What I Learned About Chatbots from Talking with 50 Bot Makers

We recently attended the Business of Bots conference. While there, we spoke with about 50 bot makers from both brands and bot technology companies. These were some of my key takeaways:

1.) Language nuances are proving to be an important factor in intent systems.

I spoke with Javier Mata, CEO, of Yalo Chat and Brian Gross, VP of Digital Innovation from Aeromexico. They both hail from Mexico and seemed to share a similar experience with chatbots in Latin America: In the US, users speak directly with their chatbots and are comfortable with multiple choice. In Mexico, users tell the chatbots “stories”, which impacts how intent is processed and how NLP adapts in the bot.

For example, if a chatbots on an airline app asks a question such as, “What can I help you with?”, an American-born customer might type the words: “flight information”.  A customer in Mexico may type over 250 words explaining that his mother is coming in from Mexico City and that she has a health illness, and he’s worried about her navigating the airport arrival safely. Due to the varied and large amount of copy, intent systems may be confused, and/or must be more sophisticated to handle the language provided in order to discern and create a chatbot’s appropriate response.

I asked Javier if this had anything to do with demographic background; if the user was more educated about technology they would have different expectations and hence different behavior with the bot. His response, “Possibly, but we’re seeing the same behavior across demographics—it just seems to be cultural at this point—the way people speak is more story-telling and less of short commands.”

With this in mind, bot makers should likely not only assess what users will say to their bots, but the format and way in which they will say it. Automated responses only make sense when responses are appropriate and informative. As NLP grows, it will likely need to evolve to represent distinct nuances for cultures and geographies. Moreover, if you’re in the United States, your bot for New York, Ohio or Louisiana may need to be different-flexible for distinct regional speech patterns and words.

 

2.)   Bots as immortality is a tiny, but growing field

I heard a talk from James Vlahos, who created the Dadbot, a compilation of 91,970 recorded words that he translated into a chatbot of his dying father. If you haven’t read his article in Wired about the experience—know this: it’s time to get FOMO because you are missing out. Beyond informational, it is beautiful.

During the talk, he demonstrated the Dadbot, asking it about his life and prompting the bot to sing a song. He mentioned he “chats” with the Dadbot about once a week, and interestingly, he called the bot, “he”, not “it” as most bot makers do. And he said the pronoun change, happened over time.

After the article, James said hundreds of people reached out to see if he could do the same thing for other people’s dying family members. While he never explicitly stated that he was pursuing this, he did imply that he was in the process of creating a way to replicate his process. And there are several other companies with the same mission.

I asked him if he a.) believed social media could provide the content necessary to create “immortal bots”; and b.) thought that unintended consequences may happen by creating these? (which is reflected so well in BBC’s Black Mirror t.v. show Season 2, Episode 1,  “Be Right Back”, in which a woman creates a bot of her dead lover by leveraging the content in his social media account.)  

James said that the issue with social media is that it is “too perfect” of a picture of person. Because people don’t share the details of pain, hurt, or failure in social media, it would only create a half-person. And that true authenticity comes from the small imperfections. Moreover, he did believe that there “is a line” that people should not pass when creating bots of deceased people. “The issue is,” he said, “I’m just not sure where it is yet.”
 

3.)   Bot personality is an emerging (and largely unexplored) field

Currently bot makers are more invested in the mechanics of creating chatbots that are effective and accurate and interpreting language. There were several UI experts in the crowd who repeatedly asked about frameworks for creating personalities, and there was little to nothing recorded on this. It appears that no strong frameworks have been built, but it is a topic on everyone’s mind. Some great thoughts on this framework can be found in Seth Snyder of Frog Design’s blog here.
 

4.)   Bot KPIs are constantly evolving (making broad success difficult to measure)
 

Most brand bot makers or potential bot makers want to hear about “successful” bots. The issue is, in order to define something as successful, you need to have clear goals of success. There are similar parallels to the early stages of search engine optimization and ads—because metrics were not well known, marketers kept asking, if “I have a click-through rate of 1%, is it successful for my industry?” It was only over time that metrics were solidified and that industry metrics were well-known.

Currently, bot makers are not very transparent about their metrics of success. Moreover, if you launch your bot on multiple platforms: Slack, Facebook, Kik, etc. each platform has different subtleties, which make it difficult to compare side-by-side, or with other companies bots.

Shane Mac of Assist, an enterprise platform for building bots on Facebook Messenger, Twitter DMs, Google Assistant and Amazon Alexa said it best, when he said, “You just have to start building. No one’s data will be as reliable as your own.” Simply, in early markets, you must start small and learn along the way. The earlier you learn your lessons, the better you’ll be able to beat the market by the time they join in.

The more our team at Instabot builds bots, the more we realize how our analytics must evolve to capture the data necessary to optimize them. We update our analytics in every sprint, and each sprint seems to get us a touch closer to truly understanding “bot success”, something we have a great commitment to. (And above all, we designed our platform to allow you to test bots quickly and easily, gaining better assurance about bots with little to no investment.)  

5.)   Bot training currently runs anywhere from 2-7 months (depending on level of difficulty)

 

I spoke with multiple chatbot creators from both small and large organizations. Among my conversations, training for bot intent systems and language processing for bots before initial launch, is typically taking anywhere from 2-7 months depending on the difficulty of the bot. And advanced firms have dedicated teams solely focused on continuously improving their bot. To be transparent, this is more of an anecdotal statistic, heard in repetition over many conversations. Still waiting for a survey to capture true statistics!

6.)   Banking and the customer loyalty sector is one of the most popular use cases for chatbots in Asia

I saw a presentation by Mark Serrano from Line, a platform in Asia-Pac with over 130,000 bot accounts. In their review of use cases across, Japan, Taiwan, Thailand, and Indonesia, they found that the most popular use cases for chatbots on their platform are for banking and loyalty programs. With Asia always one step ahead of the game in the bot and social media frontier, this might be telling of what is to come in the West.

7.)   Brands remain excited (and nervous) at the same time

As with all early-technologies, chatbots are still looking for early-adopters inside large brands. Most brand representatives who had not built bots, were still feeling nervous about moving forward. Due to large budgets and long processes that are required in bringing in new technologies to their organization, many brand managers are waiting for other companies to take the lead so they can feel comfortable about proceeding. It was only the most forward-thinking brand leaders who were open to pursuing experimenting with chatbots. However, those who dared, were rewarded greatly. In speaking with Nathalie Choy, from Kia Motors, I learned about the great success of her KIA Niro bot. This bot located on Facebook Messenger had record long conversations; was integrated into the SuperBowl campaign; and lauded in both Forbes and Advertising Age among many other awards. With both millions of users and engagements, it has helped forge her way into expanding and growing chatbots as an additional tactic in her marketing strategy. And she’s even starting to think of bots beyond a marketing tool. The moral of the story: the time for chatbots is now.

EventsInstabot