At Econsultancy we’ve always been happy about sharing our own data, as we know that our audience loves to learn from case studies.
So as well as building the chatbot, we thought it would be useful to blog about the development process so that you, dear reader, can get an insight into how it all works.
If you’re unsure of what a chatbot is, you can get a decent overview from these other posts:
- What are chatbots and why should marketers care?
- Chatbots: Are they better without the chat?
- What makes a good chatbot UX?
- Five pioneering examples of how brands are using chatbots
And now here’s the story behind our own Econsultancy chatbot.
What’s the point?
On the Econsultancy blog we often question the logic behind using tech for tech’s sake. Ben Davis has a running feud with everyone who works in the internet fridge industry, for example.
So why have we built this chatbot and what do we hope to achieve? Like a lot of other brands, one of the reasons we’ve jumped on the chatbot bandwagon is because of its PR value. The tech is still new and interesting, and it makes a nice change to our usual marketing campaigns.
However, it’s also a good fit with the themes of Supercharged and acts sort of like a proof of concept for the event. We run a number of other conferences during the year and we’re unlikely to create chatbots for any of those.
Finally, it just sounded like it would be a bit of fun and a good learning experience. Our expectations are modest – we know we’re not going to sell loads of tickets through the bot, we just want to test out the technology and create an interesting experience for Supercharged’s audience.
What does it do?
We don’t have the resources in-house to build a chatbot, so instead asked marketing technology agency Byte London to build it for us.
Byte have previously built bots for Adidas and Just Eat among others, so are very much the experts in this field. As such we leaned on them heavily for advice on what our bot should and shouldn’t do.
Having decided that we wanted to launch a bot, our first challenge was to work out what we wanted it to do. This is obviously the wrong way to approach a marketing project, but hey ho.
If our chatbot was to stand a chance of success, it needed to provide some value to the user. As such we decided to make the functionality extremely simple – in its first iteration our bot just offers information about Supercharged and explains a bit about chatbot technology. Additional functionality will be added as the event approaches, but only where it is of genuine use to our audience.
With any new tech there’s a temptation to try and push the envelope a bit, adding quirky features to surprise and delight your users. However, this chatbot has been created specifically for a marketing conference – we have to be realistic about how often people are going to use it and expect them to have limited patience for quirky functionality.
One other thing worth noting is that everyone who messages your company via Facebook will be greeted by your chatbot. That means that if anyone messages Econsultancy to ask a customer service question, they’ll get caught in an automated conversation about Supercharged.
As we only get a handful of queries via Facebook each month (Twitter is our most popular social channel), we were willing to take a risk here. Though a few unsuspecting people might be confused or even annoyed by our chatbot, the benefits of trying out the technology outweigh any potential negatives.
Byte’s strategist Isabel Perry helped lead this project, which began with a meeting where we worked out exactly what we wanted the bot to do and how it should be articulated. At this stage I should point out that our chatbot doesn’t actually use AI. As with most bots, all interactions are guided by a decision tree with pre-programmed responses.
So, having decided what we want our bot to offer (event details, speaker info, links to tickets, info on chatbots), we then had to design a conversation diagram which leads people to that information via a series of multiple choice questions. Think of it like a ‘Choose Your Own Adventure’ book.
We’ve attempted to make it feel as natural as possible, but it obviously means our bot isn’t capable of having a proper conversation. That type of tech is beyond us at the moment. So what do we do when people go off piste and type in free text?
The only solution is to try and predict what people will ask, then script relevant responses. Thankfully Byte have experience of this already so could offer advice on the types of questions users might ask, but there will always be things we hadn’t thought of (particularly when you take into account misspellings and abbreviations). We plan to keep an eye on common questions and update our script accordingly.
This Slideshare shows the full chatbot conversation script, created by Byte using Omnigraffle. You’ll notice that the bot also asks the users some questions. We added these in for two reasons: to make it more of a two-way conversation and to collect some anecdotal data that we can potentially use in blog posts or at the event.
Once the script was finalised, Byte built the bot using its own custom framework, which is based on BotKit.
If you’re tempted to build your own bot, you might be pleased to hear that you don’t necessarily need a developer for it. However, you’ll obviously be making a compromise on quality.
According to Isabel: “There are good tools available like Chatfuel, Converse and PullString that help non-developers. That said, imagine the difference between using Squarespace to make a website and working with a developer.
“It’s fine for chatbots with static content, but you quickly approach the limits of what they allow you to do. There is no backend so you can’t handle data, you can’t scale, you can’t use dynamic content and you’re limited to basic conversation logic. Developer free chatbots are easy to get started but they’re labour intensive to expand and maintain.”
Once the chatbot was complete it had to be submitted to Facebook for sign off. This is a fairly simple process and is really just to prevent businesses from spamming users with push notifications.
After our bot was given the okay we were able to demo it within Facebook using Chatfuel. This allowed us to make edits to a live version before unleashing it on the public.
Building the chatbot is perhaps the easiest part of this whole endeavour (I say that having had nothing to do with the development side other than writing some content). The bigger challenge is getting people to actually speak to it.
Discoverability of Facebook chatbots remains an issue, particularly as chatbots are still relatively new technology so people need to be educated on what they are.
Other than simply relying on people to search your brand name in Messenger, the options available for promoting a chatbot include:
- News Feed Ads that link to Messenger.
- Facebook SDK Plugins that make buttons you can add to your own website / app.
- m.me links that direct users to the bot (e.g. ours is m.me/econsultancy)
- Messenger codes. These are codes automatically generated by Facebook which users can scan using the Messenger app. This is ours:
We’ve just begun plugging the bot via our social channels and plan to also include it in emails and other Supercharged marketing. Our challenge is really two-fold – we need to both raise awareness of our new event, plus this chatbot that sits alongside it.
I’ll update you on the success of our marketing efforts and how our chatbot evolves in future posts.