Chatbot for Slush: 68 percent automation on day one
GetJenny is setting up Slush’s official chatbot. The first day after going live was promising, as based on early data the chatbot was able to handle 68% of incoming messages on its own.
Slush began with a few hundred attendees back in 2008, but for its 2017 Helsinki issue expects more than 17,000 people. Such a major event isn’t easy to run, especially not on the support side, which is where automation comes in.
In 2016, Slush received roughly a thousand requests over two days through live chat. The most commonly asked questions included “Where can I get my badge” and “How do I get to the venue” — simple enough, but once they occur in the hundreds, a team is quickly pushed to their limits.
Automate recurring questions, optimize, and keep staffing efforts low
These requests are generally similar, which means that there’s the option to handle them automatically, namely with a chatbot, a program that can have a conversation with anyone requesting help. Either the program handles all of the conversation itself, or it refers the person it is talking to to a human service agent. The latter is done if the question is recognized as too complex for the bot to handle.
GetJenny’s approach makes it possible to integrate its conversational engine, called Jenny, with a variety of different systems. In the case of Slush, this is Ninchat, a nimble Finnish chat platform.
Once set up, Jenny is then taught a large number of so-called states, particular situations the chatbot acts on depending on the input it receives, for example providing information on how to get from Helsinki airport to the venue, or retrieving up-to-date agenda information from Slush’s database through its API.
Teaching Jenny: 68 percent automation on day one
The starting point for teaching the chatbot was an analysis of the previous year’s live chat data. Through added testing and training the chatbot with hundreds of volunteers, the resulting learning curve will help the guys at GetJenny perfect their approach for this specific use of the bot, and make it possible to create a response that is both immediately useful and close enough to human interaction.
On the first day of operation, the Slush bot achieved 68 percent automation, which means that the bot was able to independently respond to more than two thirds of all the requests for information it received in its testing phase.
The chatbot won’t depend on opening hours or hired staff to man workstations. It can handle a multitude of last year’s conversations, while leaving only the most complicated matters up to agents to settle. At night, while the support agents are enjoying their time off, the chatbot keeps on answering questions and logging any question it has not been able to answer, making it a breeze to teach it about more topics the next day.
Ready for action and 17,500 attendees
Setting up the technical base for Jenny’s Slush incarnation took the team some two days, which included integrating the solution to Ninchat’s live chat platform. Training the conversational engine and feeding it with all the states and queues it will be expected to act on naturally also requires a few days, but is a straightforward process than can easily be adapted to a variety of requirements.
The team at GetJenny expect their chatbot for Slush, also called Jenny, to do well. A day before Slush kicks off, they’re busy making sure it has a broad enough answer base. Once that is given, Jenny is ready to face 17,500 attendees, 2,300 start-ups, over a thousand investors, and scores of journalists!
You can read more about the partners from Slush blog post
Originally published at Get Jenny.