How data-driven chatbots can ensure success

Data from, for example, online conversations and knowledge bases is basically up for grabs, but often remains unused when developing a chatbot. This is a shame, because data can really help your organisation get the best out of your chatbot. I spoke with Software Engineer Gerben van der Huizen and Data Analyst Koen Hallmann about the added value of data for chatbots. In this article, I’ll explain how important data is for the development of successful chatbots. Because it’s precisely the combination of insights and technology that will ensure the best results. And you have to admit: you’d like to have immediate success with your chatbot, right?

“Not because it’s necessary, but because it’s possible” is the starting point for many organisations when going into chatbot technology. After all, it is always interesting to keep up with the times, to see what technology can do and where opportunities lie. Yet it would help in general if we were to think more about the goal than the means.

For example, do you use chatbots for service? Then make sure you have a clear goal in mind. Do you want to offer faster service and better response times? Do you want to extend your accessibility, even outside the opening hours? Do you want to create efficiency for service employees by having chatbots prepare the work? The solution lies in data. Your webcare (online customer support) department holds hundreds, or perhaps even thousands, of conversations with customers every day. Use that data to predict whether a certain task would be cost-efficient to have performed by a chatbot. Use this to estimate whether the chatbot would be able to ‘pay for itself’ and avoid any additional costs for the development of a bot that may ultimately not even help with achieving your business objectives.

Software Engineer Gerben van der Huizen and Data Analyst Koen Hallmann develop data analyses to make a prediction about the performance of a chatbot. A Chatbot Feasibility Report provides the necessary insights when considering developing a chatbot for your own organisation. Data Analyst Koen explains:

“By doing an analysis, you can choose a chatbot based on hard figures, instead of relying on intuition or a fascination for the technology. By looking at a sample of webcare conversations from a real, linguistic point of view, you can also see, in advance, where a chatbot may not work properly. That saves you a lot of trouble with your customers later on. And this is something that is often overlooked.”

The underlying idea is more pragmatic, the task of linguistics that deals with the articulation and interpretation of the language.

“You have to figure out what kind of ‘language action’ is involved. Does a customer have a question, but the conversation somehow turns into a complaint? Does a chatbot then have added value? Is the answer that a chatbot gives the customer a complete answer to their question? Once you’ve gone into all this, in depth, you then come up with a numerical overview. And with that complete picture you can gain insight into whether a chatbot is for you, and what a chatbot can mean for you”, says Koen.

By analysing data and making a calculation based on it, you’re able to calculate the expected results of a chatbot for a specific task. A visual report can therefore help when trying to explain to management why a chatbot would be suitable. So, don’t just jump into implementing chatbots, but, rather, look first at where the real challenges lie. Software Engineer Gerben explains:

“Make a realistic estimate of the expected outcome beforehand before you commit to anything. But, remember, data is also of added value after the launch of the chatbot. An analysis shows exactly how a chatbot is performing and whether it pays to expand it further. The effect is made visible by this analysis and, with this, management can be included more readily in the importance, and development, of chatbots.”

An organisation that has made good use of the opportunities a feasibility report can offer is Univé. Mathijs Jilderda, Customer Experience Advisor, explains:

“The report gives us a nice and clear numerical overview and explains in a very concise way what the benefits of a chatbot are for us. Thanks to the analysis we also gained insight into the number of messages that a chatbot can take care of and which questions are suitable for the chatbot.”

When you work with data as a predictor of the success of your chatbot, it is important to guarantee the quality. Don’t immediately assume that a chatbot can be used just anywhere, because technology is not always a solution for each and every task. If analysis shows that a chatbot is not suitable for a certain task, then don’t continue with it. Showing sympathy and empathy is still really only something for a human agent. Therefore, let complaints or any other negative experiences be answered by them, instead of chatbots.

“Bots cannot do anything with empathy. A bot can help you very quickly, for example, but if you’re looking for reassurance, you’re not going to get it from a bot. A real person is an important backup that the bot can always fall back on”, says Gerben.

• Develop your chatbot based on data-driven research and make sure your chatbot is problem-oriented.
• Make a problem or opportunity demonstrable in advance by insights from data instead of your own intuition.
• Stay realistic: does a chatbot most likely not deliver what you want? Then stop early or send the process to another, better application of technology.
• Always keep in mind that a chatbot must contribute to the customer experience. Avoid irritation and make sure that a chatbot connects seamlessly with your customer, but also that your chatbot does not get in the way of the customer and pass the baton on to the employee if that is good for the conversation.