any other single touch point, have to personify your brand perfectly. If you were talking to Patek Philippe and it said "Oi oi matey! Wanna watch?" you'd a: click the little red close icon and b: instantly lose all faith in that platform.

Your tone of voice needs to be set, be right, and be codified into something the machines can replicate. You want to engage with users through chat bots - but you need to do it right or it will backfire.

Design and content.

It's taken us _many_ years to move away from designing pages and get clients to work around building content patterns. Chatbots are going to force us to take this a stage further. In a chat bot interface you don't own the look and feel, people will be engaging with you via WhatsApp, Facebook Messenger, or text messages. As such the overall look and feel is set, and so we must focus on the very small. Aside from the tone of the conversation, see above, we'll have to think very deeply about the content we'll display in the small visual gimmicks we call 'cards'.

These cards can be static content, or they can be micro applications that sit on the thread. Each has one specific job to do, and must be designed with only that in mind. You can think of them as a nano-site sitting within a text interface that exists temporarily and only to achieve one outcome. This is a new way of thinking about digital design, and I can't really write more as I've not yet formed any useful opinions.

In brief chat bots are fairly dumb fake people who you can have a scripted conversation with about whatever their owners want you to talk about. An example would be booking a taxi.

My prototype number 1.

That's a basic and easily scripted conversation. The chatbot has to do one thing, there are only so many options when booking a taxi, and it's fairly easy to keep people on track. When designing a conversation like this things are fairly simple to keep track of, at first.

If you study the section of the conversation shown above you'll see that at first glance it's all simple enough, but it very quickly moves into some really tough challenges to solve. Both in terms of technology and design. The bot needs to be able to respond to a dazzling variety of potential responses to every question, and these responses need to be mapped out in advance of them happening.

While basic AI can convert a human response into something it can understand using natural language processing (NLP) it's not perfect. But this technology offers some very interesting ways to advance the concept of what the web can be. We started with the Cab Bot test above just to get our feet wet and see what was possible to produce in under an hour. In reality it doesn't work, it just looks like it does, but that's half the battle.

In the next post I’ll demonstrate our first actual working bot talking to a large collections database.

Andrew Larking

About the author

Andrew Larking

Creative Director