Here’s What Most Businesses Get Wrong With Chatbots

AS MACHINE LEARNING systems like Gemini and ChatGPT have gone mainstream, one of the most ubiquitous (and affordable) business benefits has become the ability to offer a chatbot.

But what is a chatbot?

Let’s start by talking about what it’s not. A chatbot is not a problem-solving technology that can replace the intelligence of a trained customer success representative. As anyone who has gotten trapped in an inane chatbot maze can tell you, an out-of-the-virtual-box chatbot is no more helpful than a 5-year-old the day after Halloween.

A chatbot is a widget or app, usually found on a website, that uses Natural Language Processing (NLP) to understand visitor questions. A rule-based chatbot follows pre-programmed rules to provide answers based on keywords or patterns it recognizes in the questions. An AI-powered chatbot uses a machine learning system to learn from data that has been provided to it to offer more dynamic responses. Some chatbots go further, helping visitors book appointments, make payments, or submit forms.

I can’t talk about chatbots without thinking of this Bill Gates quote: “Automation applied to an efficient operation will magnify the efficiency. Automation applied to an inefficient operation will magnify the inefficiency.”

Therein lies the problem with chatbots. Most companies hand the implementation off to an IT or web team without understanding that the technology is the easy part. The hard part is thinking through the questions customers ask and the paths the answers may take depending on subsequent customer input.

What does that mean? I’ll illustrate with a basic chatbot logic path most retailers would use:

Greeting and Initial Question:

Chatbot: “Hi! How can I help you today?”

Customer Input:

Customer: “What’s your return policy?”

Identify Intent:

The chatbot uses NLP to recognize the customer’s need.

Provide General Information:

Chatbot: “Our return policy allows for returns within 30 days of purchase. Items must be in new condition.”

Offer Options for More Specific Information:

Chatbot: “Would you like to know more about:

Returning items by mail?

Returning items in store?

Exceptions?

Customer Input:

Customer: “Returning items by mail.”

Provide Detailed Instructions:

Chatbot: “To return items by mail:

Visit our website and log in.

Go to ‘Order History’ and select the order you wish to return.

Print a prepaid return shipping label.

Pack the item securely.

Attach the return label and drop the package at the post office.”

Check for Further Assistance:

Chatbot: “Is there anything else I can help you with today?”

This logic path must be created for every question a visitor might ask.

This detailed work is essential to a properly functioning chatbot. Do it right, and you can offset customer support costs and improve sales velocity. Do it haphazardly, and you will frustrate customers and damage your brand.

Here’s how to prepare for a chatbot:

• Create a list of every type of customer service inquiry. Ask everyone involved in customer service to note each type of customer service interaction over a period of 6-8 weeks. If you have a fully featured CRM, just pull your customer service data and save a bunch of time.
• Map each customer service interaction to identify every possible customer question and answer, including the conversational branches. What do I mean by a “branch?” If a customer answers “yes” to a question posed by the chatbot, that’s Branch A, and a “no” is Branch B. Any time a customer answer could lead to different sets of responses is a branch.
• Document the questions, options, answers, and instructions for each inquiry type.
• Program the chatbot.
• Test every possible combination of answers to findthe blind alleys and dumb answers.
• Tweak and refine.
• Test again.

Chatbots can be a marvelous business tool. Unfortunately, right now they’re not living up to their potential. But with a bit of planning and discipline, yours could be a boon to your business.