Part 2 — Can a Chatbot Prove How Scared You Are?

A case study in utilizing chatbots to augment physical promotions and movie premieres

In Part 1 of this case study, I laid out the scenario that led us to building a chatbot for the promotion of the Warner Bros. Pictures feature film Lights Out. In this Part 2 of the case study, I will bring you through the process we used to successfully complete the project and describe some of the challenges along the way.

Phase 1 — Ideation

We first had to determine what kind of chatbot we were building. It wasn’t strictly a content-only bot…it wasn’t really a conversational bot either…and it wasn’t primarily a choose-your-own adventure bot. It really required elements of all three of the general types of chatbots out there. We needed to be able to deliver movie-related content to the user. We had to be able to answer general questions posed by users and help them navigate to exactly what they were trying to find. And we also had to create an adventure for both the onboarding process and content discovery available outside of the Scare Booth experience.

Basically, this meant the chatbot became a conversational content adventure experience. Having built a number of chatbots for Facebook Messenger in the past, the QwipIt team already had a solid understanding of how to go about designing the conversation. Our team knew how to leverage third-party natural language processing services, such as Wit.ai, to make a smarter conversation that is intuitive and pleasant for users. We also had previously built content chatbots, so we knew we could blend all of the conversational elements to create a hybrid conversation that accomplished the objectives described above.

Then there was the whole hardware element. The Grandesign team had constructed large booths for users to be scared in through a number of visual effects. They had also installed cameras at various angles to capture the best fear faces during the 20 to 30 second experience. The chatbot’s job was to help the Brand Ambassadors (booth workers) in identifying users (via Messenger), prepped for the booth experience, and into the booth. Then the chatbot had to be able to receive the video from Amazon servers after it had been processed and edited, and deliver the video back to the user while his or her heart was still pounding on the other side of the booth.

This set the stage for the QwipIt team. We needed to build a chatbot that accounted for real world interactions between users and Brand Ambassadors, users and the booth, and users and their videos. We also had to make sure we provided depth of content discovery outside of the booth experience so that users could get an enticing glimpse of the upcoming Lights Out movie. Eventually we decided upon distinguishing two main conversation trees: the Booth Flow and the Content Flow.

Phase 2 — Design

We have found that the best way to go about seeing the forest for the trees is to start building and testing right away. Spending an exorbitant amount of time trying to figure out where all the dead ends and pitfalls will be in a chatbot conversation ends up being a practice in futility because you can never foresee it all without testing and trying it out.

Given that approach, we began by creating wireframes of the potential responses on a platform called Moqups. We used Moqups because we could easily create conversation trees and get visually close to the UI of Facebook Messenger’s chat window and message design. Our initial flow looked something like this:

Part of the initial response mapping on Moqups

We then began laying in the responses to our proprietary backend software so that we could test them in Facebook Messenger. Because Facebook is constantly iterating on Messenger, we continuously test our messages to make sure they are rendering correctly and that we are taking advantage of the most recent feature updates. In fact, Facebook pushed a fairly major feature release in the midst of us building this chatbot. In particular, it released “Quick-Reply” buttons and a persistent menu at the bottom of the chat that proved to be extremely useful.

During this process, we had numerous calls across the teams at Grandesign and Warner Bros. Pictures to make sure that we were all on the same page. In particular, we worked with Warner Bros. Pictures to ensure that all copy and imagery used in the chatbot conversation followed branding guidelines and satisfied promotional intent. We also worked closely with Grandesign to create a smooth user experience from end-to-end. We paid close attention to the steps a user would take to go through a Scare Booth and how a user might talk to the chatbot if they weren’t at a Scare Booth.

As we continued to build the responses, we took time to consider the cadence of conversations and made hypotheses about how users would navigate the conversation. There were two basic scenarios to consider. One, the user would be introduced to the bot by a Brand Ambassador when they came upon a Scare Booth at one of the theaters. Or two, they would find the bot naturally when interacting with the official Lights Out Facebook page and would message the page/chatbot directly.

The Lights Out Facebook Page — The “Message” button allows users to talk to the chatbot

Because the chatbot served two very different purposes (taking a user through a booth or allowing for content discovery), we had to be conscious of setting expectations at the top of the conversation so as not to cause intense confusion for the user who had no idea what a Scare Booth was. The chatbot had to be really good at explaining itself, and we needed to give users a few different ways to figure out how to navigate the conversation.

