*or website, or livechat, or voice search, or god forbid – an actual conversation with a human being?

In which this naïve writer wades into… well, not uncharted water, but mildly choppy, chatbot-filled water with nothing but his own limited experience, barely concealed prejudices and a pair of inflatable armbands that will make typing difficult.

I’ve been using Siri for the last few years with the same limited success as most people – mostly frustrating, but have found that if I ask it to “open the pod-bay doors,” it answers with a withering reply that the question deserves.

However, the world of personal voice-powered assistants suddenly opened up to me in a wonderful way last week, when I experienced Alexa (Amazon’s own voice service, which looks like a particularly fancy pedal-bin that can tell you the temperature) for the first time ever.

It was while staying in an AirBnB, heated with a Nest and lit via Hue (excuse me, I’ve finally out-millennialed myself) that I saw the potential for having my own omnipresent robot butler. One that could tell me the nearest takeaway restaurant, or read me the headlines, or play me a television programme as long as it’s available as part of an Amazon Prime subscription… OH THE POTENTIAL IS SO RICH I CAN PRACTICALLY TASTE IT.

“Please don’t put litter in me, it’s very demeaning.”

Having said that, the reality is that I couldn’t figure out how to use Alexa in conjunction with the Nest, so my family froze for the duration of our stay. Turns out neither Nest nor Alexa can tell you where the boiler is or that the pilot-light had gone out. But as I said, the potential is rich and tasty.

Unfortunately, I can’t afford or justify the need for a £150 glorified speaker, but I admire its convenience and I was suitably inspired to take my next step into the world of loquacious helper robots and began playing around with the far more affordable and portable joys of chatbots.

Yes chatbots, the interface-reliant, more accessible and slightly less ‘Skynetty’ cousin of the voice-based personal assistant.

What’s a chatbot?

A chatbot is a rule-based computer program that mimics human conversation with the user. In some cases these conversations help the user complete a simple task (ordering a pizza, booking a table in a pizza restaurant, other non-pizza related things) and in other cases they try to mimic a full-on conversation just for the sake of chatting.

The rise of chatbots

The year 2016 was described in many ways, largely as the year most ‘handy’ with the grim reaper’s scythe and as the ‘no really THIS year is DEFINITELY the year of mobile’, but also as the year of the chatbot.

This excitement stemmed from many events, including…

  • Towards the end of 2015, the ‘big 4’ messenger apps overtook the ‘big 4’ social networks in terms of monthly active users.
  • Facebook Messenger grew from 200 million active monthly users in 2014 to 1 billion in July 2016 – although that’s largely down to Facebook moving the function to a separate app that users had to download to continue using.
  • Gartner wildly predicted that by 2020, the average person will have more conversations with bots than with their spouse. (I’ve already copyrighted ‘spousebot’ FYI.)
  • And most importantly of all, Facebook opened up its Messenger app in April 2016 to all developers so that they could build their very own chatbots that work within the service.

Examples of chatbots 

Of the popular big brands using Facebook Messenger, you have the Pizza Express chatbot, which will let you reserve a table or contact customer services. The Guardian chatbot, that delivers a personalised selection of headlines at your chosen time of day. Or there’s Skyscanner, which lets you research airfare prices direct from Messenger.

Then of course there are a variety of apps outside of the realms of existing messenger apps. One of the most regularly cited examples of a ‘good chatbot’ is Poncho. He’s an anorak-wearing, weather-obsessed cat who delivers personalised weather information along with a side of flip-humour and loosely-related gifs. He’s, you know, fun.

There’s also Luka, an app that wants to position itself as the one-stop shop for chatting to all your friends and bots on a single platform (although most of its press has been around its co-founder resurrecting his deceased best friend in chatbot form).

And then there’s the Bieber Bot, which allows you to have conversations like this with a robot version of Justin Bieber…

Well it would, if I could get past the loading stage…


So, much like any tech trend, you can see that chatbots run the gamut from useful, to interesting, to stupid.

And much like any tech trend, I got very excited about chatbots for precisely two days.

I played with some chatbots, and oh boy did I have some limited fun

I realise that I have technically interacted with a chatbot before. In fact, every damn time I’ve signed into a new team on Slack. The conversation now inevitably results in this…

But this is a very basic (and modest) example of a chatbot, unskilled at coming back with a suitably witty reply. Let’s try Poncho for size…

He’s a helpful little critter, and alerts are really easy to set up and change. Poncho will also tailor reports based on your allergies or bad-hair-day awareness. Sure he doesn’t understand colloquial variations on ‘thank you’ but that’s no big deal.

He also has a nice way of deflecting ’emotionally charged’ conversations.

Alright, Poncho you can stay.

I also set up the Guardian chatbot to deliver headline articles direct to my messenger app at 7am. This lasted a couple of days before I got bored and realised that I could just go to my Guardian app for a fuller experience in my own time.

Which in turn made me realise that I could also just go to the native weather app on my iPhone and with a single tap see immediately what the weather is like in my location now and for the next 5 days.

