Why we don’t have a killer chatbot yet

Most of today’s chatbot experiences are terrible. Conversational user interface design (UI) is difficult because it is foundationally different from the web and mobile interfaces. So how can chatbot makers identify patterns of interaction that engage users and build holistic experiences that delight them?

Conversational UI design is difficult for two main reasons: 1) Chat has a “random” non-hierarchical flow, and 2) we don’t have a general purpose artificial intelligence that supports open-ended user input. Let me explain both issues.

The UIs we know are linear, not random

Most mobile and web experiences are linear. That is, you search for an item, add it to your cart, enter payment information, and then check out. Chat or conversational UI is fundamentally different from mobile and web UI. In a conversational UI, the customer’s journey starts with variable information. For example: If I want to buy movie tickets, I can start with “What’s playing at 6:30?” or “I want two tickets to The Interview at 6:30 at the AMC on main and 3rd.”

So, the first challenge for chatbot designers is that the customer’s journey is not known ahead of time. The bot must be able to assist the user and provide the right answers without requiring a linear discussion.

Artificial intelligence is not ready (yet)

The second issue plaguing chatbots is that general purpose AI is still in a nascent phase. Bot makers today rely on simple linear trees (SLTs) that use preprogrammed paths or early AI routines that pattern-match against known entities to discern the intent of the user.

These generally work well when there are a limited set of ways a user can interact with the bot. Yet for most bots, user input can be random. This leads to unexpected input that your bot doesn’t handle. Without better AI, it becomes a game of cat and mouse.

Solution: Iterate, iterate, iterate

So how do bot makers succeed given these limited tools? In an environment where the inputs are not known beforehand, and the best path is not predefined, there has to be a rapid, low-cost iteration path to success. Bot makers need to understand what their users are doing, learn how their bot is responding, and iterate to address issues that are blocking the user journey.

The tools that best support this iterative process are bot native, meaning they understand and translate the complexity of random conversational interfaces into clear metrics rather than dumping them into an endless supply of meaningless dashboards or an endless dialog.

Using these tools will allow marketing teams to identify similar segments of users and connect with them using personalized messages. They allow creative and editorial teams to identify messaging that’s off tone and off brand and to address it. And finally, these tools allow business leaders to understand their efforts in detail without needing a data scientist and an engineering team to “run the numbers.”

This ability to understand conversational UIs and iterate fast will help establish clear, differentiated, bot-native businesses that leverage the power of conversational interface.