The Right Way, and the Wrong Way, to Hand Bot Conversations to Humans

Let’s face it, everyone agrees that bots are not as smart as humans and there are certain conversations that need to be escalated to human agents. However, not everyone agrees on what’s the best way to handle this and the “conventional wisdom” often favors immediate human handoff using live chat. Not so fast, read on to see the major pitfalls of this approach.

Problem #1: If a prospect is given the option to talk to a human when using a bot, the entire conversation is doomed for failure. Why? Because instead of trying to solve the reason a prospect might want to talk to a human, the bot developer elected to take the easy route and provide an instant human-handoff option. Also, bots can solve a lot of issues if given the chance and people can navigate good bots with relative ease. But if a giant detour sign is thrown in the middle of the conversation called “talk to a human,” many consumers will elect to take that route vs. seeing the bot through.

Problem #2: Bots are fast.  Really fast.  The average response time for a bot is less than 2 seconds.  So when a consumer using a bot clicks on “talk to a human,” they go from a 2-second response time to:

  • Waiting for the bot to find a human to connect with (anywhere from 60 seconds to as long as 5+ minutes).
  • Waiting for the human to review the conversation up to that point (if they even do that).
  • An average 1 minute 16-second interaction time.  Good chat agents respond to an inquiry in less than 1 minute.  But those are rare and most chat agents are managing multiple conversations at once.  In today’s world of immediate response, going from a 2-second bot response time to a 76-second response time feels like an eternity.

Problem #3: Unless you’re using a call-center, there’s likely no human available to immediately handoff to! Leasing agents aren’t sitting around waiting for a consumer to chat with them which is why chat doesn’t work well in the Multifamily Industry (more on that in another article).  Using a call center is extremely expensive and we all know call centers aren’t nearly as informed as onsite agents. And even if you do have an expensive call center available to back up the bot, refer to the aforementioned problems #1 and #2.

 CSAT (Customer Satisfaction) Scores!

But don’t just take my word for this, let’s look at the data.  When a consumer initiates a chat conversation with another human, the CSAT score starts at around 65%.  If the conversation is successful, it ends at around 65%.  When a consumer initiates a bot conversation the CSAT score starts at around 60%, and if it’s successful can end as high as 95%*.  But, if a conversation starts with a bot then gets “hot-transferred” to a human, the CSAT score usually ends up below 30% which makes sense given the problems listed above.

The Solution…

Find a good bot that doesn’t fail and get confused. Those bots typically use Guided Conversation Methodology vs. Natural Language Processing. If the bot can’t answer a question, create an asynchronous expectation with the consumer.  In other words the bot says, “I’ll forward your question to our leasing agents and they’ll get right back to you. How would you like to be contacted?  Phone/email/text?”  In this instance, the consumer is in control of the conversation, determines how they’d like to be contacted and isn’t expecting to be “connected” right then and there.  They realize their conversation and question will be forwarded and someone will likely get back to them with an answer in the next 20-30 minutes and in their manner of choosing.

*LivePerson, 2017

Leave a reply