Everyone who has ever dealt with chatbots knows that they are not fail-proof. A hybrid model works the best, where chatbots handle basic questions and escalate more severe issues to human agents. Ideally, in this era of self-serve customer service, you would want a lesser number of customer requests to pass on to humans.

In fact, in some industries (especially B2C), we have seen the chatbot to human handoff percentages as low as 10%. Industries with a high volume of customers and low complexity of products generally have good chatbot efficiency. And a subdued chatbot to human handoff percentage is a clear indication of increases chatbot efficiency.

Decreasing the bot to human handoff percentage will help support agents get free from repeating queries coming in. This helps to reduce the first response time, resolution time and speed up the support responses. Which, in turn, helps in improving customer experience.

Chatbot to human handoff is an easy metric to calculate. It is:

(Number of conversations transferred to humans) / (the number of total conversations)

In a customer support automation tool such as Kommunicate, you can see the chatbot to human handoff rate in the analytics section.

Now that we know what the chatbot to human handoff rate is how it is calculated, let’s jump into the discussion on how to reduce it. The following instructions will help you in lowering chatbot to human handoff percentage.

Reducing chatbot to human handoff percentages

Here, I am taking an example of Kommunicate chatbots. However, you can reasonably replicate the process and ideas with a sound reporting system and analytics with any chatbot platform. So let’s jump right into it.

Visit the analytics section on the Kommunicate dashboard and navigate to the Bot section. Under the bot section, you can see Bot Messages. Here you will see the end user’s messages and the type of intent triggered for that.

Now, click on the Categories drop-down and select Fallback, as shown in the following screenshot.

Once you select the Fallback category, you can check the conversations where the chatbot has failed to answer and handed off the conversation to human agents.

Example: 1

I have picked the conversation 55473580 from the above-listed conversations. You can click on the IDs listed in the Conversations # section.

The chatbot cannot answer and hands off the conversation to humans in the above conversation because you did not add the Web installation word and related phrases to the chatbot responses.

Now, we can go to the chatbot building platform (Kompose/Dialogflow/Amazon Lex/IBM Watson) of this chatbot and add this in the training phrase into the chatbot so the chatbot can answer if the same kind of queries comes in again.

You can also try to predict related phrases that match the training phrase Web Installation and train the chatbot accordingly.

  • Web integration
  • Web install
  • Website installation

Here is how you can add it to Kommunicate’s chatbot builder, Kompose:

Example 2

Let’s pick up one more conversation from the listed conversations:

In this example, the chatbot is unable to answer the user message “Hello, do you support other languages?” because the user’s question is not trained, so the chatbot is unable to answer correctly and transferred to a human agent.

You can now add the intent/training phrases for this particular user input and train the chatbot, as we did earlier.

Wrapping Up

This way, you can check for all the conversations where the bot failed to answer and train the chatbot accordingly. This will improve your chatbot’s efficiency significantly and drop the chatbot to human handoff rates. We have seen companies improve their chatbot efficiency by 40-50% by using this simple method.

Write A Comment