7 Keys for Optimizing a CX Chatbot

As an Automated Customer Experience (ACX) Consultant here at Ada, I work with clients to launch their chatbots, gather feedback, and optimize their automation strategies as their business’ needs change. I’ve seen first-hand what great ACX looks like, and what it takes to provide consistent, top-of-the-line customer service that your clients will love.

The first secret I’ll share is that optimization is a continuous process. A bot builder’s job is never done - that’s one of my favourite things to share with my clients. 

Your bot should grow with your business. So how do you keep your bot in top shape and continue providing quality customer support as your business evolves?

There are so many possible changes that can be made to bots! It can be difficult to know where to start and what to optimize. Luckily, I have the seven key considerations to help get you started.

Understand your customer

When it comes to optimizing your chatbot, the first step is creating a clear and detailed picture of your customer.

This picture will inform the decisions and changes you make to your chatbot. These changes could include where your chatbot is living on your site, or the kinds of questions it’ll be able to answer, and can be done successfully as long as the decisions are made with your customer’s best interests at heart. If these choices are made without your customer in mind, you run the risk of making changes that do not improve your customer’s experience.

Create a chatbot persona 

Once you know your customer, it’s time to get to know your chatbot!

Creating a persona for your chatbot can guide you in two key ways. 

First, a persona will help you write consistent dialogue for your bot. Use this persona to shape how your bot talks - does it use slang? emojis? wit? -  and position your bot’s voice relative to your support agents and brand messaging. 

Second, creating a bot persona can foster trust between your customers and your bot. 

Have you ever spoken to someone who gave inconsistent answers or seemed to change their tone throughout the conversation? It probably felt hard to trust that person. It’s the same with a bot: if a customer feels like the bot’s personality is inconsistent, they may be less inclined to trust what it is saying.

Creating a bot persona doesn’t need to take lots of time or effort. Here’s a quick cheat sheet; start by jotting down a few key adjectives and phrases that will embody the bot’s voice. Then, give the bot a name, and maybe a picture to help your customers remember it. That’s it - your bot persona is ready to go!

Handle your bot’s shortcomings with grace

Bots are bound to make mistakes.

While Ada’s machine learning is ahead of the curve, it’s impossible for any chatbot software to answer every possible question that’s thrown at it.

When your bot does make a mistake, be sure to handle it with grace by acknowledging the bot’s shortcomings, and quickly getting the conversation back on track.

Being upfront about a bot’s limitations can help de-escalate frustrated customers. Use the bot’s mistake as an opportunity to educate the customer on how to correctly engage with the bot: something as simple as “I respond best to short questions” can be enough to get the conversation back on track.

Be sure to reset a customer’s expectations of the bot, too. In your bot’s response, remind your customers of what the bot can do, or provide the customer with a list of questions that people often ask the bot. This will lead to a more successful second engagement.

Use your customer’s feedback to improve your chatbot 

A bot is never truly ‘finished’. One of the most important parts of optimizing a bot is improving its AI so that it can answer more and more questions. 

The most efficient way to optimize your bot’s AI and build its knowledge is to take a look at your customers’ conversations.

Look for gaps in your chatbot’s training by analyzing any conversation where your bot didn’t understand a question. If your bot is equipped to answer that question, then add it to your bot’s training! 

Keep a running list of questions that the bot isn’t trained to answer. If you keep this list for a few weeks, you’ll notice trends in the types of content your bot is missing. 

Build fresh content to address these trends, and use your running list of questions to train this new content! Make sure to track when these changes are made, and keep an eye on the key metrics, like reduction of misunderstood questions coming in related to your new content.

Remember: it’s impossible for your bot to be able to answer everything. But it’s crucial to train your bot to answer the large number of questions it can answer!

Start from the source

There’s no better way to improve your chatbot than by starting from the source: looking at your bot’s conversations! Not only is this a great way to find gaps in your AI training, but it also gives you a holistic picture of how your customers are engaging with your bot. 

Performing spot checks on your bot will help identify areas of improvement. Take a look at how your customers access the bot, if the content makes sense, and if it flows well. Doing these spot checks may even give you ideas for new features to add to your bot!

For most clients, I recommend doing this once a week. As you flip through conversations, be sure to set a clear goal. Are you looking for feedback on a particular answer, or are you doing a general scan for trends across conversations? Digging in to your bot can be a great way to stay engaged with and find inspiration for your ACX strategy.

Keep the conversation going

Once a customer receives an answer to their question, they may not know what they should ask next. Every chatbot is different, and customers can have trouble identifying what your particular chatbot is capable of.  

Keep the conversation going by designing decision trees for your bot. Organize content in a logical way so that chatters have the opportunity to learn more and more beyond the single question they initially had for the bot. 

For example, in a telecom company’s bot, a customer may have a question about their phone plan. Once they’ve got all the information they need, keep the conversation going by offering up other topics about ongoing deals, like how they can add to their data plan.

Personalize the chatbot to fit your customer

It feels great when someone you don’t know very well remembers your name. Give your customers that feeling with your chatbot! 

Personalizing your bot’s content with something as simple as including your customer’s name will keep them coming back anytime they need customer support.

However, this strategy can - and should - go beyond a simple name recall. 

Let’s go back to the previous example of a telecom bot answering a question about phone plans. Imagine if the bot had the ability to give exact information about a customer’s phone plan. What if, based on the individual customer’s usage, the bot could ask the customer to choose between data up-sell opportunities right there in the conversation? This is all possible due to chatbot personalization! This is just one way that personalization can create an incredible customer experience while directly contributing to your organization’s bottom line.

The best way to identify opportunities for personalization is to look for existing content that seems like a blanket statement, even though it is actually unique for each customer. An example would be “You can find your bill due date on the billing page”. By using your chatbot to to give the customer their actual bill due date, the bot can make the customer feel personally addressed and keep the entire customer service experience in one place!

To learn more about optimizing your customer support strategy and increase your CSAT, book a demo with one of our experts and get the conversation started!

Kayla Thomson
Kayla Thomson

Kayla Thomson has been working at Ada since our series A, working on our ACX team, and currently our product team as an Associate Product Manager. She's passionate about solving problems, helping others succeed, and being an active listener on Zoom calls.

More info about Kayla Thomson: LinkedIn

Share This Post