Building bots is easy. Keeping users around is not.
Since Facebook Messenger bots launched at F8 in April 2016, over 30,000 bots have been created. Major brands like ESPN, Star Wars, and Yelp have released apps on Apple’s iMessage. Microsoft, Google, and Amazon have all heavily invested in platforms for bots and conversational and voice experiences. Pandorabots, one of the oldest chatbot platforms, boasts over 285,000 bots made by over 225,000 developers. As a bot maker, you’ve got competition. Lots of it. And it’s only getting worse. Dozens of startups have popped up that make it super simple to build bots. Some, like API.ai, enable you to build once and then deploy to Facebook, Slack, Telegram, Kik, and Amazon Echo.
Other bot-making tools let you create specialized bots. Olabot specializes in making bots to boost your personal brand. ManyChat helps businesses make content bots that distribute their thought leadership. Dennis Yang, cofounder of popular bot analytics platform Dashbot, has a unique perspective on what makes chatbots succeed or fail. Since launching four months ago, Dashbot has over 530 bots that have collectively pushed over 40 million messages. Some of the top-performing bots on Facebook are from their customers.
Successful bot developers employ a number of tactics to grow and keep users, but Yang and his team observed that one single tactic outperformed all the rest. Here it is:
You have to ask for feedback.
Simple and obvious, right? Yet very few chatbots employ useful feedback loops.
“When you finish a task, that is a wonderful time to ask for feedback,” says Yang. Even if a chatbot is completely utilitarian and only uses card options with users, asking an open-ended question requesting feedback after a key completed action enables developers to get valuable, qualitative insights.
Here’s an example. GameMonk, a popular Slack bot with thousands of users, gives hardworking professionals a fun break from the workday. Teams can play creative games such as guessing relevant hashtags for a GIF meme or naming as many items as possible in a given category.
After implementing feedback collection, the GameMonk team learned two key facts. First, users consistently complained of feeling rushed, so the team increased the time allotted for each question. Second, international users pointed out that the bot was English-centric. Because the company is based in Silicon Valley, the team didn’t realize how popular Slack is abroad.
When GameMonk incorporated feedback loops, the bot saw a 24 percent increase in session length and a positive 23 percent boost to sentiment.
Feedback is even more useful when you add questionnaires to segment users. Ask Haley is a Facebook Messenger bot used by parents to find activities and classes for their kids. The team implemented feedback collection using the net promoter score method.
If a user answers “Very Likely” or “Likely,” Ask Haley automatically recommends that they share the bot with their friends. This leads to a higher share rate and a boost in growth and engagement for the company. However, if a user isn’t happy with the product, Ask Haley solicits open-ended feedback to learn more.
When the bot registers a low score, a human admin from the Ask Haley team takes over from the bot and manually engages with the user. This ensures that users have an empathetic listener who adapts to their issues.
While everyone loves to get positive NPS ratings, the negative ones teach you much more. In the beginning, Ask Haley profiled users only by asking for their location and the age of their kids. Based on open-ended feedback, the team implemented a new feature that bucketed users based on their activity preferences and parenting style — for example, Crunchy Moms, Career Parents, or Vegan Families. When Ask Haley allowed parents to see what similar parents had chosen, NPS scores improved by 10 percent, and users were much more satisfied with the recommendations.
Asking for feedback won’t always improve your bot, but not implementing intelligent feedback loops means your team will miss critical information that could be used to design new features, fix frustrating bugs, and drive user engagement.