Bots are all the rage. But the first generation of bots isn’t measuring up to expectations.
People expect bots to have the intellect to do more than humans. This expectation is far from reality. Facebook bots have been panned as more pain than they’re worth. Gizmodo compared the user experience to “trying to talk politics with a toddler.”
Chatbots, personal bots, branded bots, and workbots respond well to carefully crafted commands, but they do not understand context.
The challenge for developers is to create bots that understand nuances and take actions based on them.
Users want a bot like C-3PO who understands etiquette, context, and intent. The first step is to program a bot to understand what’s important. This will vary based on historic and real-time context, and what’s important to a company, a person, or a job. The bot’s ability to take an action often depends on apps, people, policies, or workflows.
These capabilities are challenging to create. In order to break the cycle of ineffective and broken bots, we studied the way more than 1,500 businesses were using our Workbot for Slack, and also participated in a hackathon during Knowledge16 (video) with help from ServiceNow, LinkedIn and Advantage Integrated Solutions, to showcase the possibilities.
Here’s what we learned.
Use natural language, like humans
As Tom Tunguz wrote, when communication doesn’t feel natural, it becomes extra work — so conversations with bots must feel casual and human.
The bot therefore must listen and add insight to the conversation without being prompted. This goes beyond natural language processing — it requires understanding, interpreting, and following up on commands with natural language generation.
When a bot is instructed to “Add a new lead in Marketo,” the bot should understand that leads in Marketo typically flow into Salesforce, and anticipate that need by replying, “I added the new lead in Marketo, and noticed that this lead doesn’t exist in Salesforce. Do you want me to create it?”
Bots become indispensable by adding value beyond what they were asked to do.
Adapt based on a person or job function
With every interaction, bots should learn and adapt to users’ needs and styles, and act accordingly. Some users perform the same tasks each morning, such as asking for a summary of support tickets or for a list of that day’s meetings or contracts ready for renewal. Each morning, the bot should deliver that information without being asked.
However, if a user is asking for information they don’t normally access, then the bot should recognize that and act accordingly. Detecting patterns helps the bot to anticipate needs and offer solutions just like a human would.
Adapt based on the situation
Humans adapt by reading social and emotional cues. They understand what is acceptable in a team setting versus a private setting. Bots need the same understanding. But too often, bots produce a firehose of notifications in a chat channel, overwhelming users.
Instead of creating noise, bots should generate a private message for something urgent that requires action. With sensitive information like revenue reports, bots should not post openly, even if a person has requested that information via a public channel.
Know everything
Specialized bots can sometimes offer an advantage by working extremely well with one app or task. However, C-3PO is invaluable because he has context about everything and is fluent in 7 million forms of communication.
Similarly, a bot with context across all apps in use provides more relevant information and context. Customer information is often spread across all systems. A specialized bot cannot provide a holistic view of the customer.
To avoid the bot “graveyard,” developers must make bots more like humans or C-3PO, with the ability to gracefully communicate, understand context, and deliver relevant and personalized information back to the user.
[Source:-VB]