Beyond relaying clear directions in the first few messages sent to the user, we accomplished clarity through a number of different tactics including utilizing buttons and setting keywords. This is where the choose-your-own-adventure aspect came into the conversation. By clearly exposing buttons in responses to click for advancement, we made it easy for users to work down through the conversation, while avoiding dead ends at the bottom of conversation branches.

We also built a menu of keywords that could be accessed at any time. We trained our chatbot to recognize these keywords through an artificial intelligence platform for natural language processing called Wit.ai. Wit is a Facebook owned software service and is free to use. Although it is not the most robust deep learning platform, it served our purpose to recognize pre-determined keywords and a number of common phrases. We used the information we received back from Wit to deliver intelligent responses in the event that a user typed in a random question or one of the keywords.

Keyword Menu and Button Navigation

Lastly, we needed to design a way for users to view and share out their Scare Booth videos while tracking social media metrics, as well as traffic through the booths and chatbot experience. Because Facebook has not yet implemented analytics for Messenger chatbot activity specifically, we had to determine a way to gather additional information outside of the limited data we were able to glean from the Messenger conversations. We accomplished this by designing two different landing pages outside of Facebook Messenger. The first was a video page that allowed users to view and share their own Scare Booth video to Facebook, Twitter, or Instagram. The second page was a Scare Booth information page that showed users where the booths would be and made it easy for users to share out the locations and information to their friends.

Phase 3 — Implementation & Launch

There were a few steps that had to be taken in order for the chatbot go live without a hitch. We first stood up the two different landing pages mentioned above, and implemented Google Analytics and our own custom analytics on each so we could gain visibility to the users going through the experience. The video page (shown in the image below) also allowed for easy social sharing of the videos and greatly improved the number of, and visibility to, impressions created by the Scare Booth campaign. We know this due to a comparison to the other method of video delivery; that being via text message which was able to prove zero social shares or impressions created.

The video page created for each unique user and video pair

We also needed to test the video delivery process. Anytime there is a hardware component involved (the booth), there is the possibility of failures and quirks regardless of how well the hardware is built. We ran through the process a number of times and were able to start the video process via the chatbot, receive the video back from the Amazon servers administered by Grandesign, and deliver the video back to the appropriate user whom had just gone through the booth. That was definitely the tricky part.

Lastly, gaining administrative access to the Lights Out Facebook page was necessary in order to grab the correct authorization information and take over Messenger conversations. After we had access, we deployed the chatbot and prepared for users to go through the Scare Booths in LA and NYC.

Phase 4 — Reporting & Analytics

The Scare Booth videos turned out awesome. And the movie did too, receiving critical acclaim and a Rotten Tomatoes stamp of approval. Below is a sample of a social post that the chatbot was able to create for users after their Scare Booth experience.

The best part about this project was the incredible information and insight we gained on behalf of Grandesign and Warner Bros. Pictures. Because of the design of the chatbot conversation and custom analytics we were able to construct for messaging and activity information, we were able to actually prove the effectiveness of the physical promotion.

Over the course of the Scare Booth experience running from July 8th to July 25th, the chatbot exchanged over 17,000 messages with fans of the movie and people who had discovered the Scare Booths in LA or NYC. As a bit of a surprise to us, well over half (59%) of the users chatting with the bot found the conversation organically, meaning they did not go through the booth but rather found the chatbot through the Lights Out Facebook page. This gave us a multitude of data on how the users interacted freely with the chatbot, what people wanted to know about the movie, how people felt about the movie, and what people were sharing about the movie. Basically, the chatbot proved that people were scared…really scared.

The landing pages for both the videos and Scare Booth information also provided insights into sharing behavior and demographics of people interacting with Lights Out. As far as we know, this was the first time a chatbot had been used to not only augment a physical promotion of a movie, but also provide attribution information about the effect the promotion had.

Conclusion

Needless to say we considered the Lights Out chatbot a huge success. We managed to bridge the gap between digital chatbots and physical experiences while providing insights about one-on-one conversations with a brand.

The project did not come without its hiccups, as is the case with all projects that include numerous teams, custom designed software, and large-scale hardware builds. But overall, we had created a really cool user experience and were able to deliver on all of the objectives for Grandesign and Warner Bros. Pictures.

This type of project will no doubt be utilized in the future for additional physical promotions as it is cracking the code of attribution for marketing agencies and film studios alike. We are excited about the prospects of continuing to advance what we can do with a chatbot for brands of all shapes and sizes.

Follow us on Twitter 
Like us on Facebook 
Subscribe to our Subreddit