4 common chatbot implementation mistakes and how to avoid them

Level: Intermediate

Yes, artificial intelligence enhances our customers’ experience by providing immediate responses. In fact, 68% of consumers cite this as their favourite feature, according to Userlike.

“68% of consumers cite this as their favourite feature”.

Userlike

However, a good experience can turn into a nightmare if the chatbot isn’t programmed correctly.

Here are four mistakes we see on some websites and how to fix them.

1. POORLY STRUCTURED FAQS

With a conversational AI chatbot, we’re obliged to create a menu of options (similar to a written IVR menu) for users to select the most relevant question.

But what if there isn’t a suitable option, or worse still, there’s no place to ask an open-ended question? This hampers the customer experience.

The philosophy behind self-service is to automate customer service without sacrificing the quality of the agent-assisted experience.

If we don’t embrace this premise, we’ll create a bot that won’t benefit the user and will result in a flood of phone complaints.

SOLUTION 1: OFFER A VARIETY OF FAQS

Especially at the beginning, when you are unsure of which FAQs to offer, we recommend offering a range of options, from pricing and product information to job applications or demos. Once the bot is up and running, it is important to continually improve these FAQs.

Using the bot’s configuration panel, you can track how many queries each topic receives and replace those that have no queries within a month.

SOLUTION 2: IMPLEMENT A HYBRID CHATBOT

Another approach is to enable a hybrid chatbot that combines a conversational AI tree menu with open-ended, generative AI queries. This reduces user frustration when they can’t ask for what they want.

While generative AI can be creative and susceptible to trolling, it is important to limit it to specific knowledge bases and test it before activating it on our website.

This option is particularly attractive for contact centres, which can use open-ended questions to incorporate the most common ones into FAQs.

2. INABILITY TO SPEAK TO AN AGENT

Customers who contact a chatbot are generally looking for a quick, out-of-hours response. However, some may have more complex queries and prefer not to call.

“The customer is always right” is a mantra when dealing with the public. Even if phone channels are reserved for incidents, if a user contacts the chatbot with a serious problem, it is necessary to allow the chatbot to transfer the request to an agent.

TRAINING THE BOT PROPERLY

We train our generative AI to automatically transfer the chat to an agent if it doesn’t understand the nature of the query, without the customer explicitly requesting it.

Firstly, because with the new Spanish Customer Service Act, we must always offer human support as an alternative to chatbots, or face fines of up to £100,000.

And secondly, it significantly improves consumer readiness: according to Simplr, 80% of consumers are more likely to engage with a chatbot if they know the conversation will be transferred to an agent.

“80% of consumers are more likely to engage with a chatbot if they know the conversation will be transferred to an agent”.

Simplr

3. NOT ASKING FOR FEEDBACK

If we’ve recently activated the chatbot on our website, we’re missing a significant opportunity to measure customer satisfaction by not requesting feedback after each interaction.

The AI can analyse and identify user sentiment during conversations, and send us a rating for each one. However, it’s also beneficial to contrast the bot’s automated assessment with direct user feedback.

Consumers like to be heard, so while we can measure their sentiments independently, it is always valuable to involve them in the bot’s improvement process.

4. 4. ARTIFICIAL INTELLIGENCE HALLUCINATIONS

AI hallucinations occur when the bot confidently gives a wrong answer because it thinks it is the right answer.The purpose of AI in customer service is to free agents from routine queries, not to give us headaches.

To avoid this, we recommend that knowledge bases are well structured. This means avoiding mixing concepts in the same article.

For example, if a document explains how to activate a service, don’t include the steps to deactivate it.

READY TO IMPLEMENT A CHATBOT FOR YOUR BUSINESS?