Conversational UX and UI for Spikes Asia Festival 2017, Singapore.
Creative skills - User experience research | User experience design | Information Architecture | Wireframing | Prototyping | Interaction design | UX writing | Product strategy & roadmap | Product design | Product marketing & growth.
Management skills - Product pitch | Product and project management | Client and stakeholder management | Business
At Kontiki Labs, I spearheaded Spikey, the official chatbot on Facebook Messenger for Spikes Asia Festival of Creativity 2017, Singapore. For us, as well as the organisers of the Festival, this was our very first event bot. I bagged this project through several product pitches and new initiative efforts.
Spikes Asia is a collaboration between Ascential Events (Europe) Limited and Haymarket that takes place once every year in Singapore. The Festival and Awards provide the Asia Pacific region’s creative communications industry with a platform to network and exchange ideas, bringing together the finest creative thinkers from around the world.
In this article, I’m going to cover the well thought out approach that simplified the complex program structure, setting the bot’s purpose right on point and made Spikey a success. This will also serve as a great event chatbot use case to follow!
So, here’s how we nailed it 🎯.
STRUCTURING 🎛️
Firstly, what was the need for a chatbot at Spikes? Here’s why —
Know who their audiences really are and what do they engage more with at the event (they usually have multiple components — the talks, awards, workshops, trainings, screening of the work, networking)
Make it easier for the attendees to browse through the exahustive festival itinerary and plan it in advance based on their interest areas. Events like Spikes are usually spread across multiple venues and feature a complex itinerary. This makes it difficult for attendees to plan their personal event schedules and then keep a track of everything that’s going on
Mobile winds up being the primary tool of the digitally savvy attendees. A bot would allow them to get quick replies in a familiar messaging/chat interface
Also, in today’s age, we cannot really expect everyone to download the event app. The push was to move away from app. Another point to note, app hardly gives any deep analytics around the user behavior, etc.
Secondly, what’s the problem statement —
To begin with, we outlined the central problem and challenge areas specific to Spikes after multiple discussions with the Spikes team and also taking into consideration based on their prevoius experiences. This helped us deconstructing the core and work backwards.
This made us go back to white boarding and break down all the parts involved. Doing this exercise helped us cut through the unnecessary detail and get right down to the heart of what content to serve up to our users.
Breaking the complex event structure down. White boarding to the rescue!
USER PERSONA 👩💻👨🎨
On to the next thing — ‘who is our audience?’ In this case it was folks from advertising and marketing industry. Understanding this demographic was one of the most critical steps from a conversation voice and tone perspective. This, in turn, defines the tech liberties that both developers, UX writers and designers can take.
SOUL OF THE BOT — THE CONTENT & CONTEXT 💙
Conversational Interfaces need a specific content strategy along with a special kind of Content Management System with NLP built-in.
We used our customized CMS platform, KONTIKI — that we have built from ground up. Kontiki is an AI driven CMS for voice and chatbots that powers the content using principles of existing robust CMS’s like WordPress and marrying them to conversation design guidelines and NLP techniques.
Kontiki Labs AIaaS Platform
The platform was used for establishing the rock solid foundation of the bot conversation by:
Linking the automatic content feeds from the web
Authoring conversations in a rich interface
Bit sized content structures suitable for bot interface
Providing in-chat support for emojis / images/videos
GIVING SHAPE TO THE BOT — THE FEATURES 🤖
Feature proposition during initial discussions.
