Early Learnings from the Bot-verse

If you have just 60 seconds….

Facebook announced a bot store for Facebook Messenger at its developer conference this week. Bots, over time, are set to replace apps. As we’ve seen already through our own experiments for Sony Pictures, the potential for brands here is massive. However, a superior user experience is key to truly capitalize. This requires 1) building a conversational user experience from the ground up; 2) employing the smarts, through machine learning, to enable a fruitful conversation; 3) having a long-term view where your brand bot becomes your ultimate brand ambassador, building one to one customer relationships, at scale, over time. 

Facebook announced a Bot Store for Messenger, at F8, its annual developer conference in San Francisco this week. This means that now anyone can have a go at creating a bot on the platform, with the potential to reach the 900M people who are using Messenger today. You can now explore and start a chat with the official F8 launch bots here – quite aptly, via the dedicated F8 bot!

bot1-400

The potential here is gigantic. If Facebook gets this right, the Bot Store will replace the App Store; developers and brands will build bots instead of apps; brands will finally become more easily (and socially) discoverable on mobile; consumers’ ability to engage with brands will be more seamless, and they will no longer need to commit to a lengthy and data-sapping app downloads. Overall, bots promise to provide a better user experience, with the potential to convert to massive untapped consumer engagement and revenue opportunities.

For the past six months, we at UM have been quietly testing messaging bots through an exclusive partnership with the startup, msg.ai

This partnership resulted in the first ever AI bot in Messenger, to support the launch of Sony’s Goosebumps Movie, back in October. The results from this pilot were staggeringly promising. Slappy, the main character in the movie, which we chose as the voice of our bot, had conversations with consumers which lasted an average of 10 minutes each – and in some cases, surpassed the length of the movie itself. A third of those consumers returned to have a further conversation. Slappy not just entertained, but sold movie tickets and directed consumers to merchandise, growing interest in the franchise. You can check out his highlights reel, here.

This sort of result, while evidently attainable, is by no means guaranteed. The F8 launch bots are a case in point. They currently perform rudimentarily, and the experience today is not yet better than an app or website. True, we are only at the beginning of an exploration into a whole new user interface, and these bots will get better as they are optimized and get to know you better – but there are some clear lessons off the bat.

While there is still much to learn all round, our experience of the F8 launch bots has helped reinforce some core principles in our approach for Goosebumps Movie pilot.

  1. The bot experience must be natively conversational. It must be designed that way – by human specialists – from the ground up.
  2. The bot needs to be smart for the experience to be good and enduring.
  3. Bots are best as a long-term play where the relationship builds and matures.

1. Be natively conversational from the ground up

Not many of the F8 launch bots are designed conversationally. The CNN bot (which you can find via the CNN Facebook page or the F8 bot) provides a news digest with options for dig deeper. But it is fairly wooden, with each shipment of news introduced with the same statement: “there is a lot happening in the world. Here are your top stories.”

bot2-400

The news, already developed, has been repurposed into digestible chunks. It is a decent mobile experience, and personalized to an extent, but it is not truly conversational. Same with the Spring bot, which seemed to replicate the user flow of the e-commerce site – again, not native to chat, and not providing a better user experience than the app format. You can sense that she is starting to get what I mean (second screenshot), but she is not set up to deliver the information in the format I expect it in, through a conversation.

bot3-400

bot4-400

We have seen conversational news, and shopping, done well via the Quartz app, and PS Dept, respectively. In both cases, the experience is designed that way from the ground up, involving expert human editors and stylists. Both also have a human, first person conversational voice. In the Goosebumps case, the voice was Slappy. We took advantage of his large personality, and gave him lots of props (campaign assets), so we could engage and entertain at length.

bot5-400

bot6-400

2. Be smart  

The F8 launch bots frequently failed to deliver the answer I was looking for. Poncho, the “weathercat,” was unprepared to answer my question about the weather in New York. This is presumably because it did not recognize the term “NYC.” Given that this is all I wanted to know from my conversation with Poncho, this made for a rather poor user experience.

bot7-400

This reinforces the need for machine learning. For Goosebumps, we mapped out the range of possible questions and answers manually, as a cross-functional team. We then applied machine learning to automatically generate multiple iterations for each question. This meant that Slappy would have known that NYC meant New York, in the same way that he understood colloquial terms, such as “love it… sure…. Will do.”

bot8-400

3. In it for the long haul

Conversations work as a highly effective medium, because they contain implicit understanding of context and tone. In a real relationship, this understanding only gets better. The same goes for bots – as long as they are built with machine learning. We are finding that this a requirement for bots, and what sets bots 2.0 apart from the chat bots of old.

You can see the difference in the exchange below. Slappy’s friend decides to re-engage him after a few weeks of silence. Because of their historical relationship, Slappy was able to come back with a very human sounding response. He disarms his friend with a reminder that their relationship went way back!

bot9-400

When done right, bots can be our best brand ambassadors and customer support agents, which people actually want to interact with. When the initial pilot ended, we had people saying how much they missed talking to Slappy.

Taking the long-term view, bots have the ability to single-handedly drive customer relationship management. A bot can manage thousands of individual customer relationships at the same time, and through machine learning, nurture and convert those relationships over time.

Bots represent a massive opportunity for brands. However, as the inevitable bot-rush happens, brands will still be competing to be heard and seen, just on a different playing field. Brands that do it best, and learn this new platform early, will win.

Leave a Reply

Your email address will not be published. Required fields are marked *