Sentry Scouts: Bots — A Recap
Sentry Scouts Meetups are organized by software developers for software developers, about the future of building and using software. Sign up for the next Sentry Scouts Meetup or enjoy a video of a past Meetup.
Or do both. Or do neither. Or simply become paralyzed by choice and take a nap.
Let’s take a moment to remember the holiday season (hello, 2019) and reflect on interactions with customer support. Maybe, like me, you had to contact a specific store’s customer support every day for a week. Fun times. Most people dread these conversations — listening to hold muzak, explaining the same thing to multiple people, wondering if the issue will actually be solved anytime this century.
Bots, the focus of June’s Sentry Scouts Meetup, alleviate the pain of customer support interactions and expedite the process by answering (certain) questions quickly, passing us to a real live person when needed, and lightening the mood with a joke or two.
In case you haven’t heard, Sentry Scouts are an opportunity for like-minded friends and professionals to swap stories around a faux campfire (a fauxre, if you will). Don’t worry — we also provide plenty of camp-related non-pizza snacks and drinks to go around. We gathered a fine group of
bots completely verifiable humans:
Bots simplify processes, both internally and for end-users. Potential use cases for bots are seemingly endless, and as the technology improves, so do those use cases.
Many of our panelists emphasize the benefit of using bots for customer service. Bots can provide quick answers about shipping, promotions, and even help a customer place orders. Allowing customers to have their FAQs answered by bots cuts time to issue resolution while also keeping agents available for problems that need additional attention.
Sarah Barnekow, Partner Engineer at Slack, explains that “you can use bots for anything you experience in day-to-day life, including HR and project management to restaurant and travel reservations,” and our panelists certainly demonstrate that variety. Ernesto Soltero, a Software Engineer at ETRADE, is replacing virtual assistants on stock application pages and experimenting with stock trading directly through chatbots. Bex Warner, Developer Advocate Intern at GitHub, works on bots that help developers by “running checks on commits in a PR and give line-by-line output.”
With bot usefulness established, the next question turns to personality: should a bot have one? Or should a bot be your stereotypical emotionless robot? Some of our panelists, like Kevin Poorman (Sr. Developer Evangelist at Salesforce), enjoy bots with more personality. Poorman tries to add as much character as possible in the bots that he builds. Dennis Yang, Co-Founder & CPO at Dashbot, explains that personas allow companies to communicate and reinforce their brand.
Including personality begins with an understanding of how people talk. Poorman recognizes that it’s “harder to have [personality] that will go over well with everyone.” Lauren Kunze, CEO at Pandorabots, Inc., advocates for using a cross-functional team during bot development, including allowing legal to approve everything the bot can say. Warner recommends looking at the problem you are trying to solve first and allowing personality to follow naturally.
Natural language understanding isn’t a solved problem, or we would be talking to software, and it would talk back.
Even if developers have a full grasp of how their bot should interact with customers, training bots is a challenge. Yang encourages the review of as many conversations and questions received by bots as possible. This data analysis allows bot developers to figure out if they’re handling those conversations and questions the right way.
In other words, developers can use natural language processing (NLP) to understand what people are intending to say, retraining bots to recognize and respond appropriately. Kunze warns that “natural language understanding isn’t a solved problem, or we would be talking to software, and it would talk back.” While deep learning and neural networks have made tremendous breakthroughs in AI, the technology isn’t quite good enough to completely rely on NLP platforms.
Bots that follow specific patterns tend to success really well.
Barnekow suggests that sometimes a more structured way to get data, like rules, makes more sense. Warner agrees that there are perks to allowing customers to click a button over giving open-ended responses. “Bots that follow specific patterns tend to success really well.”
However, as Poorman explains that there’s no reason why developers can’t combine NLP and rules. Kunze agrees, advocating to use all tools available to developers. While there’s no silver bullet solution currently, Pandorabots has seen good results with a hybrid of machine learning and rules.
Thank you to everyone who joined us for Sentry Scouts: Bots. We hope to see you next time. In the meantime, you’ll want to check out our fantastic blog, educational (yet fun) tutorials, and other helpful resources.
See you there! Or here, on the internet!