10 Lessons for Effective Bots in 2017

Chatbots, or conversational interfaces powered by artificial intelligence, were thought to be the next great platform in 2016. Facebook introduced the bots for Messenger Platform at their annual developers conference, F8. The intention was that bots would enable new ways for brands to connect directly with consumers, welcoming a whole new paradigm for communication [eyeroll].

It’s time to check in with the hype vs. reality of chatbots.

The reality is that conversational interfaces, as sold by the most ardent evangelists, lack the level of sophistication that users expect. But that doesn’t mean that it’s time to hang up chatbots and move on to The next big thing—the real opportunity isn’t artificial intelligence powered human-computer conversations, it’s instant interactions available where your customers are.

Some Lessons for Effective Bots in 2017

We’ve had the opportunity to build a few chatbots over the last year and have collected some lessons that can hopefully help others build successful bots of their own.

1. Natural Language Processing isn’t there… yet

NLP can be really effective if the scope of the interaction is narrowly defined, but we’re too far from general AI for most chatbot interfaces to reflect a real conversation.

2. Offer prompts

Prompts can help direct users towards more successful interactions. Facebook offers several helpful interface templates that provide a visual cue for what’s possible, which can provide more direction than expecting users to memorize commands as though they’re on Windows DOS. Nobody wants to return to the command line interface.

3. Infuse some personality

Just because you won’t likely fool people into thinking they’re chatting with a real person doesn’t mean the experience should lack personality. There is room for surprise and delight in these early experiments. Self-deprecating humor and a well placed emoji can go a long way towards creating a light hearted, casual experience.

4. Don’t dismiss web views and extensions

Facebook offers web views and extensions, in addition to the interface templates referenced above. Web views allow you to build more custom experiences by integrating additional functionality directly into the messenger environment. That being said…

5. Focus on utility

This might sound obvious, but you have to say no to deliver the best products and experiences. Overly ambitious ideas lead to a lack of focus and set the wrong expectations. What’s more, your vision for replicating your brand voice in a chatbot will simply fall flat due to the current limitations of the technology. Focusing the experience around utility maximizes the value of the bot by creating quick interactions in context while minimizing the chances for disappointment.

6. Consider interactions over conversations

One of the unique opportunities for chatbots is the ability to provide an instant interaction in the context where users spend their time. Consumers spend 85% of their time within five apps on average, with messaging apps increasingly dominating that time. Instead of fighting to be yet-another-app on a user’s home screen, users could be interacting with your brand in between making plans for dinner.

7. Failure is an option

Expect that your bot will misunderstand an input. Always have a response that allows the bot to fail gracefully and get the user back on track. This could be as simple as a prompt that offers some tips for re-engaging in conversation, to offering a customer service number to call for more complicated requests. Never lead a user to a dead end in the conversation.

8. Don’t forget about discovery

Unless you already have a high user engagement with your existing Facebook presence, you will likely need to provide some form of discovery for your bot. Thankfully, Facebook offers a variety of ways for users to discover your bot, both on and off platform.

9. Learn as you go

Don’t let the current limitations of AI stop you from building your first bot. The beauty of machine learning, the underlying technology powering bots, is that it gets better over time. Part of the process of building a bot requires training the learning engine to detect user intent. Sometimes the robot understands what you type and other times you have to guide it towards a more complete understanding. We do this process before launching the bot so that it can at least perform the most basic tasks, but there’s no reason you can’t continue teaching the bot new tricks over time.

10. Start building!

It’s never been easier to build an application for a new platform. There are a ton of tools out there that can help you get started, even if you don’t have a technical background. The sooner you get started building your bot, the sooner you’ll have learnings under your (and your bot’s) belt to have a sophisticated and useful tool in the market for your customers.