6 Tips For Realising True Value From Your Chatbot

How can you continually improve and scale your chatbot to serve more customers, staff, and partners better, with a broadening domain of knowledge? Here are six tips for improving your chatbot.

The business case for chatbots is logical: Provide 24/7 on-demand information and support, plus a consistent customer experience, at a low cost. But how can you continually improve and scale to serve more customers, staff, and partners better and with a broadening domain of knowledge?

The answer lies in how you go about building your natural-language processing (NLP) data model and, moreover, understanding the impact of making future changes to the training data or adding new intents.

With that in mind, here are my top six tips for building a better bot.

1. Give Your Chatbot A Personality, Not Just A Name

Think about your brand and the audience you intend to serve. Tone of voice is important to gain an instant connection and empathy with the people you are hoping to engage. Take into consideration the culture of the audience. We see significant differences between how people interact with chatbots in different geographies.

2. Don’t Start Off Too Small

Many professed chatbot experts will advise you to start off small. Would you release your human agents to the general public with minimal training and subject matter knowledge? Of course you wouldn’t. Be ambitious. Look at the domain and subject area you want to deliver, and plan to build a model that initially supports 50 intents.

3. Ensure Iterative Testing Before Going Live

Remember you are dealing with human beings who can be incredibly unpredictable, so you need to try and cover a lot of bases. Create a testing pool of people from your business or organisation, making sure you have a diverse mix of roles, age, gender, and cultures. You can also include friends, family, and even your best and most loyal customers. Start with approximately 10 people, gather their feedback, and analyse their questions and the chatbot’s responses. You can then retrain and adjust before releasing it to the next wave of testers, maybe 20 this time. Repeat step one and push out to around 50 people. Repeat again, and if you are confident, release your chatbot to the world.

4. Test And Retest

Look for the emerging chatbot testing solutions that are coming to market and ensure they are optimised to support your NLP provider. The ability to identify regression, confusion, and ambiguity across intents means your chatbot performs optimally and you have the confidence to scale your model with a high degree of accuracy without disrupting continuity or quality of service. In addition, the black box is changing continuously, which can also impact performance, so having this visibility is crucial.

5. Continuous Analysis And Retraining Is Essential

Conversational data model analysts and administrators are some of the new roles being created with the advancement of AI in the enterprise and customer experience ecosystem. It is important that you monitor and assess daily the conversations people are having with your chatbot. As with tip No. 4, you’ll likely need the support of a chatbot testing solution or corpus management system. You want to ensure that you are capturing new lines of questions and conversations that can add value to your chatbot’s performance. The ability to react to sentiment or an event quickly means your chatbot becomes the information touch point of choice. Look to retrain at least once a week in the first months after going live.

6. Consider Chit-Chat And Be Able To Handle Certain Questions

People are generally social and like to have a chat every now and then. With this in mind, think about some of the random questions people might ask your chatbot, such as, “What’s the weather today?” or “How old are you?” or “Can you tell me a joke?” or “Can you cheer me up?” These are all genuine questions our conversational platforms have been asked. By training your chatbot to respond with an appropriate answer, the conversation is more likely to continue, and the human is also more likely to return. Again, this is where your conversational data model analyst comes in. By monitoring the conversations, more insight can be gained and more training can be done to improve the customer’s experience.