How to avoid the chatbot graveyard

Tips for launching your future-proof virtual assistant

Hero graphic

Recently we’ve learned that the average lifespan of a digital assistant is just 1,5 years. This is often due to unmet expectations, negative customer feedback or low usage. We believe we have found the right multi phased approach to help mitigate these risks and deliver future-proof digital assistants.

This approach consists of four key phases; strategy, design, launch and optimization. In this article we will zoom in on the launch of your virtual assistant. We show how to use a staggered deployment with voicebots or chatbots and how to grow users in a controlled way.

Limit scope

Our first tip is to avoid trying to help your customers with all questions they might have, with the first version of your digital assistant.

From our experience with elaborate digital assistants, we learned to launch with a limited scope. This allows testing and refining the chatbot in a controlled environment. This can mean the digital assistant will launch first in a single subdomain of your services. Before extending the digital assistant’s responsibilities, data from users is used to validate the performance and customer experience.

Let’s say you are responsible for the contact center of a tour operator. Instead of launching a digital assistant to answer questions about all travel options you start with inquiries regarding booking a hotel within Europe. You notice that questions regarding cancellations are not fully recognized. When you have improved your NLU, for example by training the AI, and have validated the virtual assistant again, you can start adding other domains to your digital assistant such as booking outside of Europe or booking rental cars.

“Wouldn’t this lead to a lot of disappointed customers asking questions the virtual assistant cannot answer?” I hear you thinking. We haven’t forgotten about them. With a warm handover from the virtual assistant to a human agent or request form outside opening hours we can help these customers without losing satisfaction. Furthermore, we track all questions leading to handover with which we determine how the scope can be extended after validation.

Limit exposure

Another way to create a staggered and controlled start is to expose the virtual assistant to only a part of your customers initially. That way, if (or: when) bugs or issues arise, only a small group of users will be affected. After a bad experience it can be difficult to persuade people to use the chatbot again, leading to a low adoption rate. As you may have heard ‘You never get a second chance to make a first impression’!

To solve this, we start introducing the digital assistant to a small number of customers. For chat this can be done by showing the virtual assistant only on certain pages or by using a specific introduction to send customers to a certain use case. For voice bots we can introduce the bot to a subgroup of your customers by hanging it under a specific route in your IVR. Another way is giving the voice bot a separate phone number, linked to a specific question or target audience. Lastly, customers can be asked randomly if they want to skip the cue and talk to a digital assistant instead.

Testing with a small representative part of your audience gives insights in adoptation of this new way of communication. Furthermore, the focus group can be asked their opinion on the digital assistant. We use this information, along with handled rate and completion rate to validate the digital assistant. Next, we benefit from the data gathered from the focus group to improve the customer experience of the digital assistant. After optimization through processes such as A/B-testing, adding unhandled questions and training the AI, the digital assistant can be introduced to more customers.

Validate assumptions

To check if your digital assistant is ready to be introduced to more customers, it is important to remove uncertainties and validate that the conversational solution meets your business objectives and expectations.

For a digital assistant focused on customer service you can imagine asking the following questions when validating.

We want to know:

 

1)         Can we accurately recognize the question the user might have?

To answer this, you start by looking at how often fallbacks are triggered. If relevant questions are not recognized, the digital assistant must be enhanced and trained before extending their responsibilities. It is also important to check for mismatches, especially when you expect questions that are often similarly worded. Find out if your digital assistant can distinguish between those correctly.

 

2)         Does the customer adopt the digital assistant as a way of communication?

Not all customers are used to talking with a digital assistant. By providing multiple ways to contact your company, you can discover user preferences and see when customers are choosing a virtual assistant for support. A common mistake is removing other channels before knowing in which cases customers want to use a digital assistant.

 

3)         Are customers satisfied with the digital assistant?

Before rolling out the digital assistant for every customer on a specific platform you must check if users are happy with the service level the digital assistant provides. This can be different between platforms. Don’t assume customers who often call will be similarly open to the virtual assistant as customers that often chat.

When all these questions can be answer with a resounding ‘Yes’ your digital assistant is ready to take on more customers and more tasks. If not, find where the assistant needs improvement. Benefit from the user data to discover topics for development and test your new digital assistant again through a staggered start.

Next step

Our strength lies in working together with your team to introduce future-proof digital assistants. We help picking the best strategy for launch that fits the ambition of the company. Together with you we develop the right approach that removes constraints and doubts. Curious how we can help you with launching your digital assistant? We are ready to assist.