That’s when I silenced the chatter, as there’s nothing I’m being provided with here that I can’t get in my existing apps, apart from a ton of cluttered notifications in Facebook Messenger.

I then turned to some of the more ‘off the beaten path’ examples. I’ve heard good things about Luka – especially in terms of being more about ‘shooting the shit’.

This lead to an exchange with a Prince bot, that felt more icky and confusing than entertaining or useful.

I guess the spelling is fairly accurate, but missed a chance to shoehorn in a ‘I Would Die 4 U’ reference.

Luka’s weather app also doesn’t seem to like the way I structured my question or the exasperated tone of my voice.

You can understand why people either lose their patience or just get bored after a few minutes.

So what’s the point of chatbots?

Well to paraphrase Ted Livingston, co-founder of messenger app Kik, which launched its own ‘bot shop’ in 2016 – it’s a question of consolidation. Natively carrying out tasks via your most frequently used messenger app is potentially better and more convenient than scurrying around multiple apps on your phone.

To that I would counter: is using more than one app on your phone really that much of a hardship, especially for apps that work far better than they do in the messenger app?

In case you forgot what apps on an iPhone look like. It’s probably been a minute or two since you glanced at it.

Livingston also mentions that there’s less friction in using a chatbot for the first time. “There’s no new app to download, no new account to create, and no new interface to learn.”

Now that’s more interesting. I don’t have the Skyscanner app, and actually the chatbot experience was so promising that I would probably not bother using it via any other platform. However I do still need to visit a third-party airfare site to pay for my ticket, as Skyscanner just aggregates prices.

Chatbots are also terrible for form-filling. In a recent talk at The UX Conference, UX designer Bride Trozelli revealed that the average form can take seven taps on the standard mobile experience, as opposed to 110 taps via chatbot.

The fall of chatbots

Yes it’s that soon. Much like how proper news journalists already have the queen’s obituary written in advance, marketing writers like to pre-write their ‘DEATH OF X TREND’ piece as soon as there’s a whiff of popularity. However there is a feeling this year that perhaps chatbots aren’t quite as exciting as we first believed.

Facebook is rethinking its chatbot strategy, and is now encouraging developers to ease up on the more ‘chatty’ qualities of their chatbots, with a new format that strips it down to interactive menus and clickable buttons.

“You may have been tempted to use the conversational flow to present your bot’s main features and flows… Consider stripping such exchanges down and cutting to the chase by putting the most important features in your menu.”

Read as: people mainly just use chatbots to tell Pizza Express to fuck off, so let’s try and fix that.

Ted Livingston has also reigned in his enthusiasm for the format. In a blog post from August 2016, Livingston realises that “so far, there has been no killer bot. This is not yet the world that the early hype promised.”

Notice the use of ‘yet’, which suggests it’s only a matter of time, but as Livingston goes on to say, “I believe we’ll look back on the early emphasis on ‘conversational commerce’ as a mistake.”

Perhaps the problem is one of expectation.

I’ve been careful to avoid using the term AI so far, as I (and other more level-headed) experts don’t believe that many of these bots are truly an example of artificial intelligence. Instead they’re guided by decision trees, which map out a finite number of possible responses to a finite number of queries.

Thanks Wikipedia!

These can certainly be incredibly complex, but essentially the user is playing a digital version of a Choose Your Own Adventure novel, not something that’s learning and adapting based on the behaviour of you and other users.

We’ve been promised AI-powered chatbots, but instead we’re met with a weather app that doesn’t understand natural language phrasing, or a Prince avatar that mercifully feels far from the real thing.

Some, dare I say, best practices

I’m always wary of promoting ‘best practices,’ as one industry/company/person’s best practice doesn’t necessarily map to another’s. However, Bride Trozelli did make some common sense recommendations if you want to develop your own chatbot:

  • Always introduce yourself – be friendly, personable, and polite
  • Immediately follow with a CTA – don’t let your users second-guess your potential, and ultimately leave disappointed
  • Don’t pretend to be human – the uncanny valley is an unnerving place
  • Give something before asking for something back – if you’re asking for a location, tell the user how helpful that will be for them
  • Use structured input where you can – URL cards, buttons instead of keyboard input, and geolocation
  • Interactions should be short and precise – users should understand the value of the chatbot within 10 seconds
  • Have an exit strategy – if it all goes wrong, make sure there’s a contactable human there to help

My opinionated conclusion

Aside from the limited structure and mis-sold potential, I think the major stumbling block is the reliance on the keyboard and for humanity in general, the reliance on the screen.

Apps allow for taps and swipes to get you where you want to go. Chatbots require you to type on a fiddly keyboard and wrestle with the vagaries of autocorrect. Sure, Alexa can mishear you – but the error rate in speech recognition is now at 5.9%, which is about equal to professional transcriptionists.

Perhaps voice-based personal assistants have spoilt us. Technology is trying to move us away from screens – for the sake of our own safety and sanity – with haptics (touchy stuff), gestures and voice becoming the next frontiers of user interface. Chatbots feel more like a step backwards into text-based RPGs than a leap forwards into a more personalised, butler-robot assisted future. Now Alexa, where’s my dirty martini!?