Having established the big picture and the content, we came up with a list of features and core elements that would support the objectives we had defined. These were:
User Onboarding — Crucial because it helps set the right expectations, gives context and guides the users
Capturing interests for tailored recommendations — Knowing that our attendees belonged to several different domains within advertising, such as creative, digital, tech, branding, social, and so on, we created domain streams using Kontiki CMS to serve personalized content tied to these domain interests
Program Filters — We gave users the option to filter programs based on days, themes, speakers and tracks of the event
Detailed Schedule — We also kept this other channel open in case users wanted to browse through talks day-wise. (And guess what, the analytics have confirmed that this feature was very heavily requested)
Setting Alerts — Users could set reminder alerts for their favorite talks (5/15/30 minutes in advance) and the bot would notify them accordingly.(Another popular feature as per the FB analytics)
View Shortlisted Entries — During the event each day there used to be category wise shorlisted entries announcements. User could view the same in the web view format
Locating Venues — The event was spread across 3 different venues — main festival, award ceremony and after party. The bot could help with navigating and locating the venue
Hotel and Accomodation — Based on Spikes tie up with some of the local hotels which was further segmented budget range wise, bot could recommend stay options
Free Text Search — Users could also select to directly type out their query
Spikey’s conversation screenshot
BEAUTY THAT BROUGHT THE BOT TO LIFE — NLP 😎
The magic sauce in this recipe was the NLP that allowed Spikey to intelligently understand whatever query users posed to it — whether relating to the event, festival program, speakers, location, food, coffee or things related to on ground support
A question can be asked in a zillions of ways. Instead of training just the keywords we trained the phrases by first categorizing the query buckets such as — on ground support, program related, venue related, etc. — and then by creating separate entities around each of those. It could also detect the typos from users’ end and still come back with relevant response. For instance —
“Where is the festival held”/ “How to reach the festival venue?”/ “Venue?”/ “Directions to the venue” / “Where can I find the direction?” / “Directions?” / “Festival venue” / “Show me festival venue”/…
Knowing that the audience belongs to the advertising industry, we packed some fun elements and made Spikey smarter (of course by training it using NLP techniques) to respond to greetings, its whereabouts/personality and some other questions related to bots in general. We drafted a list of questions based on the entities we mapped out earlier
At the same time, we didn’t want our bot to take expletives or harassment lying down. Hence, training for this aspect of interaction was something we gave a lot of thought to it
We constantly tested the bot day in and out for several days to see how differently users will ask and further train the bot accordingly before it was set to deal with the real world out in the public
The approach of neatly structuring and defining our objectives right at the outset helped us produce great results and validated our efforts towards building a purposeful event chatbot.
We were clear that our chatbot would be more helpful to the users if it could be kept as simple to navigate and browse. And, ‘keeping it simple’ is so much easier said than done!
We are not claiming this, but the numbers certainly are. That brings me to the last part of this article — analytics.
THE MAGICAL NUMBERS — BOT ANALYTICS 📊
Following the launch of our bot, user activity touched 5.9K. Besides, the bot was never turned off at any point by a single user.
Another interesting insight that emerged which indicated a healthy sign of stickiness and user engagement was that —
Bot had sent out as many as 4.2K messages
Bot received 789 user responses
Bot Call To Actions were clicked 907 times
The most popular actions performed by the users were these —
> Accessing festival program
> Setting alerts
> Getting direction to the venue
OUR LEARNINGS 📝
The bad news: Bot discovery is still a missing part of the puzzle and this insight arises from my firsthand experience at the event. Surprisingly, there were many who were not aware of the bot or the process of locating it. Even though the communication regarding bot launch and how to use the bot had been done well and circulated through the right channels multiple times, there remained a number of attendees who didn’t actually latch on it. I personally gave some attendees a demo on how to scan the bot’s M code (Messenger should probably give me some brownie points for educating their users).
The Good news: Those who did discover the bot told us they’d been glued to it and had continued engaging with it all through the event. This was validated by the numbers as well. Once again, this proves that conversational user interfaces are here to stay eradicating the need for accessing multiple and cumbersome interfaces.
There are tons of chatbots out there. But what makes a really good one? The answer is simple — a well-structured chatbot serving a definitive purpose and making the users smarter. That’s really what it all boils down to. Ultimately, the bots depend on us humans to be at their best! :)
The bot is not there in messenger now since the event is over.
However, you can take a look at the full demo on the bot functionalities in the walkthrough video below